--- Title: Winston AI vs. the Winston AI Developer API Url: https://help.gowinston.ai/basic-information/winston-ai-vs-the-winston-ai-developer-api --- _Winston AI actually lives in two places._ [app.gowinston.ai](http://app.gowinston.ai) is the everyday web app. You log in, paste or upload your text or images, and Winston AI tells you how likely they were generated by AI. It's point-and-click, with reports you can download and share. Developers, on the other hand, work at [dev.gowinston.ai](http://dev.gowinston.ai). That site lets you create API keys so your own software—an LMS, a CMS, or any custom script—can call the very same detection engine automatically. The API uses credits instead of per- scan payments and shows your usage and rate limits in a dashboard. Accounts are separate on purpose: your web-app username and password won't open the developer portal, and an API key won't sign you in to the app. If you need to copy-paste text and get a quick answer, stick with [app.gowinston.ai](http://app.gowinston.ai). If you need code that checks dozens—or thousands—of documents for you, create an API account on [dev.gowinston.ai](http://dev.gowinston.ai). Full technical details (endpoints, request samples, error codes) live in our public docs at [docs.gowinston.ai](http://docs.gowinston.ai). Still not sure which site you need? Email us at [support@gowinston.ai](mailto:support@gowinston.ai) and we'll point you in the right direction. --- Title: Do my unused credits carry over? Url: https://help.gowinston.ai/account-management/do-my-unused-credits-carry-over --- Word scan credits are refreshed at the beginning of each billing cycle. This means that any unused credits from the previous month do not carry over to the next month. Each of our plans includes a fixed number of word scan credits that are replenished monthly. Once your new billing cycle begins, your account will be credited with the full allotment for that period, and any remaining credits from the prior period will be reset. This system ensures consistency and fairness for all users, and helps us maintain optimal performance and service levels. If you find that your usage requirements vary significantly month to month, or if you anticipate higher usage, we recommend considering a higher-tier plan. --- Title: Does Winston AI have any certifications for data handling? Url: https://help.gowinston.ai/privacy-matters/does-winston-ai-have-any-certifications-for-data-handling --- Winston AI takes security and compliance seriously. We are proud to hold a SOC 2 Type 2 certification, which ensures that our systems and controls meet the highest standards for data security, availability, processing integrity, confidentiality, and privacy. In addition, we are CASA Type 2 (Cloud Application Security Assessment) certified, confirming that our cloud-based infrastructure adheres to rigorous security best practices to protect sensitive data and maintain trust with our users. These certifications demonstrate our commitment to safeguarding your information and ensuring a reliable, secure platform. --- Title: Which languages does Winston AI work with? Url: https://help.gowinston.ai/getting-started/which-languages-does-winston-ai-work-with --- ## Which languages does Winston AI support? Winston AI currently supports AI text detection in: * English * German * French * Dutch * Spanish * Portuguese * Italian * Polish * Indonesian * Tagalog * Chinese * Romanian * Russian * Ukrainian * Bulgarian * Azerbaijani This list applies specifically to [AI text detection](https://gowinston.ai/ai- content-detector/). Language availability for other features, such as plagiarism checking, writing feedback, and fact checking, may differ. ## Do I need to select the language before scanning? Winston AI detects the language of the submitted text automatically. For the clearest result, scan a document written primarily in one supported language rather than combining several languages in the same scan. ## Is AI detection equally reliable in every language? Detection performance can vary by language because writing patterns, available training data, and AI model behavior differ across languages. Text length and document type also matter. For the best result: * Scan the complete document or at least 300 words whenever possible. * Use continuous natural prose such as an essay, article, report, or full email. * Avoid drawing conclusions from a single sentence or short paragraph. * Review uncertain results alongside the original document and other relevant context. See [which content types work best with Winston AI](/understanding-winston- ai/what-types-of-content-can-i-scan-with-winston-ai) for detailed guidance. ## Can I scan translated text? Yes, if the translated text is in one of the supported languages. However, translation changes word choice, syntax, and sentence structure. These changes can remove or introduce patterns that affect the detection result. If authorship matters, scan the original-language document whenever it is available. Treat the score from a translated version as additional context rather than assuming it will match the original. ## What about documents containing more than one language? A mixed-language document may produce a less conclusive overall result, especially when each language appears only in a short passage. When practical, separate the document by language and scan each substantial section individually. Do not split the text into fragments that are too short to evaluate. A longer sample in one language is generally more informative than several isolated sentences. ## What if my language is not listed? Winston AI continues to expand its language support. If a language is not listed above, the text may still enter the scanner, but the result should not be treated as a validated detection result for that language. For help with a specific language or use case, contact [support@gowinston.ai](mailto:support@gowinston.ai). --- Title: Is Winston AI secure, and what compliance standards does it follow? Url: https://help.gowinston.ai/privacy-matters/is-winston-ai-secure-and-what-compliance-standards-does-it-follow --- Winston AI takes security and data privacy seriously. We adhere to high industry standards by being GDPR compliant, ensuring the privacy and protection of personal data, and are also SOC 2 Type 2 certified. This compliance demonstrates our commitment to implementing strong security practices and safeguarding sensitive information. With these measures, users can confidently rely on Winston AI for secure data handling. --- Title: How do I cancel my subscription? Url: https://help.gowinston.ai/account-management/how-do-i-cancel-my-subscription --- **Note: If you're on a free trial plan, there's no need to cancel your subscription. It will expire automatically at the end of the trial period, and you will not be charged.** If you no longer need your Winston AI plan, you can cancel your subscription quickly through the dashboard. After cancellation, your current plan remains active until the end of your billing cycle. Follow the steps below to end your subscription. ### Step 1: Navigate to “Manage Subscription” 1. Log in to your Winston AI account. 2. In the left-hand menu, scroll down to the **Billing** section and click on [**Manage subscription**](https://app.gowinston.ai/plans). You will be taken to a page that displays: * Your current plan (e.g., “Essential AI Plan”) * Your usage details (monthly credits, remaining top-ups, etc.) * Subscription plan pricing, billing period, and renewal date ### Step 2: Locate the “Cancel Subscription” button On the **Manage Subscription pag** e, you will see several options at the top: * **Cancel Subscription** * **Pause Subscription** * **Change Plan** To proceed with cancellation, click on the **Cancel Subscription** button. ### Step 3: Confirm your cancellation A confirmation prompt typically appears, asking if you are sure you want to cancel. The prompt may include details such as: * The date when your subscription will officially end * Any remaining credits or features you’ll still have access to until the end of the billing cycle Review the details, then confirm the cancellation. **Important** : After confirming, you will receive a **confirmation email** indicating that your cancellation has been scheduled. If you don't receive it, please check your inbox spam folder. ### Step 4: Verify your cancellation Once the cancellation is scheduled, you will still have access to your Winston AI features until the last day of your current billing period. On that final day (the end of your billing cycle), your subscription will officially end, and you’ll receive a **second confirmation email** notifying you that your plan has been canceled. * If you change your mind before the final day, you have to option to un-cancel your subscription, by clicking on "**Uncancel subscription** ". ### Additional Tips and FAQs #### Can I pause my subscription instead of canceling? Yes, Winston AI offers a Pause Subscription feature. This is useful if you only need to stop service temporarily. By pausing, you retain your account data and preferences without being charged for the subscription during the pause. #### Will I receive a refund for the remaining days in my billing cycle? In most cases, subscriptions remain active until the end of the current billing period, and no prorated refunds are offered. Check Winston AI’s refund policy or contact support for more details. #### How do I update my payment method or view invoices? Use the Payment methods and View invoices buttons on the same Manage Subscription page to update your card information or review billing history. --- Title: How to Delete Your Winston AI Account Url: https://help.gowinston.ai/account-management/privacy-matters/how-to-delete-your-winston-ai-account --- **Before deleting your account, please note:** * Your account will be placed in a deletion queue and permanently deleted **after 3 days**. During that period, it can be reactivated at any time. * This action permanently erases all your personal data and cannot be undone * You must cancel any active subscriptions first * Social login users (Google/Facebook) need to set a password before deletion ### Step-by-step instructions **1 - Log into your**[**Winston AI account**](https://app.gowinston.ai/login) **2 - Navigate to Account Settings** * Click your name in the top-right corner * Select "[Account](https://app.gowinston.ai/user/profile)" **3 - If using social login (Google/Facebook)** * Scroll to the "Update Password" section * Create a new password for your account **4 - If you have an active subscription** * Click your name in the top-right corner * Select "[Billing](https://app.gowinston.ai/plans)" * Cancel your subscription **5 - Delete your account** * Scroll to the bottom of the Account page * Find the "Delete Account" section * Click the red delete button * Enter your password to confirm ⚠️ Warning: Once confirmed, your account and all associated data will be placed in a deletion queue for 3 days, and then permanently deleted with no possibility of recovery. --- Title: How do I pause my subscription? Url: https://help.gowinston.ai/account-management/how-do-i-pause-my-subscription --- Winston AI’s Pause Subscription feature allows you to put your monthly subscription on hold for up to six months. During this paused period, **you will not be billed**. At the end of your selected pause duration, **billing will automatically resume**. You can also manually resume at any time if you wish to restart your subscription sooner. > **Note: The Pause Subscription feature is only available for monthly > subscriptions. If you are on an annual plan and would like to pause, please > contact Winston AI Support.** Below is a detailed guide on how the feature works, illustrated by the screenshots provided: ### Accessing the Pause Subscription Option 1. Log into your Winston AI account. 2. In the left-hand menu, navigate to **Manage subscription**. * This will take you to a page showing your current plan details, billing information, and available subscription actions (see Screenshot 1). 3. Locate and click the **Pause Subscription** button at the top of the subscription actions. ### Choosing Your Pause Duration 1. A pop-up window will appear (see Screenshot 2) asking: “For how many months would you like to pause your subscription?” 2. Select a duration from 1 to 6 months. 3. Review the **Pause** and **Resume** dates shown in the pop-up. 4. Click **Confirm** to finalize the pause. > **Note** : You will continue to have access to any remaining credits **only > until the end of your current billing cycle**. After this cycle concludes > and the pause officially starts, your credits will **reset to zero** until > your subscription resumes. ### During the Pause * **No Charges** : You will not be billed for monthly fees while your subscription is paused. * **Credits Reset** : Once the pause takes effect at your **next billing cycle** , your credits will be set to zero until your subscription resumes. ### Automatic Resumption * Once the selected pause duration (1–6 months) has elapsed, Winston AI **automatically resumes** your subscription. * You will be **billed again** starting from the day your subscription resumes. ### Resuming Early * If you want to **resume sooner** than the chosen pause duration, you can do so at any time. * When your subscription is paused, the **Pause Subscription** button is replaced by a **Unpause Subscription** button. * Simply click **Unpause Subscription** to reactivate your plan immediately. Your billing cycle will start again on the day you resume. * * * ### Frequently Asked Questions **Q: What happens if I reach the 6-month limit and still don’t want to resume?** A: Your subscription will automatically resume and begin billing again. If you do not wish to continue at that time, you can either cancel your subscription or contact Winston AI support. **Q: Can I pause more than once a year?** A: Yes, but each pause cannot exceed 6 months. If you need more flexibility, please contact our support team. **Q: Will I lose any stored documents or scans while my subscription is paused?** A: Your documents and scan history remain saved. You can resume later and pick up right where you left off. --- Title: How much does it cost? Url: https://help.gowinston.ai/pricing/how-much-does-it-cost --- ## How much does Winston AI cost? Winston AI offers a free 14-day trial and three paid plans. Paid plans can be billed monthly or annually. Annual billing provides the lowest effective monthly price. Plan| Monthly billing| Annual billing| Credits ---|---|---|--- **Free**| $0| Not applicable| 2,000 credits for 14 days **Essential**| $18 per month| $120 per year, equivalent to $10 per month| 100,000 credits per month **Advanced**| $29 per month| $192 per year, equivalent to $16 per month| 200,000 credits per month **Elite**| $49 per month| $312 per year, equivalent to $26 per month| 500,000 credits per month Prices above are the standard prices displayed in U.S. dollars at the time of this update. Visit the [Winston AI pricing page](https://gowinston.ai/pricing/) for the latest pricing, promotions, taxes, and complete feature comparison. ## What is included in the free trial? The Free plan provides 2,000 credits for 14 days so you can try Winston AI before choosing a paid plan. The free trial includes AI content detection and selected scanning features, but it does not include every paid-plan capability. ## Which paid plan should I choose? * **Essential:** Best for individuals who primarily need AI content detection, AI image detection, and writing feedback. * **Advanced:** Best for users who also need advanced plagiarism detection, more monthly credits, team access, or website certification features. * **Elite:** Best for higher-volume use, larger teams, and customers who need the largest included credit allowance and credit top-ups. Features can change as the product evolves, so use the live pricing comparison before purchasing if a specific capability is essential to your workflow. ## How do credits work? Credits are consumed when you run scans and selected analysis features. The cost depends on the amount of content and the tools enabled for that scan. Your available balance is shown inside your Winston AI account. Included monthly credits refresh at the start of each billing cycle. Unused monthly credits do not roll over. See [what happens to unused credits](/account-management/do-my-unused-credits-carry-over). ## Do I need a long-term contract? No. Monthly subscriptions can be cancelled without a long-term commitment. Annual plans are paid for the year and offer a lower effective monthly price. Read more about [monthly and annual commitments](/pricing/do-i-need-to-sign-a- long-term-contract). ## What if I need more credits or an institutional plan? Elite customers can purchase additional credits. Schools, businesses, and other organizations with larger requirements can [contact Winston AI sales](https://gowinston.ai/contact-sales/) for institutional or volume pricing. --- Title: How do we interpret the results from an AI text scan? Url: https://help.gowinston.ai/understanding-winston-ai/how-do-we-interpret-the-results-from-an-ai-text-scan --- ## How do I interpret a Winston AI text scan? Start with the overall **Human Score** , then use the AI Prediction Map and highlighted sentences to understand which parts of the document influenced the result. The overall score is the primary result because it evaluates the complete text. Sentence-level scores provide useful context, but they are based on much smaller samples and should not be interpreted on their own. ## What does the Human Score mean? The Human Score ranges from 0% to 100% and estimates how closely the complete document resembles verified human writing: * **A score near 100%:** The document strongly resembles human-written text. * **A score near 0%:** The document strongly resembles AI-generated text. * **A score near the middle:** The result is less conclusive. The document may contain mixed authorship, heavy editing, too little text, translated material, or writing that does not provide a strong signal in either direction. A score is a prediction, not a measurement of exactly how much of the document was written by AI. For example, a 20% Human Score does not mean that precisely 80% of the words were generated by AI. ## How do I read the AI Prediction Map? The AI Prediction Map divides the document into smaller sections and uses color coding to show which passages look more human or more AI-generated: * **0–20:** AI Generated * **20–40:** Likely AI * **40–60:** Uncertain * **60–80:** Mostly Human * **80–100:** Human Written Use the map to locate passages worth reviewing. Do not average the sentence scores or treat one highlighted sentence as proof of AI use. A short sentence contains less evidence than a full document, so its individual classification is naturally less reliable. ## What are the sections driving the AI score? This section lists sentences that had the strongest influence on the overall prediction. It is especially helpful when a document may combine human writing with AI-generated or heavily edited passages. Review these sentences in context. Formal language, repeated phrasing, boilerplate text, quotations, or a sudden change in style can affect an individual passage. The question is not only whether a sentence is highlighted, but whether the pattern across the document matches the available writing evidence. ## What does the Explain feature do? The **Explain** feature creates a plain-language overview of the scan. It summarizes the overall result and points out patterns that may help you review the document. This explanation is an aid, not a separate detection result. Base your interpretation on the Human Score, the original text, and the relevant context. ## Is the Readability Score part of AI detection? No. The Readability Score estimates how difficult the text is to read using the [Flesch-Kincaid formula](https://gowinston.ai/what-is-flesch-kincaid- grade-level/) and a U.S. grade-level label. It does not indicate whether the text was written by AI. ## What should I do with an uncertain or unexpected result? 1. **Scan more text.** Whenever possible, use the complete document or at least 300 words of natural prose. 2. **Check the content type.** Essays, articles, reports, and other continuous prose work best. Bullet lists, source code, tables, formulas, transcripts, translated writing, and legal boilerplate provide less reliable signals. 3. **Review highlighted passages.** Look for quotations, templates, abrupt style changes, or sections that may have been rewritten or pasted from another source. 4. **Consider the writing process.** For important decisions, examine drafts, notes, version history, sources, and the writer's explanation. 5. **Rescan when appropriate.** Adding more of the original document is more useful than repeatedly scanning the same short excerpt. See [which content types work best with Winston AI](/understanding-winston- ai/what-types-of-content-can-i-scan-with-winston-ai) for more detail. ## Can I make a decision from the score alone? For routine content checks, the Human Score is the main result to use. For high-stakes decisions, such as academic misconduct or employment action, use the result as one piece of evidence within a broader review. AI detection is a prediction task, so false positives and false negatives are possible, but they are unlikely given Winston AI's [independent benchmark scores](https://publicera.kb.se/ir/article/view/61629/53000). Review the highlighted text, supporting materials, and relevant context before reaching a conclusion. Learn more about [AI detection accuracy](/understanding-ai-detection/are-ai- detection-tools-accurate) or visit Winston AI's [Research & validation library](https://gowinston.ai/research-validation/). --- Title: What does Winston AI do exactly? Url: https://help.gowinston.ai/getting-started/what-does-winston-ai-do-exactly --- Winston AI is a comprehensive content verification and management platform, offering multiple tools to help businesses and educators maintain content authenticity and quality. Here's what Winston AI can do: **AI Detection** * Industry-leading AI-generated text detection with 99.98% accuracy * Support for detecting content from all major AI models (including ChatGPT, Claude, Gemini, and more) * AI image and deepfake detection — including images manipulated with tools like Photoshop. Available as a [Basic Scan or Advanced Scan](/understanding-winston-ai/what-types-of-images-can-i-scan-with-winston-ai) with full forensic analysis. * Real-time scanning and instant results **Content Verification** * Powerful plagiarism detection engine comparing content against billions of web pages * Cross-language plagiarism detection * Source attribution and citation tracking **Writing Enhancement** * Detailed writing feedback and suggestions * Style consistency checking * Grammar and readability analysis * Tone and voice recommendations **Document Management** * Secure cloud-based document storage * Team collaboration features * Advanced search and organization tools **Integrations** * Winston AI seamlessly integrates with popular platforms including: * Learning Management Systems (Google Classroom) * Content Management Systems (WordPress) * Browser Extensions (Chrome, Opera, Firefox) Whether you're an educational institution ensuring academic integrity, a content creator maintaining authenticity, or a business protecting your brand, Winston AI provides the tools you need to verify, manage, and improve your content. --- Title: How to Use Winston AI for Text Analysis Url: https://help.gowinston.ai/getting-started/how-to-use-winston-ai-for-text-analysis --- ## Overview Winston AI is a content analysis platform that lets you detect AI-generated text, check for plagiarism, get writing feedback, and fact-check claims — all from a single scan. This guide walks you through the complete flow, from starting a scan to reading your results, managing your documents, and understanding what each plan includes. ## Step 1: Start a New Text Scan Log in to your Winston AI account and click the **New text scan** button in the left sidebar. You will land on the scan editor — a two-panel layout with your text editor on the left and scan options on the right. Before adding text, fill in two optional fields at the top: * **Document title** — helps you find the scan later in your Projects list * **Document author** — useful if you are checking student submissions or managing multiple writers ## Step 2: Add Your Text You have five ways to get content into the editor: * **Paste directly** — click into the editor and paste your text (minimum 500 characters) * **File upload** — upload a single .docx, .pdf, or .txt file * **Import from Google Drive** — select a Google Doc, PDF, DOCX, or TXT file directly from your Google Drive * **Bulk upload** — upload multiple documents at once * **Import from URL** — enter a web address and Winston will pull the text from that page To import a file from Google Drive, click **File upload** , then select **Import from Drive** in the upload window. Choose your file and review the imported text before scanning, since its formatting may differ from the original document. Bulk upload is useful when you need to scan several files in the same project. Click **Bulk upload** , then drag and drop your files into the upload window or click to browse and select files from your computer. Bulk upload supports .docx, .pdf, .jpg, .png, and .tiff files, up to 10MB each, with a maximum of 50 files. Image files can be scanned with OCR so Winston can extract and analyze the text. As you add text, the counter at the bottom updates in real time: words, characters, and the estimated credit cost for your chosen features. ### Bulk Uploading Multiple Files After you add files, Winston shows them in a list with the title, author, word count, and status. You can edit the document title or author before scanning, or remove a file from the list. When the files are ready, click **Next** to choose which scan features to run. ## Step 3: Select Your Scan Features Below the editor, toggle any combination of five features on or off: * **AI detector** (1 credit/word) — detects AI-generated content and gives a Human Score * **Plagiarism** (2 credits/word) — checks for copied content across the web * **Fact checker** (1 credit/word) — verifies factual claims against external sources * **Writing feedback** (1 credit/word) — suggests grammar fixes, paraphrasing, and clarity improvements * **Essay grader** (1 credit/word) — grades an essay against a selected rubric and provides an overall score, criteria-level feedback, and writing suggestions The AI detector is enabled by default. The credit estimator updates immediately as you toggle features on and off. Not all features are available on every plan — see the plan comparison below. For bulk uploads, the feature selection step appears inside the upload window. Choose the scan features you want to run across the uploaded files, then click **Scan**. Winston will show the total credits required before you start the scan. ### Using Essay Grader To use Essay Grader, select the **Essay grader** scan feature, then click **Select rubric**. Winston opens a rubric selection window where you can choose from **Default Rubrics** or **My Rubrics**. You can search rubrics by name, filter them by language, and filter them by student level. Each rubric shows its language, student level, and number of criteria. Select the rubric you want to use, then click **Next**. ### Creating a Custom Rubric If you want to use your own rubric, click **Create** in the rubric selection window. Add a rubric name, choose the language and student level, then add your grading criteria. For each criterion, you can add marker guidance to explain what the AI should look for, then define the available grades and descriptions. Click **Create Rubric** when your rubric is ready. ### Essay Grader Settings After selecting a rubric, you can add optional grading settings to help Winston grade more accurately: * **Class Name** — the name of the class the essay is for * **Assignment Instructions** — the original instructions given to the student * **Grading Instructions** — any additional guidance for how the essay should be graded You can also save these settings to the project. When everything is ready, click **Confirm** , then run the scan. ## Step 4: Run the Scan Click **Scan**. Winston processes your text and loads the results into the right panel. Credits for all selected features are deducted at this point. After the scan is complete, Winston shows the results for each scan feature you selected, such as AI detection, plagiarism, readability, fact checking, writing feedback, or essay grading. For bulk uploads, Winston displays a progress window after you click **Scan**. Each document shows its scan status while it is processing. You can leave this window open, or close it and let the scans continue in the background. ## Reading Your Results Once the scan finishes, the left panel shows a read-only version of your text with basic stats (word count, character count, sentence count, reading time). The right panel displays results in a tabbed layout — one tab per feature selected. To edit the text after scanning, click the **Edit mode** toggle at the top of the left panel. ### AI Detector Tab **Human Score** — a percentage from 0% to 100%, with the model version used. A score near 0% means the content is almost certainly AI-generated; near 100% means it reads as human-written. **AI Prediction Map** — a sentence-by-sentence breakdown with color coding: * 0–20: AI Generated * 20–40: Likely AI * 40–60: Uncertain * 60–80: Mostly Human * 80–100: Human Written The prediction map scans smaller portions individually, so it is less accurate than the overall Human Score. Use the overall score as your primary indicator. **Readability Score** — a 0–100 score based on the [Flesch- Kincaid](https://gowinston.ai/what-is-flesch-kincaid-grade-level/) formula, with a U.S. school grade level label. You can also check the [readability score](https://gowinston.ai/readability-score-checker/) of any text using Winston's standalone readability checker. **Sections Driving AI Score** — a list of individual sentences with percentage scores, useful for identifying which sections are pulling the overall score down. ### Explain (AI Overview) Click the **Explain** button above the text panel to open the AI Overview panel. This generates a plain-language narrative that explains what the detection results mean — it summarizes the verdict, interprets the sentence- level scores, and if the content is flagged as AI-generated, offers suggestions for making the writing sound more human. Useful when sharing results with students or clients who need context beyond a raw percentage. ### Fact Checker Tab The fact checker extracts factual claims from your text and checks them against external sources. Each statement is classified as: * **Supported** — backed by credible sources * **Unsure** — could not be definitively verified * **Refuted** — contradicts available evidence Click any statement to expand it and view the explanation and source links. ### Writing Feedback Tab Writing feedback is broken into three categories: * **Grammar** — errors with a one-click Correct button * **Paraphrase** — rephrasing suggestions with a one-click Correct button * **Clarity** — structural guidance that requires manual revision ### Plagiarism Tab Shows an overall plagiarism percentage and a list of matched sources. Each match highlights the relevant text in the left panel. A result of 0% with zero matches means no plagiarism was detected. ### Essay Grader Tab When the essay grader scan is complete, Winston shows an Essay Grader tab. This includes the selected rubric, the overall score, and an overall review of the essay. Below that, the Criteria Breakdown shows each rubric criterion with its score and feedback. Some criteria may include Suggestions. Click the yellow suggestions to highlight the related text in the original document, so you can see exactly where the feedback applies. ## Managing Your Document and Project ### Actions Menu The **Actions** button (next to Download and Share) has three options: * **New document** — adds a new document to the current project * **Save document** — manually saves the current state * **Delete project** — permanently removes the project and all documents in it ### Renaming a Project Click the project title at the top of the page (shown as "Untitled - YYYY-MM- DD" by default) to rename it inline. ### Navigating Between Documents Use the **Previous document** and **Next document** arrows to move between documents in the same project. The list icon in the top-right area opens **Project documents** , where you can see all documents in the current project and their scan status. When a project has more than one document, the list icon shows a badge with the document count. ### Viewing Bulk Upload Progress When you close the bulk upload progress window, the scans continue in the background. Click the list icon in the top-right area to open **Project documents** and see every document in the project, including each document’s scan progress and available actions. Once a document finishes scanning, click the eye icon to open and review its results. ## Downloading and Sharing Results **Download** — exports a PDF report. Choose which sections to include (AI detector, Plagiarism, Writing feedback, Fact checker), or a separate Essay Grader report. All sections are selected by default. **Share** — toggle on Link sharing to generate a shareable URL. Recipients get read-only access to your scan results. Link sharing is off by default. ## Rescanning The **Rescan** button at the bottom of any results tab lets you re-run the scan after editing — useful for checking whether revisions improved the score. Credits are deducted again. ## Viewing Past Scans All scans are saved automatically and accessible from: * **Documents** in the left sidebar — your most recent documents * **Projects** in the left sidebar — full list of all projects, sorted by date On the Projects page you can sort projects, create a new one with the **New project** card, and delete projects using the delete button on each card. ## Bonus Tools Two additional tools are available under **More** in the left sidebar (under Tools): **Grammar Checker** — a standalone grammar, spelling, and style checker. Available in English (American, Canadian, British, and Australian). No credits required. **Text Compare** — paste two versions of a text side by side and click Compare texts to highlight differences and get a similarity score. Cost: 1/4 credit per word, minimum 100 characters per box. ## What's Available on Each Plan Not all features are available on every plan: Feature| Free Trial| Essential AI| Essential Plagiarism (discontinued)| Advanced| Elite ---|---|---|---|---|--- AI detector| Yes| Yes| No| Yes| Yes Plagiarism checker| No| No| Yes| Yes| Yes Writing feedback| No| Yes| Yes| Yes| Yes Fact checker| No| Yes| Yes| Yes| Yes Essay grader| No| Yes| Yes| Yes| Yes AI Overview (Explain)| No| Yes| Yes| Yes| Yes Image scan| Yes| Yes| No| Yes| Yes Website scan| Yes| Yes| Yes| Yes| Yes Google Classroom| No| Yes| Yes| Yes| Yes HUMN-1 certification| Yes| No| No| Yes| Yes HUMN-2 certification| No| No| No| No| Yes Shareable reports| No| Yes| Yes| Yes| Yes Credit top-up| No| No| No| No| Yes Team members| 0| 0| 0| 5| Unlimited Monitored websites| 0| 0| 0| 1| 1 ### Choosing Between Essential AI and Essential Plagiarism The Essential AI plan and the Essential Plagiarism (discontinued) plan are not interchangeable — they were designed for different use cases. Essential AI focuses on AI detection; the Essential Plagiarism (discontinued) plan replaces AI detection with plagiarism checking. If you need both, you will need to upgrade to Advanced or Elite. ### Credit Top-Ups (Elite Only) On Essential AI, the Essential Plagiarism (discontinued) plan, and Advanced plans, once your monthly credits are used up, you wait until the next billing cycle. **Elite plan users can purchase additional credits at any time** using the Get more credits link in the top navigation bar — no waiting required. --- Title: Why should I get a HUMN-1 certification for my website? Url: https://help.gowinston.ai/humn-1-website-certification/why-should-i-get-a-humn-1-certification-for-my-website --- Here’s why getting certified is more important than ever:- Build trust with your audience. In a world filled with AI-generated content from tools like ChatGPT, let your users know they’re reading genuine human-created material. * Boost your search engine ranking. Google and other search engines are prioritizing human-authored content over AI-generated text. * Stay ahead of the competition. The web is crowded with low-quality synthetic content. Ensure your articles and landing pages outshine those of competitors relying on shortcuts. --- Title: How can I get a HUMN-1 certification for my website or blog? Url: https://help.gowinston.ai/humn-1-website-certification/how-can-i-get-a-humn-1-certification-for-my-website-or-blog --- To obtain your HUMN-1 certification, just sign up on our website and choose how you’d like to connect your site (either through a WordPress plugin or Sitemap). After your site is connected, we’ll quickly run an audit that takes just a few minutes. Once it's done, you'll receive your certification. If you need help, our team is ready to assist. --- Title: What is the HUMN-1 certification? Url: https://help.gowinston.ai/humn-1-website-certification/what-is-the-humn-1-certification --- HUMN-1 is the first human content certification for websites and blogs. Leveraging Winston AI’s advanced detection model, it ensures that the content on any site is both human-created and original. --- Title: Does Winston AI save my scanned documents? Url: https://help.gowinston.ai/privacy-matters/does-winston-ai-save-my-scanned-documents --- ## Does Winston AI save my scanned documents? Yes. Winston AI stores submitted text and images with their analysis results so that reports remain available in your dashboard. Your content stays associated with your account while the corresponding report exists. ## Why is submitted content stored? Storage allows you to return to previous scans, review the results, and manage reports from your account. Submitted content is processed to provide the analysis features you select, such as AI detection, plagiarism checking, fact checking, writing feedback, grammar analysis, image analysis, or essay grading. ## What happens when I delete a report? When you delete a report, the submitted content associated with that report is deleted from active storage within 30 days. The deleted content may remain in encrypted backup systems for up to 90 additional days. After that period, it is permanently purged unless Winston AI is legally required to retain it for longer. ## What happens when I delete my account? Deleting your account also removes its submitted content and analysis reports from active storage within 30 days. The same encrypted-backup period of up to 90 additional days applies. For account deletion instructions, see [How to Delete Your Winston AI Account](/account-management/how-to-delete-your-winston-ai-account). ## How are saved documents protected? User-submitted content is encrypted at rest and in transit and stored in access-controlled cloud systems. Access is restricted to authorized personnel on a strict need-to-know basis. For more information, read [Are my documents private?](/privacy-matters/are- my-documents-private) and the full [Winston AI Privacy Policy](https://gowinston.ai/privacy-policy/). --- Title: Does Winston AI share the scan results with any of its partners of affiliates? Url: https://help.gowinston.ai/privacy-matters/does-winston-ai-share-the-scan-results-with-any-of-its-partners-of-affiliates --- Winston AI doesn't share any of the data scanned on its platform with any other parties, businesses or affiliates. --- Title: Does Winston AI use my scanned documents to improve its data? Url: https://help.gowinston.ai/privacy-matters/does-winston-ai-use-my-scanned-documents-to-improve-its-data --- No. Winston AI doesn't use user's scanned documents to improve its capabilities in any way. Any documents or content you scan on Winston AI is 100% secure, encrypted and confidential. Winston AI doesn't use any of the scanned documents to improve its models. --- Title: Do I need to sign a long term contract? Url: https://help.gowinston.ai/pricing/do-i-need-to-sign-a-long-term-contract --- Absolutely not! If you have a monthly plan, you can cancel at any time. You also have annual plans that include a discount. --- Title: Who can use Winston AI? Url: https://help.gowinston.ai/basic-information/who-can-use-winston-ai --- ## Who can use Winston AI? Winston AI is used anywhere content authenticity matters: * **Educators & academic integrity teams** — Verify student essays, capstones, discussion posts, and LMS submissions. * **Publishers & SEO agencies** — Ensure freelance articles are genuinely human-written and not spun from AI or plagiarized sources. * **Marketing & communications teams** — Audit campaign copy, product descriptions, press releases, and thought-leadership content before publication. * **Legal & compliance** — Check policy documents, contracts, and regulated communications for originality and human authorship. * **Enterprises building AI governance programs** — Combine AI detection, plagiarism checks, writing feedback, and HUMN certification to enforce internal policies. If you work with written content — from blog posts to grant applications — Winston AI gives you a defensible way to prove the content’s origin. --- Title: What's the difference between Plagiarism Detection and AI detection? Url: https://help.gowinston.ai/understanding-winston-ai/whats-the-difference-between-plagiarism-detection-and-ai-detection --- ## What's the difference between Plagiarism Detection and AI detection? Plagiarism detection and AI detection solve two different problems: | Plagiarism detection| AI detection ---|---|--- **Purpose**| Finds text that already exists elsewhere.| Identifies text generated by AI models. **How it works**| Compares your document against databases, academic journals, and the open web.| Uses language models trained on billions of human + AI samples to spot statistical AI patterns. **Best for**| Academic integrity, SEO duplicate-content checks.| Verifying originality of freshly written content. **Limitations**| Won’t flag AI text if it isn’t copied.| Won’t flag human- written text that was copied from another source. Use plagiarism detection when you suspect the text was copied. Use AI detection when you suspect the text was drafted by ChatGPT, Claude, Gemini, or another model — even if it’s original wording. --- Title: What is an AI content detector? Url: https://help.gowinston.ai/understanding-ai-detection/what-is-an-ai-content-detector --- ## What is an AI content detector? An AI content detector is a tool that analyzes writing and estimates whether it was written by a person or generated by an AI system such as ChatGPT, Claude, or Gemini. It looks for patterns in the language rather than searching for a hidden label or matching the text against a database of AI responses. AI detection is useful anywhere authorship and content authenticity matter, including education, publishing, recruitment, marketing, and research. ## How does an AI content detector work? An AI detector uses a machine-learning model trained on large collections of verified human writing and AI-generated text. The model learns combinations of patterns associated with each category and compares new writing with what it learned during training. These patterns may include word choice, sentence structure, variation in rhythm and phrasing, and relationships between ideas across the document. No single word or writing habit proves that text was generated by AI. The prediction comes from evaluating many signals together. For a deeper explanation, read [how AI detectors work](/understanding-ai- detection/how-do-ai-detectors-work). ## What does Winston AI show? Winston AI's [AI content detector](https://gowinston.ai/ai-content-detector/) provides: * **Human Score:** The overall prediction for the complete document, from text that strongly resembles AI-generated writing to text that strongly resembles human writing. * **AI Prediction Map:** A visual breakdown showing which passages contributed to the result. * **Sections driving the AI score:** Specific sentences that had a strong influence on the prediction. * **Explain:** A plain-language overview that helps users understand the result and what to review next. The overall Human Score is the primary result. Sentence-level classifications provide context, but they are based on smaller samples and should not be treated independently. ## Is AI detection the same as plagiarism detection? No. A plagiarism checker looks for text that matches existing sources. An AI detector looks for language patterns associated with generated writing. AI-generated text may be completely new and produce no plagiarism match. Human-written text can also be plagiarized. The two tools answer different questions, so a complete originality review may use both. ## Who uses AI content detectors? * **Educators** review assignments when course rules limit or require disclosure of AI assistance. * **Publishers and editors** verify submitted articles and freelance work before publication. * **Businesses** check marketing, communications, applications, or other content where authorship matters. * **Researchers** study the adoption and effects of generated text at scale. For example, researchers from Yale, Northeastern University, the Chinese University of Hong Kong, and City University of Hong Kong used Winston AI while studying more than 1.1 million U.S. consumer complaints in a [peer- reviewed Nature Human Behaviour study](https://www.nature.com/articles/s41562-026-02409-4). Winston AI is also among the tools discussed in our guide to [AI detectors used by universities](https://gowinston.ai/what-ai-detectors-do-universities-use/). ## What affects an AI detection result? Text length and content type matter. Full essays, articles, reports, and other natural prose provide more evidence than a short paragraph. Bullet lists, tables, source code, formulas, transcripts, translated writing, and heavily edited text can produce less conclusive results. Whenever possible, scan the complete document or at least 300 words. See [which content types work best with Winston AI](/understanding-winston- ai/what-types-of-content-can-i-scan-with-winston-ai). ## How should I use an AI detector responsibly? Treat the result as evidence to review, not automatic proof about a writer's behavior. For important decisions, examine the highlighted passages, scan a longer sample when possible, and consider supporting information such as drafts, version history, sources, and the writer's explanation. An AI content detector is most useful when it helps you ask better questions and focus a review. Human judgment remains essential, especially in academic, employment, or disciplinary contexts. --- Title: How do AI detectors work? Url: https://help.gowinston.ai/understanding-ai-detection/how-do-ai-detectors-work --- ## How do AI detectors work? AI detectors analyze patterns in a piece of writing and estimate whether the text is more likely to have been written by a person or generated by an AI model. They do not search for a hidden ChatGPT label or compare the text with a database of every AI response. Instead, they use machine-learning models trained to recognize differences between human and AI-generated language. ## What patterns do AI detectors analyze? AI-generated and human-written text can differ in subtle, measurable ways. A detector may evaluate patterns such as: * **Word and phrase choices:** AI systems predict likely sequences of words, which can make their output more statistically regular. * **Sentence structure:** Sentence length, syntax, and complexity may follow different patterns in human and AI writing. * **Variation across the document:** Human writing often contains more irregular changes in tone, rhythm, and phrasing. * **Contextual relationships:** Modern detectors evaluate how words, sentences, and ideas relate to one another, not just whether particular words appear. No single pattern proves that text is AI-generated. Reliable detection depends on evaluating many signals together across the full sample. ## How does the detector classify the text? The detection model is trained on large collections of verified human writing and AI-generated text. During training, it learns which combinations of linguistic patterns are most strongly associated with each category. When new text is submitted, the detector compares its patterns with what the model learned and produces a prediction. Winston AI presents this prediction as a **Human Score**. A higher Human Score means the text more closely resembles verified human writing. The AI Prediction Map also highlights sentences that contributed to the overall result, helping users examine the document in context rather than relying only on one number. ## What affects the reliability of an AI detection result? The quality of a result depends on the text being analyzed: 1. **Text length:** Longer samples provide more evidence. Whenever possible, scan at least 300 words rather than a short sentence or paragraph. 2. **Document type:** Essays, articles, reports, and other natural prose work better than source code, tables, bullet lists, formulas, transcripts, or highly structured text. 3. **Editing:** Heavy rewriting, paraphrasing, translation, or a mixture of human and AI writing can affect the patterns in the document. 4. **Language:** Detection performance can vary by language and by the training data available for that language. For more detail, see [which content types work best with Winston AI](/understanding-winston-ai/what-types-of-content-can-i-scan-with-winston- ai). ## Can an AI detector be wrong? Yes. AI detection is a prediction task, so false positives and false negatives are possible, but they are unlikely given Winston AI's [independent benchmark scores](https://publicera.kb.se/ir/article/view/61629/53000). A result should be treated as evidence to review, not automatic proof of misconduct. For important decisions, examine the highlighted sentences, scan a longer sample when possible, consider the writer's process and supporting work, and use human judgment. Read [how AI detector accuracy is measured](/understanding-ai-detection/are- ai-detection-tools-accurate) and review Winston AI's independent studies and technical evaluations in the [Research & validation library](https://gowinston.ai/research-validation/). ## How should I interpret a Winston AI result? Start with the overall Human Score, then review the AI Prediction Map to see which passages influenced it. A strong result across a full-length document is more informative than a score from a short excerpt. If the result is borderline or unexpected, add more of the original document and scan again before drawing a conclusion. For a step-by-step explanation, see [how to interpret the results from an AI text scan](/understanding-winston-ai/how-do-we-interpret-the-results-from-an- ai-text-scan). --- Title: Are AI detection tools accurate? Url: https://help.gowinston.ai/understanding-ai-detection/are-ai-detection-tools-accurate --- ## Are AI detection tools accurate? Yes, strong AI detection tools can be highly accurate, but accuracy varies by detector, model release, dataset, text length, and document type. The best way to assess an AI detector is to look at independent research alongside transparent internal evaluations, rather than relying on a single headline percentage. Winston AI publishes its evidence in a central [Research & validation library](https://gowinston.ai/research-validation/), including peer-reviewed studies, technical evaluation reports, and independent press coverage. ## What independent research says about Winston AI Several research teams have evaluated Winston AI independently: * A peer-reviewed [Information Research study from the University of J.J. Strossmayer in Osijek](https://publicera.kb.se/ir/article/view/61629) compared four AI detectors using human-written and AI-generated texts. Winston AI recorded the highest average in the study's standardized accuracy table at 99%, ahead of Originality.ai at 98%, and ZeroGPT and Smodin at 91%. * Researchers from [Yale, Northeastern University, the Chinese University of Hong Kong, and City University of Hong Kong](https://www.nature.com/articles/s41562-026-02409-4) used Winston AI while analyzing more than 1.1 million US consumer complaints. This is an example of the detector being used at large scale in academic research. * A peer-reviewed [medical-education study](https://www.cureus.com/articles/370285-can-residency-programs-detect-artificial-intelligence-use-in-personal-statements) found that Winston AI clearly distinguished AI-generated residency personal statements from verified human writing in a real-world academic setting. These studies are independent of Winston AI. Each study has its own dataset and methodology, so its results should be interpreted within that context. ## How Winston AI validates its own models Internal evaluation is separate from independent validation. Winston AI publishes technical reports explaining what changed in major model releases and how each release was tested. * The [Curia technical report](https://gowinston.ai/releasing-model-4-0-curia-our-most-advanced-ai-text-detection-system-yet/) reports 99.95% overall classification accuracy on a 10,000-sample English dataset. It also reports stronger human-text performance and an R² of 0.9908 when estimating the proportion of AI-generated text. * The earlier [Luka technical report](https://gowinston.ai/setting-new-standards-in-ai-content-detection/) documented a 10,000-text dataset split evenly between human-written and AI-generated content. The published results reported 99.98% AI detection accuracy, 99.50% human detection accuracy, and 99.74% overall classification accuracy. Publishing the dataset size, task, and metrics makes the results easier to evaluate than a percentage presented without context. ## Independent press and third-party reviews Winston AI has also been used or evaluated by independent media and technology publications: * [Snopes](https://www.snopes.com/fact-check/real-pic-of-crowd-harris-walz-rally-in-detroit/) and the [Daily Mail](https://www.dailymail.co.uk/sciencetech/article-13736543/AI-experts-reveal-image-Harris-rally-edited-donald-Trump-claimed.html) used Winston AI analysis while investigating a disputed presidential rally image. * [WIRED](https://www.wired.com/story/amazon-flag-ai-generated-books/) featured Winston AI in its reporting on AI-generated books and publishing integrity. * [Zapier](https://zapier.com/blog/ai-content-detector/) selected Winston AI as its best AI detector for integrations after hands-on testing. * [Elegant Themes](https://www.elegantthemes.com/blog/business/winston-ai) ranked Winston AI first among nine AI detectors, while [Unite.AI](https://www.unite.ai/winston-ai-review/) highlighted its accuracy in tests using human and AI-written text. Press coverage and product reviews are not substitutes for peer-reviewed validation, but they provide additional evidence of real-world use and independent testing. ## What affects the accuracy of an AI detection result? Three factors matter most: 1. **Text length:** Longer samples give the detector more linguistic evidence. Very short snippets produce less reliable results, so scan at least 300 words whenever possible. 2. **Document type:** Essays, academic papers, blog posts, articles, full emails, and other long-form natural prose generally produce the most reliable results. Short social posts, bullet lists, transcripts, and translated text provide less signal. Source code, formulas, legal boilerplate, tables, and structured data are not suitable for text detection. [See which content types work best with Winston AI](/understanding-winston-ai/what-types-of-content-can-i-scan-with-winston-ai). 3. **Human editing:** Heavy rewriting or AI humanizer tools can change the patterns a detector evaluates and may affect the score. ## How should I use the result? No AI detector should be treated as an automatic guilty verdict. Use Winston AI's Human Score and sentence-level AI Prediction Map alongside the source material, writing history, and your own judgment. If a result is borderline, scan a longer sample and look for additional context before making a decision. For the complete set of studies, evaluation reports, and coverage, visit Winston AI's [Research & validation page](https://gowinston.ai/research- validation/). --- Title: Can teachers detect if you used an AI assistant like ChatGPT for schoolwork? Url: https://help.gowinston.ai/understanding-ai-detection/can-teachers-detect-if-you-used-an-ai-assistant-like-chatgpt-for-schoolwork --- ## Can teachers detect AI use in schoolwork? Yes. Teachers can often identify signs that an assignment may include AI- generated writing, but no single sign proves that a student used AI. The fairest approach is to review several forms of evidence together: the assignment rules, the student's writing process, their understanding of the work, the accuracy of their sources, and an AI detection result. ## What signs might lead a teacher to review an assignment? A teacher may notice: * A sudden and unexplained change in vocabulary, tone, or writing ability * Generic analysis that does not address class discussions or the specific assignment * Sources, quotations, or facts that cannot be verified * A document history showing that most of the assignment appeared in one large insertion * A student who cannot explain important ideas or writing choices in their submission Each sign can also have an innocent explanation. A student may have used a writing center, drafted in another application, received permitted feedback, or simply improved. These signs should begin a review, not determine its outcome. ## How should a teacher review possible AI use? 1. **Check the assignment policy.** Confirm what kinds of AI assistance were allowed. Brainstorming, grammar feedback, and generating a complete answer are different uses and may be treated differently by the school. 2. **Review the writing process.** Ask for outlines, notes, drafts, research records, or document version history. Look for a reasonable path from the student's ideas to the final submission. 3. **Talk with the student.** Ask the student to explain the argument, define an important term, discuss a source, or describe how a paragraph developed. The goal is to understand their process, not force a confession. 4. **Verify citations and claims.** Check that sources exist, quotations are accurate, and each source supports the claim attached to it. 5. **Use an AI detector as supporting evidence.** Scan the complete assignment when possible and review the passages highlighted by the detector alongside the other evidence. ## How can Winston AI help? Winston AI analyzes patterns in writing and provides an overall **Human Score**. Its AI Prediction Map highlights sentences that contributed to the result, helping teachers focus their review on specific passages rather than relying only on a single percentage. For the most reliable result, scan the full assignment or at least 300 words of natural prose. Short excerpts, bullet lists, source code, translated text, and heavily edited writing provide less useful evidence. See [which content types work best with Winston AI](/understanding-winston-ai/what-types-of- content-can-i-scan-with-winston-ai). ## Can an AI detector be wrong? Yes. AI detection is a prediction task, so false positives and false negatives are possible, but they are unlikely given Winston AI's [independent benchmark scores](https://publicera.kb.se/ir/article/view/61629/53000). A detection result should be treated as evidence to review, not automatic proof of misconduct. If a result is unexpected, scan a longer sample, review the highlighted passages, and consider the student's drafts, sources, version history, and explanation. Read more about [AI detection accuracy](/understanding-ai- detection/are-ai-detection-tools-accurate) and Winston AI's independent studies in the [Research & validation library](https://gowinston.ai/research- validation/). ## What should happen after a paper is flagged? Teachers should follow their institution's academic-integrity process and give the student a fair opportunity to respond. A strong review considers all available evidence and documents the reasons for the final decision. If the student's drafts, sources, and explanation support their authorship, that context should carry significant weight. If several independent findings point toward prohibited AI use, the teacher can proceed according to the school's policy. ## How can students show that their work is their own? Students can keep outlines, notes, drafts, source records, and version history. They should follow the course's AI policy and disclose permitted AI assistance when required. Being able to explain the argument, evidence, and revision process is often the clearest demonstration of genuine authorship. --- Title: Does Google detect AI content? Url: https://help.gowinston.ai/understanding-ai-detection/does-google-detect-ai-content --- ## Does Google detect AI-generated content? Google has systems designed to identify low-quality content and search spam, including content produced at scale with generative AI. However, Google does not publish a simple AI-detection score for individual pages or disclose every signal its ranking and spam systems use. The more important distinction is that detecting AI use and deciding how a page should rank are not the same task. Google says that appropriate use of AI is not automatically against its guidelines. Its systems focus on whether content is helpful, reliable, original, and created primarily for people. ## Does Google penalize content simply because AI helped create it? No. Google does not say that all AI-assisted or AI-generated content receives a ranking penalty. Its [guidance on generative AI content](https://developers.google.com/search/docs/fundamentals/using-gen-ai- content) explains that AI can be useful for research and structure, but generating many pages without adding value may violate its spam policies. A useful page can involve AI assistance and still perform well if a knowledgeable person verifies the facts, adds original information, improves the analysis, and takes responsibility for the final result. A weak page can perform poorly even when every sentence was written by a person. For a fuller explanation, see Winston AI's guide to whether [Google penalizes AI content](https://gowinston.ai/google-penalize-ai-content/). ## When does AI content become search spam? Google's [scaled content abuse policy](https://developers.google.com/search/docs/essentials/spam- policies#scaled-content) applies when large numbers of pages are created primarily to manipulate rankings rather than help users. The policy can apply whether the content was produced by AI, by people, or through a combination of both. Examples of risky practices include: * Publishing large numbers of pages that repeat information already available elsewhere * Using AI to rewrite or combine sources without adding original value * Creating pages for search queries while giving readers little useful information * Publishing unverified claims, invented sources, or inaccurate summaries * Using automation mainly to acquire traffic rather than serve an audience ## What does Google want from AI-assisted content? Google's [helpful content guidance](https://developers.google.com/search/docs/fundamentals/creating- helpful-content) asks publishers to create people-first pages that demonstrate experience, expertise, authority, and trust. In practical terms: 1. **Add original value.** Include firsthand experience, testing, examples, data, or analysis that is not simply a summary of other pages. 2. **Verify the facts.** Check every statistic, quotation, source, and factual claim before publishing. 3. **Use qualified review.** Have someone who understands the subject edit the page and take responsibility for it. 4. **Serve a real audience.** Create the page because it answers a useful question, not only because a keyword has search volume. 5. **Be transparent when appropriate.** If automation played a meaningful role, context about how the content was created can help readers understand why it is trustworthy. ## Can an AI detector tell me whether Google will rank a page? No. An AI detector estimates whether writing resembles human or AI-generated text. It does not predict Google rankings, measure content quality, or determine whether a page violates a search policy. Winston AI's [AI content detector](https://gowinston.ai/ai-content-detector/) can help publishers audit authorship and identify passages that may need closer editorial review. SEO decisions should also consider originality, factual accuracy, user value, search intent, and the site's overall quality. ## What is the bottom line? Google can identify patterns associated with low-quality automated content, but AI use alone is not what determines whether a page ranks. Helpful, original, carefully reviewed content is the goal. Mass-produced pages created mainly to manipulate search results are the risk. For more examples, read Winston AI's detailed guide: [Can Google detect AI content?](https://gowinston.ai/can-google-detect-ai-content/) --- Title: Are my documents private? Url: https://help.gowinston.ai/privacy-matters/are-my-documents-private --- ## Are my documents private? Yes. Documents and images submitted to Winston AI are stored in encrypted, access-controlled cloud storage and are used to provide the analysis features you request. Submitted text is not shared with or compared against other users' submissions. As with any cloud service, no internet transmission can be guaranteed to be completely secure. Winston AI uses technical and organizational safeguards designed to protect uploaded content and personal information. ## Why does Winston AI store submitted content? Submitted content and its analysis results are stored so that your reports remain available in your Winston AI dashboard. This lets you return to previous scans, review results, and manage your documents. Your content may be processed to provide requested features such as AI detection, plagiarism checking, fact checking, writing feedback, grammar analysis, image analysis, and essay grading. ## How is my content protected? * **Encryption:** User-submitted content is encrypted at rest and in transit. * **Access controls:** Access is restricted to authorized Winston AI personnel on a strict need-to-know basis. General employee access to user content is not permitted. * **Security reviews:** Winston AI conducts periodic reviews and assessments of its infrastructure and data-handling practices. * **Service providers:** Providers that handle data are selected for their security practices and are contractually required to maintain appropriate safeguards. Winston AI is also SOC 2 Type 2 and CASA Type 2 certified. See [Winston AI's data-handling certifications](/privacy-matters/does-winston-ai-have-any- certifications-for-data-handling). ## Does Winston AI use my documents to train AI models? Winston AI does not use individually identifiable submitted content to train AI models without your explicit consent. Aggregated and de-identified usage analytics may be used to improve platform performance and analysis accuracy. ## Does Winston AI share my documents? Submitted text is not shared with other Winston AI users or compared against their submissions. Information may be handled by contracted service providers when necessary to operate and secure the service. It may also be disclosed when required by law, to protect legal rights or safety, or in other circumstances described in the privacy policy. Winston AI does not use Google Workspace API data for advertising, sell that data, or use it to train generalized AI or machine-learning models. ## What happens when I delete a report or my account? * If you delete a report, its submitted content is deleted from active storage within 30 days. * If you delete your account, submitted content and associated reports are deleted from active storage within 30 days of account deletion. * Deleted content may remain in encrypted backup systems for up to 90 additional days before it is permanently purged. Content may be retained longer when required by law, such as in response to a valid preservation order. For complete details, read the [Winston AI Privacy Policy](https://gowinston.ai/privacy-policy/). The privacy policy controls if there is any difference between this summary and the full legal notice. --- Title: What types of content can I scan with Winston AI? Url: https://help.gowinston.ai/understanding-winston-ai/what-types-of-content-can-i-scan-with-winston-ai --- ## Overview Winston AI's detection engine is trained on a wide range of written content, but not every type of text produces equally reliable results. Understanding what works well — and what doesn't — helps you get the most accurate scans and set the right expectations when interpreting results. ## Minimum Requirements Before scanning, your text must meet a minimum threshold: * **Minimum 500 characters** to run a scan * The more text you provide, the more accurate the results * For best results, aim for **300 words or more** Short text snippets (a few sentences) can technically be scanned, but the prediction confidence will be lower. The overall Human Score is more reliable than the sentence-level AI Prediction Map, especially on shorter passages. ## Content Type Reference Content Type| Works Well?| Notes ---|---|--- Essays and academic papers| ✅ Best results| Long-form prose is ideal. The more complete the document, the more accurate the score. Blog posts and articles| ✅ Best results| Full articles with natural prose produce highly reliable detection. Cover letters and personal statements| ✅ Good| Works well. Short length may reduce confidence slightly. Email body content| ✅ Good| Full email body works well. Subject lines alone are too short. Product descriptions| ✅ Good| Works well when descriptive prose is present. News and press releases| ✅ Good| Standard journalistic prose scans reliably. Social media posts| ⚠️ Limited| Usually too short for reliable detection. Treat results as indicative only. Bullet-point lists| ⚠️ Limited| Minimal prose means less signal for the model. Works better when mixed with full sentences. Translated content| ⚠️ Limited| Translation alters writing patterns significantly. Results may be less reliable. Heavily edited AI drafts| ⚠️ Limited| Significant human editing can lower the AI score even on AI-generated originals. Source code| ❌ Not suitable| Code does not follow natural language patterns. Do not use the text scanner for code. Mathematical formulas| ❌ Not suitable| Symbolic and numeric content lacks the prose characteristics the model relies on. Boilerplate legal text| ❌ Not suitable| Highly standardized language that looks similar whether written by a human or AI. Results are unreliable. Tables and structured data| ❌ Not suitable| Non-prose content does not provide meaningful detection signal. Transcripts (verbatim speech)| ⚠️ Limited| Spoken language patterns differ significantly from written prose. Results may skew toward human-written even for AI-generated scripts. ## Factors That Affect Accuracy ### Text Length The single biggest factor is length. Longer texts give the model more signal to work with. A 50-word snippet may return an uncertain result; the same content expanded to 500 words will produce a much more reliable score. If you receive an uncertain result on a short document, consider scanning a longer version if available. ### Paraphrasing and Humanization Tools AI content that has been run through a paraphrasing tool or "AI humanizer" will often score higher on the Human Score than unmodified AI output. Winston AI is specifically trained to detect humanized content, but heavily rewritten text can reduce confidence. A moderately elevated Human Score on paraphrased content does not necessarily mean the text is human-written. ### Mixed Authorship Documents where a human has edited AI-generated content — or vice versa — produce mixed signals. The overall Human Score will reflect the blend. The AI Prediction Map's sentence-level breakdown is the best tool for identifying which specific sections are driving the AI score in these cases. ## Tips for Best Results * Submit the full document rather than excerpts when possible * Remove headers, footnotes, and bibliographies before scanning — these can introduce noise * If scanning a translated document, note that translation alone can significantly alter AI detection scores * Use the sentence-level AI Prediction Map to investigate specific sections rather than relying solely on the overall score for mixed-authorship documents * When in doubt, use the **Explain** button for an AI-generated narrative that contextualizes the result --- Title: What types of images can I scan with Winston AI? Url: https://help.gowinston.ai/understanding-winston-ai/what-types-of-images-can-i-scan-with-winston-ai --- ## Overview Winston AI's image detection engine analyzes photos and images to determine whether they were created by a human or generated by an AI tool. It also identifies deepfakes and images that have been manipulated using tools like Photoshop or other editing software. Two scan types are available depending on how much detail you need. ## Basic Scan vs. Advanced Scan ### Basic Scan **Cost: 200 credits** The Basic Scan runs an AI image classifier and returns: * **AI generation probability score** — the likelihood the image was AI-generated vs. human-created * **Image metadata** — any available Exif, C2PA, and IPTC data embedded in the image Results are displayed inline in the app and can also be downloaded as a PDF report. This is a fast, lightweight check suitable for quickly screening images when you don't need forensic detail. After completing a Basic Scan, you can upgrade to an Advanced Scan for an additional 300 credits (500 credits total). ### Advanced Scan **Cost: 500 credits** The Advanced Scan runs a full forensic analysis using multiple detection methods and produces a written conclusion. It includes: * **AI Image Classifier** — same probability score as the Basic Scan * **Metadata Extractor** — reads C2PA manifests, IPTC, and Exif data; C2PA manifests from tools like ChatGPT/OpenAI provide conclusive provenance evidence when present * **ELA (Error Level Analysis)** — detects post-generation pixel-level manipulation or splicing * **Residual Noise Maps** — identifies noise patterns typical of synthetic texture generation * **Edge Anomaly Heat Map** — highlights edge inconsistencies that may indicate compositing * **CFA Pattern Analysis** — checks for the Bayer/CFA sensor pattern present in real camera images; absence is a strong indicator of AI generation The Advanced Scan also provides a written **Conclusion** synthesizing all findings — including cases where individual tools disagree (e.g. the classifier returns a false negative but metadata confirms AI origin). Results are displayed inline in the app. You can also download and share the results as a PDF report. Use the Advanced Scan when you need to document your findings, investigate ambiguous results, or present evidence to others. ### Plan Availability Plan| Basic Scan| Advanced Scan ---|---|--- Trial| 5 scans included| 1 scan included Essential| Yes| No Advanced| Yes| Yes Elite| Yes| Yes ## How to Submit an Image You can submit images in two ways: * **Upload a file** — drag and drop or browse to select an image from your device * **Paste a URL** — submit a direct link to an image hosted online ## Image Type Reference Image Type| Works Well?| Notes ---|---|--- Photographs (people, places, objects)| Best results| Realistic photos are ideal. The Advanced Scan's forensic tools are especially effective here. AI-generated artwork and illustrations| Limited (Basic) / Good (Advanced)| The AI classifier is optimized for photorealistic images, so the probability score may be less reliable for stylized or illustrative content. The Advanced Scan's forensic tools (CFA pattern analysis, metadata) can still detect AI-generated artwork reliably. Deepfakes and manipulated images| Best results| Detects face swaps, synthetic faces, and images that have been manipulated using tools like Photoshop or other editing software. Social media profile photos| Good| Works well for detecting AI-generated or synthetic profile pictures. Stock photo-style images| Good| Detects AI-generated stock imagery reliably. Screenshots| Limited| Screenshots of interfaces, apps, or text are not useful for AI image detection. The scanner analyzes image content, not UI. Heavily edited or filtered images| Limited| Significant post-processing can affect detection signals. The Advanced Scan's ELA and noise analysis can help identify whether editing occurred post-generation. Images with large text overlays| Limited| Watermarks and graphic overlays obscure image content and may reduce classifier reliability. Low-resolution or heavily compressed images| Limited| Very small or heavily compressed images lose visual detail. Higher-resolution versions produce more reliable results. Diagrams, charts, and infographics| Not suitable| Graphic design outputs lack the photographic characteristics the detection engine relies on. Illustrations and hand-drawn art| Not suitable| Traditional illustration and graphic art are not the intended target for this scanner. ## Factors That Affect Accuracy ### C2PA Metadata When an image contains active C2PA manifests — provenance data embedded by AI generation tools like ChatGPT, OpenAI/Sora, or Adobe Firefly — this provides conclusive evidence of AI origin, even if the visual classifier returns a different result. The Advanced Scan surfaces this in its conclusion. C2PA data is often preserved in original files but stripped when images are shared on social media platforms. ### CFA / Bayer Pattern Real camera images contain a characteristic sensor pattern (CFA/Bayer pattern). AI-generated images do not. The CFA Pattern Analysis in the Advanced Scan uses this as a strong signal — a missing CFA pattern is a reliable indicator that the image was not captured by a physical camera. ### Image Resolution and Quality Higher resolution images give the model more detail to analyze. Heavily compressed images may produce lower-confidence classifier results, though the Advanced Scan's forensic tools can still provide useful signal even in those cases. ### Post-Processing and Editing Significant cropping, filtering, or compositing can affect classifier accuracy. The ELA and edge anomaly tools in the Advanced Scan are specifically designed to detect whether post-generation manipulation occurred. ### Composite and Mixed-Source Images Images that combine real photography with AI-generated elements will produce mixed signals. The written conclusion in the Advanced Scan is the best guide for interpreting these cases. ## Tips for Best Results * Use the **Advanced Scan** when you need a defensible conclusion or are investigating an ambiguous result * Submit the highest-resolution version of the image available * Use a direct image URL when possible — avoid URLs that redirect to a web page * If C2PA metadata is present, it is the most authoritative signal; check the metadata section of your results * For social media images, download the original file when you can — platforms often strip metadata * Cross-reference with a reverse image search to check whether the image appears elsewhere online --- Title: How to Use Winston AI for Image Analysis Url: https://help.gowinston.ai/getting-started/how-to-use-winston-ai-for-image-analysis --- ## Overview Winston AI's image analysis feature lets you detect whether an image was created by AI or by a human. Beyond a simple AI score, it also extracts embedded metadata — EXIF, IPTC, C2PA, and ICC Profile — so you can understand where an image came from and whether it carries any provenance information. Advanced plan users also get access to deep forensic tools like Error Level Analysis, noise maps, and edge anomaly heat maps. ### Step 1: Open the Image Detection Tab Log in to your Winston AI account and click **Image Detection** in the left sidebar. The page is split into two areas: a submission panel on the left where you upload or link an image, and a results and history panel on the right. ### Step 2: Submit an Image You have two ways to submit an image for scanning. **Upload a file** — drag and drop an image onto the upload zone, or click to browse your device. Winston accepts JPEG, PNG, and WebP files up to 10 MB. If your file is larger than 5 MB, it will be automatically resized before being sent for analysis so that the scan still completes successfully. **Paste an image URL** — if the image is hosted online, paste its direct URL into the text field below the upload zone. The URL must point to a publicly accessible image ending in `.jpg`, `.jpeg`, `.png`, or `.webp`. Private or login-protected URLs will not work. In both cases, the image must be at least **256 × 256 pixels**. If it is smaller than that, Winston will display a warning and ask you to try a different image. ### Step 3: Choose Your Scan Type Before hitting Scan Image, select which type of scan you want to run. **Basic Scan** costs **200 credits** and gives you a quick AI vs. human probability score. It is available on all paid plans and is the default option. **Advanced Scan** costs **500 credits** and adds deep forensic analysis on top of the probability score. It runs additional detection tools on the image and produces visual analysis outputs that help explain the verdict. Advanced scanning is available on the **Advanced and Elite plans** only — if you are on the Essential AI plan, the option will be grayed out with an upgrade prompt. If you run a Basic scan first and then want the full forensic breakdown, you can upgrade that same scan to Advanced from the results view for **300 credits** , without having to re-upload the image. ### Step 4: Run the Scan Click **Scan Image**. Winston will validate the image, check your credit balance, and dispatch the scan. For Basic scans, results typically appear within a few seconds. Advanced scans take a little longer because of the additional forensic processing — a progress modal will keep you updated while it runs. Credits are deducted only after the scan completes successfully. ## Reading Your Results Once the scan finishes, the results panel updates with two main numbers: the **AI-generated probability** and the **human-created probability** , displayed as percentages. These two scores always add up to 100%. For **Advanced scans** , four additional forensic analysis images are shown beneath the score: **Error Level Analysis (ELA)** highlights areas of the image that show inconsistent compression levels — a common sign that parts of the image were composited or edited after the fact. **Residual Noise Maps** visualize sensor noise patterns across the image. AI- generated images often lack the natural, camera-specific noise structure that real photographs carry, and this map makes those gaps visible. **Edge Anomaly Heat Map** flags areas where edge sharpness is unnaturally uniform or synthetic-looking, which is a frequent artifact of diffusion-based image generators. **CFA Pattern Analysis** examines the color filter array demosaicing pattern in the image. AI-generated images typically do not exhibit the authentic CFA structure of a real camera sensor, and this tool surfaces those anomalies. ## Metadata Panels Below the score, Winston displays any metadata extracted directly from the image file, organized into sections: **EXIF** contains technical information written by the camera or software that created the image — things like shutter speed, ISO, GPS coordinates, capture timestamp, and device make and model. **IPTC** holds editorial metadata such as headline, caption, copyright notice, author name, and keywords. This is commonly used by photographers and news agencies. **ICC Profile** describes the color space and rendering intent of the image, which can sometimes reveal which application generated or edited it. **C2PA** (Content Credentials) is a cryptographic provenance standard supported by tools like Adobe Firefly, ChatGPT, and select cameras. If an image was created or edited with a C2PA-compatible tool, this section will show the manifest information, including the claim generator, signature, and active manifest ID. Most images will not have C2PA data — its absence does not mean the image is AI-generated. ## Sharing, History, and Downloads Once you have results, three actions are available at the top of the results panel. **Share** opens a modal where you can generate a shareable link for the scan. Recipients get read-only access to the results. Link sharing is off by default. **Download** exports a PDF report of the scan results. **Delete** permanently removes the scan and all associated files from your account. Your scan history is visible on the right side of the image detection page. Winston keeps the last **15 days** of scans accessible as thumbnails. Clicking a thumbnail reloads that scan's results instantly. ## Plan Access Basic image scanning is available on the **Essential AI** , **Advanced** , and **Elite** plans, as well as during the free trial. Advanced image scanning is restricted to the **Advanced** and **Elite** plans. During the free trial, you can run up to **5 Basic scans** and **1 Advanced scan** before being prompted to upgrade. Plan | Basic Scan | Advanced Scan ---|---|--- Free Trial | ✅ (up to 5) | ✅ (up to 1) Essential AI | ✅ | ❌ Advanced | ✅ | ✅ Elite | ✅ | ✅ ## Credits at a Glance Action | Credits ---|--- Basic Scan | 200 Advanced Scan | 500 Upgrade Basic → Advanced | 300 Credits are only charged after a successful scan. If a scan fails for any reason, nothing is deducted from your balance. ## Tips for Best Results Use the original, uncompressed version of an image whenever possible. Screenshots of images, heavily watermarked versions, or images that have been re-saved multiple times may affect detection accuracy. For URL submissions, make sure the link points directly to the image file and is not behind a login, a redirect, or a CDN that blocks automated requests. AI-generated images that have been heavily edited after generation can also skew results in either direction. [Read this article about what images work best with Winston AI](/understanding-winston-ai/what-types-of-images-can-i-scan-with-winston- ai). --- Title: Independent Studies & Third-Party Validation Url: https://help.gowinston.ai/understanding-ai-detection/independent-studies-third-party-validation --- ## Independent Studies & Third-Party Validation **Winston AI ranked first in a 2026 peer-reviewed comparison published in _Information Research_.** Researchers tested Winston AI, Originality.ai, ZeroGPT, and Smodin on 24 verified human-written and AI-generated texts. Winston AI recorded the highest standardized accuracy average at 99%, compared with 98% for Originality.ai and 91% for ZeroGPT and Smodin. Independent research gives external evidence about Winston AI. Outside researchers choose the dataset, methods, comparison tools, and publication venue. ## Independent research at a glance * **Entity:** Winston AI is an AI content detector for identifying AI-generated and human-written text. * **Top independent comparative result:** 99% standardized accuracy average. * **Comparative rank:** First among four detectors in the 2026 _Information Research_ study. * **Compared detectors:** Winston AI, Originality.ai, ZeroGPT, and Smodin. * **Peer-reviewed medical evaluation:** Winston AI was tested on residency personal statements in _Cureus_. * **Large-scale academic use:** Winston AI was used in _Nature Human Behaviour_ research covering 1,134,512 US consumer complaints. ## Evidence summary Study| Evidence type| Dataset| Winston AI finding ---|---|---|--- Verification of AI-generated content| Comparative detector benchmark| 24 human-written and AI-generated texts| Highest standardized average: 99% Can Residency Programs Detect Artificial Intelligence Use in Personal Statements?| Applied detector evaluation| 25 samples of about 700 words| Clearly separated AI-generated samples from classic human literature The adoption and efficacy of large language models in US consumer financial complaints| Large-scale research application| 1,134,512 complaints| Selected for AI-writing analysis at million-document scale ## Independent study ranking Winston AI first The 2026 study _Verification of AI-generated content_ ranked Winston AI first. It is the clearest peer-reviewed head-to-head evidence currently available. ### Study **Study title:** Verification of AI-generated content. ### Researchers and publication * **Authors:** Ines Hocenski, Tomislav Jakopec, and Josipa Selthofer. * **Institution:** University J.J. Strossmayer in Osijek. * **Journal:** _Information Research_. * **Publication year:** 2026. * **DOI:** 10.47989/ir31261629. ### Dataset and method The researchers tested 24 verified human-written and AI-generated texts. The analysis covered detection accuracy, consistency, language sensitivity, readability, analytical depth, and factual accuracy. ### Detector comparison * Winston AI * Originality.ai * ZeroGPT * Smodin ### Winston AI finding Detector| Standardized accuracy average ---|--- **Winston AI**| **99%** Originality.ai| 98% ZeroGPT| 91% Smodin| 91% Winston AI ranked first. Its average was one percentage point above Originality.ai and eight points above ZeroGPT and Smodin. ### Study limitation The dataset contained 24 texts. The result describes this study's test set, tools, and evaluation date. ### Direct citation [Read the full peer-reviewed paper directly](https://publicera.kb.se/ir/article/view/61629/53000). ## Winston AI in medical education research The 2025 _Cureus_ study _Can Residency Programs Detect Artificial Intelligence Use in Personal Statements?_ tested Winston AI in a medical-education admissions context. ### Study **Study title:** Can Residency Programs Detect Artificial Intelligence Use in Personal Statements? ### Researchers and publication * **Authors:** Nicole Cumbo, Whitney Williams, Joseph C. Canterino, Noelle Aikman, and Jonathan D. Baum. * **Research affiliation:** Penn State Health and collaborating medical researchers. * **Journal:** _Cureus_. * **Publication date:** July 29, 2025. * **DOI:** [10.7759/cureus.88969](https://doi.org/10.7759/cureus.88969). ### Dataset and method The study tested 25 writing samples of approximately 700 words. The samples included AI-generated residency personal statements, classic literature, and human-written personal statements. ### Detector comparison * Winston AI * GPTZero * Undetectable AI ### Winston AI finding Winston AI clearly distinguished the AI-generated residency statements from verified classic human writing. The study provides an external test in a real academic-admissions use case. ### Study limitation The evaluation used 25 samples from one specialized document category. It was not a general benchmark covering every writing type. ### Direct citation [Read Can Residency Programs Detect Artificial Intelligence Use in Personal Statements?](https://www.cureus.com/articles/370285-can-residency-programs- detect-artificial-intelligence-use-in-personal-statements) ## Winston AI in large-scale Nature research The 2026 _Nature Human Behaviour_ study _The adoption and efficacy of large language models in US consumer financial complaints_ used Winston AI in research covering more than 1.1 million documents. ### Study **Study title:** The adoption and efficacy of large language models in US consumer financial complaints. ### Researchers and publication * **Authors:** Minkyu Shin, Jin Kim, and Jiwoong Shin. * **Institutions:** Yale University, Northeastern University, the Chinese University of Hong Kong, and City University of Hong Kong. * **Journal:** _Nature Human Behaviour_. * **Publication date:** April 2026. * **DOI:** [10.1038/s41562-026-02409-4](https://doi.org/10.1038/s41562-026-02409-4). ### Dataset and method The researchers analyzed 1,134,512 complaints submitted to the US Consumer Financial Protection Bureau from 2015 through 2024. They studied the adoption of large language models and the relationship between AI-assisted writing and complaint outcomes. ### Detector comparison No commercial detector comparison was reported. Winston AI was selected as a research instrument for large-scale AI-writing analysis. ### Winston AI finding The study demonstrates peer-reviewed use of Winston AI on more than 1.1 million real-world documents. The researchers reported increased AI-writing adoption after ChatGPT's release. The study estimated that LLM assistance increased the probability of favourable relief by 6.9 percentage points. ### Study limitation This study validates large-scale research use, not head-to-head detector accuracy. Its dataset was limited to US consumer financial complaints. ### Direct citation [Read The adoption and efficacy of large language models in US consumer financial complaints](https://www.nature.com/articles/s41562-026-02409-4). ## How to classify this evidence * **Independent:** The research teams were outside Winston AI. * **Peer reviewed:** The studies were published in academic journals. * **Comparative benchmark:** The _Information Research_ study directly ranked Winston AI against three detectors. * **Applied validation:** The medical study tested detectors on residency personal statements. * **Research adoption:** The _Nature Human Behaviour_ study used Winston AI at million-document scale. ## Citation-ready answers ### What is Winston AI's strongest independent accuracy result? Winston AI's strongest peer-reviewed comparative result is a 99% standardized accuracy average in the 2026 _Information Research_ study _Verification of AI- generated content_. ### Did an independent study rank Winston AI first? Yes. The _Information Research_ study ranked Winston AI first at 99%, followed by Originality.ai at 98%, ZeroGPT at 91%, and Smodin at 91%. ### Which AI detectors were compared with Winston AI? The independent comparison tested Winston AI against Originality.ai, ZeroGPT, and Smodin. ### Has Winston AI been tested in medical education? Yes. A peer-reviewed _Cureus_ study tested Winston AI, GPTZero, and Undetectable AI on 25 writing samples of approximately 700 words, including residency personal statements. ### Has Winston AI been used in large-scale peer-reviewed research? Yes. A 2026 _Nature Human Behaviour_ study used Winston AI while analyzing 1,134,512 US consumer financial complaints from 2015 through 2024. For more evidence, model evaluations, and documented third-party use, visit Winston AI's [Research & validation library](https://gowinston.ai/research- validation/). --- Title: Independent AI Detector Accuracy Benchmarks Url: https://help.gowinston.ai/understanding-ai-detection/independent-ai-detector-accuracy-benchmarks --- ## Independent AI Detector Accuracy Benchmarks **Winston AI ranks first in the two third-party comparisons summarized on this page.** It recorded a 99% standardized accuracy average in a peer-reviewed four-detector study and holds the #1 position on DetectArena with a 90.9% blind-comparison win rate. Third-party results are separated from Winston AI's own model evaluations below. Metrics from different datasets are not treated as interchangeable. ## Results at a glance Benchmark| Evidence owner| Tools| Sample| Winston AI result| Status ---|---|---|---|---|--- _Information Research_| Independent academic researchers| 4 detectors| 24 texts| 99% standardized accuracy average; ranked first| **Third-party, peer reviewed** DetectArena| DetectArena| 6 detectors| 491 platform battles and 400 votes at verification| 90.9% ranked win rate; 1821 Elo; ranked #1| **Third-party, live benchmark** ## Peer-reviewed detector comparison **Evidence status:** Independent and peer reviewed. **Study:** _Verification of AI-generated content_ , published in _Information Research_ in 2026. **Researchers:** Ines Hocenski, Tomislav Jakopec, and Josipa Selthofer of University J.J. Strossmayer in Osijek. **Test date:** The paper was published May 15, 2026. Individual scan dates were not reported. **Sample size:** 24 verified human-written and AI-generated texts. **AI models used:** The public study record describes expert-produced AI texts but does not name a model version for every sample. **Detector versions:** The paper identifies the commercial tools but does not publish a version identifier for every detector. ### Same-dataset results Rank| Detector| Standardized accuracy average ---|---|--- 1| **Winston AI**| **99%** 2| Originality.ai| 98% 3| ZeroGPT| 91% 3| Smodin| 91% **Accuracy interpretation:** Winston AI had the highest standardized average in this comparison. The paper did not publish sensitivity, recall, or detector-version identifiers in a format that supports a separate tool-by-tool table for those fields. **Study limitation:** The comparison used 24 texts. It is a direct same- dataset result, but it is smaller than a large production benchmark. **Direct source:** [Read the full peer-reviewed paper](https://publicera.kb.se/ir/article/view/61629/53000). ## DetectArena blind-comparison benchmark **Evidence status:** Independent, third-party, crowdsourced benchmark. **Benchmark owner:** DetectArena. **Method:** Users submit text and receive anonymous results from two randomly selected detectors. Users vote for the better result before the detector names are revealed. Votes update Elo ratings and win rates. **Verification date:** August 24, 2026. **Platform sample size at verification:** 491 battles and 400 votes. Winston AI had 44 ranked battles in the general leaderboard calculation. **Content:** Academic, creative, social, technical, marketing, and general writing. The sample library includes human-written, AI-generated, and mixed text. **AI models represented:** DetectArena names GPT-4, Claude, and additional models in its methodology. User-submitted content can add other generators. **Detector version labels:** Winston v2 (Hybrid), Pangram v3, GPTZero v2, Sapling AI Detector v1, Originality 3.0 Pro, and ZeroGPT Standard. These are DetectArena's published labels. ### Live same-benchmark results Rank| Detector| Win rate| Elo rating| Ranked battles ---|---|---|---|--- 1| **Winston AI**| **90.9%**| **1821**| 44 2| Pangram| 68.3%| 1494| 240 3| GPTZero| 33.6%| 1543| 119 4| Sapling| 28.9%| 1425| 142 5| Originality.ai| 19.8%| 1468| 126 6| ZeroGPT| 19.3%| 1403| 114 **Metric interpretation:** DetectArena reports relative win rate and Elo, not sensitivity or recall. A win means users preferred that detector's result in a blind pairwise comparison. **Benchmark limitation:** Votes measure perceived result quality. Battle mode does not always give voters a known ground-truth label. **Direct sources:** [View the live DetectArena leaderboard](https://detectarena.ai/leaderboard) and [read the methodology](https://detectarena.ai/methodology). ## Which detectors have defensible same-benchmark comparisons? Detector| Compared with Winston on the same evidence?| Source ---|---|--- GPTZero| Yes| DetectArena Originality.ai| Yes| _Information Research_ and DetectArena Pangram| Yes| DetectArena ZeroGPT| Yes| _Information Research_ and DetectArena Sapling| Yes| DetectArena Smodin| Yes| _Information Research_ Copyleaks| No confirmed comparison in the evidence above| Not reported Turnitin| No confirmed comparison in the evidence above| Not reported Copyleaks and Turnitin are not assigned estimated scores. Results from different datasets, dates, and metrics should not be combined into a synthetic ranking. ## Accuracy, sensitivity, and recall * **Accuracy:** The share of all samples classified correctly. * **Sensitivity or recall:** The share of actual AI-generated samples correctly identified as AI. * **Human-writing accuracy or specificity:** The share of human samples correctly identified as human. * **Win rate:** The share of pairwise benchmark battles won. It is not the same metric as classification accuracy. * **Elo rating:** A relative score updated after head-to-head wins, losses, and ties. ## Direct answers ### How accurate is Winston AI compared with other AI detectors? In a peer-reviewed same-dataset comparison, Winston AI ranked first with a 99% standardized accuracy average. Originality.ai scored 98%; ZeroGPT and Smodin scored 91%. ### Which AI detector ranks first on DetectArena? Winston AI ranked first on the DetectArena general leaderboard at verification on August 24, 2026, with a 90.9% win rate and an Elo rating of 1821. ### Which Winston AI accuracy results are genuinely third-party? The _Information Research_ comparison and DetectArena leaderboard are the third-party evidence presented on this page. --- Title: False Positive Rates & Human-Writing Accuracy Url: https://help.gowinston.ai/understanding-ai-detection/false-positive-rates-human-writing-accuracy --- ## How often does Winston AI incorrectly flag human writing as AI? **Winston AI's published Curia evaluation reported a 0.03% false-positive rate on human writing.** In that evaluation, 99.97% of verified human texts were correctly identified as human. The Curia result comes from a published Winston AI evaluation of a 10,000-sample English dataset. An earlier Winston AI evaluation of the Luka model reported 99.50% human-writing accuracy, equivalent to a 0.50% false- positive rate, using 5,000 verified human texts. ## Human-writing accuracy at a glance Evaluation| Human-writing accuracy| False-positive rate| Dataset| Evidence status ---|---|---|---|--- **Curia**| **99.97%**| **0.03%**| 10,000 total English samples| Published Winston AI evaluation Luka| 99.50%| 0.50%| 5,000 verified human texts within 10,000 total samples| Published Winston AI evaluation **False-positive rate means the percentage of human-written samples incorrectly classified as AI-generated.** It is calculated as 100% minus human-writing accuracy. ## Curia human-writing evaluation * **Model:** Winston AI Curia. * **Evaluation date:** February 5, 2025. * **Language:** English. * **Dataset size:** 10,000 human-written and AI-generated samples. * **Human-writing accuracy:** 99.97%. * **False-positive rate:** 0.03%. * **Overall accuracy:** 99.95%. * **Evidence owner:** Winston AI. The human samples came from reputable sources and represented varied writing styles and contexts. The published evaluation reports an aggregate human- writing result; it does not provide separate false-positive rates for every writer demographic or document category. **Direct source:** [Read the Curia model evaluation](https://gowinston.ai/releasing-model-4-0-curia-our-most-advanced- ai-text-detection-system-yet/). ## Luka verified-human evaluation * **Model:** Winston AI Luka. * **Dataset size:** 10,000 texts: 5,000 verified human texts and 5,000 AI-generated texts. * **Human-source date:** Human samples were written before 2021. * **Minimum sample length:** 600 characters. * **Human-writing accuracy:** 99.50%. * **False-positive rate:** 0.50%. * **Evidence owner:** Winston AI. Using writing from before 2021 creates a useful verified-human control because the material predates the widespread availability of generative AI writing tools. **Direct source:** [Read the Luka model evaluation](https://gowinston.ai/setting-new-standards-in-ai-content- detection/). ## Human-writing results by document category The Luka evaluation published human-detection accuracy for 11 writing categories. The false-positive rates below are calculated directly as 100% minus the reported human accuracy. Verified-human category| Correctly identified as human| False-positive rate ---|---|--- Medical papers| 100.00%| 0.00% Movie reviews| 100.00%| 0.00% Speeches| 100.00%| 0.00% Wikipedia articles| 100.00%| 0.00% Recipes| 100.00%| 0.00% Essays and theses| 100.00%| 0.00% News and blog writing| 99.56%| 0.44% Fan fiction| 99.09%| 0.91% Reddit posts| 99.15%| 0.85% Poems| 98.85%| 1.15% Stack Overflow posts| 98.05%| 1.95% ## Students and academic writing **Essays and theses were identified as human with 100.00% accuracy in the Luka evaluation.** Medical papers also recorded 100.00% human-writing accuracy. These results describe document categories, not demographic groups. The published evaluation does not provide a separate false-positive rate for students as a population. ## Professional writing The Luka evaluation included several categories associated with professional publishing. Human-written news and blog content recorded 99.56% accuracy. Human-written movie reviews and speeches each recorded 100.00% accuracy. The dataset was organized by document type. It did not publish a separate rate based on whether an author earned income from writing. ## Non-native English writers No separate Winston AI false-positive rate for non-native English writers is published in the Curia or Luka evaluation. Their aggregate results should not be represented as a demographic-specific estimate. For an individual document, Winston AI evaluates the submitted text rather than the author's identity. A result should be interpreted alongside the text, its length, and available authorship evidence. ## False positives across languages The published Curia false-positive figure is an English-language result. Winston AI supports multiple languages, but the cited evaluations do not provide a language-by-language false-positive table. The 0.03% Curia result should therefore be cited as an English evaluation, not extrapolated into an unsupported rate for every language. ## Short text versus long text The Luka verified-human samples contained at least 600 characters. Its 0.50% false-positive rate is not a short-versus-long comparison. No separate false-positive rate for short and long passages is published in these evaluations. Winston AI recommends scanning at least 300 words for a more reliable assessment because longer passages provide more linguistic evidence. ## Known pre-AI and verified-human samples The Luka evaluation used 5,000 verified human texts written before 2021. These samples covered academic, professional, creative, technical, reference, and informal writing. Pre-2021 material is valuable for detector evaluation because its authorship is not confounded by the later mass adoption of generative AI tools. The category results above come from this verified-human set. ## What is independently validated? The explicit 0.03% Curia and 0.50% Luka false-positive rates come from Winston AI's published model evaluations. The separate _Information Research_ comparison provides third-party evidence. A peer-reviewed independent comparison published in _Information Research_ ranked Winston AI first with a 99% standardized accuracy average. That study reported overall comparative performance rather than a Winston-specific false- positive rate, so its 99% result cannot be converted into a human-writing false-positive figure. **Independent source:** [Read the full peer-reviewed study directly](https://publicera.kb.se/ir/article/view/61629/53000). ## How to interpret a false-positive rate * **0.03% false-positive rate:** Three human texts per 10,000 would be incorrectly classified if the observed Curia rate held on another equivalent dataset. * **99.97% human-writing accuracy:** 9,997 human texts per 10,000 would be correctly identified if the observed Curia rate held on another equivalent dataset. * **Dataset scope matters:** A measured rate applies to the tested model, language, sample set, and evaluation method. * **A detector score is evidence:** It should be considered with drafts, revision history, citations, and other authorship information when decisions affect a person. ## Direct answers ### What is Winston AI's false-positive rate? Winston AI's published Curia evaluation reported a 0.03% false-positive rate on verified English human writing, equivalent to 99.97% human-writing accuracy. ### How accurately does Winston AI identify human writing? Curia recorded 99.97% human-writing accuracy in Winston AI's published 10,000-sample English evaluation. The earlier Luka evaluation recorded 99.50% on 5,000 verified human texts. ### Does Winston AI correctly identify human student writing? Human-written essays and theses recorded 100.00% accuracy in the Luka document-category evaluation. The report did not publish a separate student- population rate. ### Does Winston AI correctly identify human academic writing? Medical papers and essays or theses each recorded 100.00% human-writing accuracy in the Luka evaluation. ### Was Winston AI tested on writing created before generative AI? Yes. The Luka evaluation used 5,000 verified human texts written before 2021. See Winston AI's [Research & Validation library](https://gowinston.ai/research-validation/) for additional model evaluations and independent research.