Winston AI logo
Content Authenticity Detection · Academic Research

Winston AI

Winston AI is a content integrity platform that detects AI-generated text and images, checks for plagiarism and verifies facts. It scores submissions sentence by sentence and produces shareable authenticity reports for educators, publishers and professional writers.

Active GDPR compliant Free trial Subscription API available 18+ Verified by Guidaio
Overview

What is Winston AI?

Winston AI is a content integrity platform published by Winston AI inc., a Montreal company, built around a single question: was this written by a human, or by a machine? Its core product is an AI content detector that returns a score from 0 to 100 on the likelihood that a text is machine-generated, alongside an AI Prediction Map that colour-codes the assessment sentence by sentence. The published figures come from two model generations. Version 3.0, Luka, reports 99.98% AI-detection accuracy, 99.50% human-detection accuracy and 99.74% overall classification on a 10,000-text dataset split evenly between human and AI writing. Version 4.0, Curia, reports 99.95% overall classification on a 10,000-sample English dataset and an R2 of 0.9908 when estimating what proportion of a document is AI-written. Detection algorithms are updated weekly, and the model is trained on a human-reviewed dataset specifically to keep false positives low and to catch paraphrased or heavily edited passages.

Around that detector sits a wider suite. An AI image detector flags deepfakes and pictures from Midjourney, Google Nano Banana, ChatGPT Image, Stable Diffusion, Adobe Firefly and Meta AI, returning EXIF, IPTC and C2PA forensics alongside the verdict. An advanced plagiarism checker compares text against online content and publications, sorting matches into uncited, cited and inaccessible buckets. A fact checker, an essay grader, a grammar checker, writing feedback, a citation generator, a readability score, a text comparison tool and a word counter round out the workspace, while an OCR layer reads scanned documents and handwriting. HUMN-1 and HUMN-1+ site certifications let a publisher signal that its pages are human-written.

Three audiences are addressed directly: education, where the emphasis is on academic integrity and defensible evidence; SEO and publishing, where the concern is authenticity before release; and individual writers. Independent validation is claimed from three peer-reviewed studies, at Osijek, Yale and Penn State, and from DetectArena, a third-party blind benchmark where Winston AI ranked first with an 1821 Elo rating and a 90.9% win rate as of 20 August 2026. Access is through the web application, a Chrome extension, a WordPress plugin, Zapier or a version 2 REST API.

What it does

  • Detect AI-generated text from ChatGPT, Claude, Google Gemini, LLaMA and other known large language models
  • Detect AI-generated images and deepfakes, with EXIF, IPTC and C2PA forensics
  • Check text for plagiarism against online content and published sources
  • Map AI probability sentence by sentence with a colour-coded AI Prediction Map
  • Generate authenticity reports that can be shared by link or exported as printable PDFs
  • Verify factual claims, grade essays, score readability and return writing feedback
  • Plug detection into any application through the REST API, or into Chrome, WordPress and Zapier
Audience

When to use Winston AI / When not to

A quick filter to help you decide if Winston AI is the right fit.

When to use Winston AI

  • Teachers, professors and academic institutions reviewing student submissions for AI use and plagiarism
  • Publishers, editors and SEO teams verifying content authenticity before anything goes live
  • Journalists and fact-checkers assessing whether a text or an image has been AI-generated or manipulated
  • Freelance writers, authors and content agencies who need to evidence the human origin of their work
  • Developers and platforms embedding detection into their own product through the REST API

When not to use Winston AI

  • Anyone looking to write, rewrite or "humanise" content: Winston AI analyses text, it never generates it
  • Users who need a mobile app, since access is web-based through a browser, a Chrome extension, a WordPress plugin, Zapier or the API
  • Organisations that require a self-hosted or on-premise deployment, which is not offered
  • European data controllers who need a signed Data Processing Agreement and a published subprocessor list before onboarding a vendor
  • Anyone expecting a permanently free tool: the free tier is a 14-day, 2,000-credit trial that is suspended at the end
Get started

How to use Winston AI

A typical end-to-end flow, from setup to results.

  1. Create a free account with an email address, a Google or Facebook login, or single sign-on; no credit card is requested
  2. Paste the text to analyse, upload a document in .docx, .png or .jpg format, or import it from a URL
  3. For images, drag and drop a JPEG, JPG, PNG or WEBP file of up to 10 MB
  4. Choose the depth of an image scan: Basic for a quick verdict, Advanced for full forensic analysis
  5. Run the scan and read the 0 to 100 score, returned within seconds
  6. Open the AI Prediction Map to see which sentences carry the strongest machine-writing signals
  7. Switch on the plagiarism checker to cross-reference the same text against online sources
  8. Label and file documents in the dashboard by title, author and type so they can be retrieved later
  9. Share the result as a link, or export it as a printable PDF report
  10. For an integration, open a developer account on the API environment, generate keys and call the version 2 endpoints
Quick read

Pros & Cons

Pros

  • One of the strongest accuracy claims on the market, backed by a published evaluation dataset and methodology
  • Genuine third-party validation: three peer-reviewed studies and first place on the DetectArena blind benchmark
  • A single workspace for AI detection, images, plagiarism, fact-checking and grading rather than five separate tools
  • The detection model is retrained weekly to keep pace with new large language models
  • A solid data posture: submitted content is never used for training, hosting is in Canada, and users keep full ownership
  • GDPR compliance claimed throughout and a SOC 2 Type 2 certification announced for enterprise buyers
  • A documented API with 2,500 free credits, and a low paid entry point at 10.00 USD per month billed annually

Cons

  • No permanent free plan: the "Free" tier is a 14-day, 2,000-credit trial after which the subscription is suspended
  • Most headline accuracy figures come from evaluations the vendor published itself; only the Osijek study and DetectArena are third-party
  • No Data Processing Agreement is published, no subprocessor list exists and no Article 27 EU representative is designated
  • No mobile application on iOS or Android, and no self-hosted or on-premise option
  • The advertised language coverage differs from one page of the site to another
  • The legal and security email addresses printed on the terms and disclosure pages are broken hosting placeholders
  • Credit-based pricing means heavy volumes require top-ups, an API plan or an enterprise contract
Pricing

Pricing & Plans

There is no permanent free plan. New users are offered a 14-day free trial carrying 2,000 credits and requiring no credit card; at the end of that period the subscription is suspended unless it is upgraded to a paid version. The lowest paid entry point is the Essential plan at 10.00 USD per month when billed annually, that is 120.00 USD per year, or 18 USD per month on a month-to-month basis, the annual commitment being advertised as a 45% saving. Payments are processed by Stripe.

Free, 0 USD
  • 2
  • 000 credits over a 14-day trial
  • AI content detection
  • AI image and deepfake detection
  • document scanning
  • email and chat support
  • and HUMN-1 site certification for the trial period
Advanced, 16.00 USD per month billed annually (192.00 USD per year) or 29 USD per month month-to-month
  • 200
  • 000 credits per month
  • everything in Essential plus advanced plagiarism detection
  • HUMN-1+ site certification and up to 5 team members
Elite, 26.00 USD per month billed annually (312.00 USD per year) or 49 USD per month month-to-month
  • 500
  • 000 credits per month
  • everything in Advanced with unlimited team members
Enterprise and custom plans, price on request through the sales form
  • tailored detection capability
  • custom integrations with existing tools
  • dedicated support and implementation assistance
Plan 6
  • API access through a separate developer account
  • with 2
  • 500 credits granted on sign-up
Plan 7
  • Top-up credits are available on every plan
  • and all paid plans allow up to 200
  • 000 characters per scan with unlimited scans per hour
Special offers — A 14-day free trial carrying 2,000 credits and requiring no credit card · A 45% saving advertised on annual billing compared with month-to-month subscriptions · 2,500 free credits granted when opening a developer account for the API · School and academic access, listed as a dedicated enquiry option on the sales form, with pricing on request and no published discount rate · Reseller and enterprise arrangements, also handled through the sales form
Prices and plans listed above may evolve. Always check the official pricing page before subscribing.
Trust & Privacy

Data, GDPR & hosting

A consolidated view of how Winston AI handles your data.

GDPR overview

GDPR compliance is claimed explicitly and repeatedly, on the homepage, the product page and the sales page, which also announces a SOC 2 Type 2 certification. The privacy notice, effective 17 July 2026, sets out the GDPR and UK GDPR legal bases relied on (consent, contract, legitimate interests, legal obligation, vital interests) and lists EEA and UK rights of access, rectification, erasure, restriction, portability and objection, with links to the European supervisory authorities and the Swiss regulator. A Privacy Officer is designated under Quebec's Law 25, reachable at privacy@gowinston.ai, with a 30-day response commitment, and privacy impact assessments are run on high-risk projects. Three gaps remain in the pages collected: no Article 27 EU representative is designated, no Data Processing Agreement is published or offered, and no subprocessor list exists.

Who owns the data?

The terms are unusually clear on this point: Winston AI asserts no ownership over submitted content, and users retain full ownership of everything they upload along with the associated intellectual property rights. Submissions and their detection reports are private by default and reachable only through the account; a report becomes visible to others only if the user chooses to generate a shareable link. Where a submission carries a third party's personal data, the privacy notice makes the submitter the data controller and Winston AI the processor. Two clauses run the other way: product feedback becomes the company's exclusive property, and enterprise customers' names and logos may be used in marketing unless they opt out in writing.

Reuse rights

Users may freely reuse what they submit and what comes back. They keep all rights to their own content, and detection reports can be exported as printable PDFs or circulated through a shareable link without asking anyone's permission. The restrictions concern Winston AI's own material and service rather than the user's. The terms forbid systematically retrieving site content to build a collection, compilation, database or directory without written permission, and forbid using the site to compete with the company or for revenue-generating endeavours outside the limited personal, non-commercial licence granted on the site's own content. API access carries its own rule: it may not be resold, sublicensed or redistributed without express prior written permission, and rate limits and fair-use policies apply and may be revised. Submitters must also warrant that they hold the rights to whatever they upload.

Data retention & training

Retention summary
Account data, meaning name, email address and settings, is kept for as long as the account exists plus up to six months from the start of any idle period, or until deletion is requested. Content submitted for analysis is kept for as long as its report exists in the dashboard. Deleting a report removes the associated content from active storage within 30 days, and deleting the account removes all content and reports within the same 30 days. Copies may survive in encrypted backups for up to 90 days afterwards, then are permanently purged. Log and usage data is kept for up to 12 months, then deleted or anonymised. Nothing is retained beyond these periods except where the law requires it. Where deletion is temporarily impossible, the data is isolated from further processing until it becomes possible.
Trains on customer data
No
GDPR contact

Hosting summary

The terms state plainly that the site is hosted in Canada, and that accessing it from anywhere else amounts to transferring your data to Canada for processing there. The publisher, Winston AI inc., is a Quebec company; the terms are governed by the law of the Province of Quebec and the federal laws of Canada, with Quebec courts holding jurisdiction. The privacy notice describes submitted text and images as held in encrypted, access-controlled cloud storage, encrypted at rest and in transit, with access restricted to authorised personnel on a strict need-to-know basis rather than open to employees generally. Periodic security reviews of the infrastructure are claimed, and third-party providers handling the data are said to be selected on their security practices and contractually bound to appropriate safeguards. Payment data is stored by Stripe rather than on the vendor's own servers. What the pages do not say matters too: no hosting provider is named, no cloud region is specified and no subprocessor list is published. A SOC 2 Type 2 certification is announced on the sales page.

Hosting countries
🇨🇦 Canada
Watch-outs

Things to keep in mind

Risks and trade-offs to weigh before adopting Winston AI.

  • False positives: a human-written text flagged as AI can cost a student a grade or a writer a contract. The vendor claims minimisation, not elimination, and the site itself concedes that performance varies with the source model, the level of editing and the document's length
  • Treating a score as a verdict: the education page recommends combining detection with the student's writing history, a review of sources and a conversation, never a sanction based on the number alone
  • Shared reports: a report circulated by link becomes visible to anyone holding that link, original submitted content included
  • Third-party data: uploading someone else's essay or photograph makes the submitter the data controller for personal data that person never consented to have processed
  • Erosion of judgement: outsourcing the question "does this read as human?" to a score can dull the editorial and pedagogical instinct it was meant to support
  • Terms that cut one way: product feedback becomes the company's exclusive property with moral rights waived, and enterprise customers' names and logos may appear in marketing unless they opt out in writing
  • Commercial lock-in: purchases are non-refundable in principle, annual plans cannot be cancelled mid-term, and heavy usage pushes towards credit top-ups
Setup

Setup & Integrations

Technical difficulty

Very low for everyday use. Creating an account takes seconds, by email, Google or Facebook login or single sign-on, and no credit card is asked for. Pasting a text or uploading a file is all that is needed to obtain a result. The Chrome extension, the WordPress plugin and the Zapier connector install without writing any code. Only the API raises the bar: it requires a separate developer account, key generation and REST calls to the version 2 endpoints, which is developer work, straightforward but developer work nonetheless.

Deployment

Web appAPIChrome extensionPlugin

Integrations

Zapier WordPress Google Classroom Google Chrome Google Docs ChatGPT

Supported languages

EnglishFrenchSpanishGermanPortugueseDutchPolishItalianRomanianIndonesianTagalogRussianBulgarianUkrainianChinese (Simplified)
Company

Behind Winston AI

Company name
Winston AI inc
Founded
28/01/2023
Country of origin
🇨🇦 Canada
Headquarters
215 Avenue du Mont-Royal O, Montréal, QC H2T 2T2, Canada
UBO
Thierry Lavergne
UBO country
🇨🇦 Canada
Domain registrar country
🇺🇸 United States
Legal contact
Support contact

Social

Official links

Resources

All the official URLs gathered for verification and reference.

Compare

Alternatives

Tools that compete with or complement Winston AI.

G GPTZeroO Originality.aiC CopyleaksQ QuillbotS ScribbrZ ZeroGPTS SaplingP PangramS Smodin
FAQ

Frequently asked questions

What exactly does Winston AI do?
It analyses text and images to estimate whether they were produced by a human or by a generative model, checks the same text for plagiarism, verifies factual claims, grades essays and returns a shareable report. It detects content; it never writes or rewrites it.
How accurate is the detection?
The vendor reports 99.98% AI-detection accuracy for its Luka 3.0 model and 99.95% overall classification for Curia 4.0, both on 10,000-sample datasets it published itself. Independently, a peer-reviewed study at the University of J.J. Strossmayer in Osijek recorded 99%, and the DetectArena third-party benchmark ranked Winston AI first with an 1821 Elo rating and a 90.9% win rate as of 20 August 2026.
Which AI models can it detect?
For text, ChatGPT, Claude, Google Gemini, LLaMA and the other known large language models. For images, Midjourney, Google Nano Banana, ChatGPT Image, Stable Diffusion, Adobe Firefly and Meta AI, among others.
Which languages are supported?
Around fifteen, including English, French, Spanish, German, Portuguese, Dutch, Polish, Italian, Romanian, Indonesian, Tagalog, Russian, Bulgarian, Ukrainian and Simplified Chinese. Be aware that the lists published on different pages of the site do not entirely agree with each other.
Is there a free plan?
Not a permanent one. New accounts get a 14-day trial with 2,000 credits and no credit card requirement, after which the subscription is suspended unless it is upgraded. The cheapest paid plan, Essential, costs 10.00 USD per month billed annually, or 18 USD per month month-to-month.
Is my content used to train the models?
The site states that it is not, on the homepage, on the education page and in the API FAQ. The privacy notice adds that training on submitted content would require explicit, opt-in consent from the submitter, so it is switched off by default rather than something to opt out of.
Where is the data hosted and how long is it kept?
The terms state that the site is hosted in Canada, under the law of the Province of Quebec. Submitted content is deleted from active storage within 30 days of a report or an account being deleted, may persist in encrypted backups for up to 90 days after that, and log data is kept for up to 12 months.
Who owns what I upload?
You do. The terms state that Winston AI asserts no ownership over submitted content and that you retain full ownership and all associated intellectual property rights. Product feedback and suggestions are the exception: those become the company's exclusive property.
Is there an API, and is there a mobile app?
There is a version 2 REST API covering text detection, plagiarism, fact-checking and image detection, publicly documented, with 2,500 credits granted on developer sign-up. There is no iOS or Android application: access is through the web app, a Chrome extension, a WordPress plugin or Zapier.
Can a detection score on its own prove that a student cheated?
The site itself says no. Its education page recommends combining the result with the student's writing history, a review of sources and a direct conversation, and notes that performance varies with the source model, the level of editing and the length of the document. Users must also be at least 18 years old.
Conclusion

Should you pick Winston AI?

Winston AI is one of the few AI detectors that argues its case with evidence rather than adjectives. Three peer-reviewed studies, a published evaluation dataset, first place on a blind third-party benchmark and documented use by the Daily Mail, Snopes, WIRED and Zapier put it on firmer ground than most of its category, and weekly model retraining is the right answer to a problem that moves every month. The suite around the detector, covering image forensics, plagiarism, fact-checking, essay grading and OCR, makes it a workspace rather than a widget, which matters most in education, where a defensible report is worth more than a number.

The data posture is genuinely good. Submitted content is never used for training, hosting is in Canada under Quebec law, a Privacy Officer is designated, users keep full ownership of what they upload, and SOC 2 Type 2 is claimed. The gaps are administrative rather than technical, but they are real for European buyers: no published Data Processing Agreement, no subprocessor list, no Article 27 representative. The broken hosting placeholders standing in for the legal and security email addresses on two policy pages do not help either.

Two reservations deserve stating plainly. The headline 99.98% figure is the vendor's own measurement, and the site is candid enough to note elsewhere that performance varies with the source model, the amount of editing and the document's length. And the Free plan is a 14-day trial, not a free tier. Read the accuracy claim as a strong indication rather than a verdict, keep a human in the loop as the education page itself advises, and Winston AI is a credible, reasonably priced choice from 10.00 USD per month.