Winston AI Review: Accuracy, Pricing and Terms (2026)

Winston AI's homepage calls it "the only AI detector with a 99,87% accuracy rate". Its terms and conditions say the tool "may mistakenly label content as AI-generated approximately 1 time out of every 10." Those two sentences sit on the same website, and they cannot both describe the same experience of using it.

I read both documents on 3 November 2026, along with the pricing pages and the independent tests I could find. I did not run my own scans for this review. What follows is what Winston says about itself, what outsiders measured, and where the two disagree.

My view: Winston is a serious detector with a clear free trial and fair monthly pricing. The headline accuracy number is a marketing figure and the terms of service are a more honest description of what a score means.

What Winston AI says about itself

The homepage features a "99.87% Accurate" badge and quotes "99.89% of AI text detected at a 1% false-positive rate, including adversarial attacks." The company cites a Chicago Booth benchmark of 8,697 documents with zero false positives and zero missed AI documents and states that Winston is ranked first on RAID.

Two small things stand out. The headline figure is represented both as 99.87% and 99,87%, which is different from the 99.89% RAID figure. Also, the page does not include a link to an independent check of the headline number, showing that the number was provided by Winston alone.

The page also lists languages that the software supports, including English, French, Spanish, German, Portuguese and more. The software also claims to be able to detect the output of models like ChatGPT, Claude, Google Gemini and Llama. The FAQ lists fewer languages, so be sure to check the current list if you do not read English.

What the terms of service say

Section 23 of the terms, last updated 18 March 2026, is the most useful page on the site. The section states, "The accuracy of the returned result is not guaranteed," that the score is "a probabilistic estimate," and that the tool may mislabel content as AI-generated roughly one time in ten.

One in ten is not a 1% false-positive rate. These can both be true if they describe different texts and settings, but a reader deciding whether to accuse a student or a freelancer needs to know which one applies to the text in front of them. Winston does not say.

Independent tests

Winston's help center lists two outside benchmarks. A peer-reviewed study in Information Research, by researchers at the University of Osijek and published on 15 May 2026, tested 24 texts and reported a standardized accuracy of 99% for Winston, 98% for Originality.ai and 91% each for ZeroGPT and Smodin. The page itself notes that 24 texts is small and that scan dates and detector versions were not reported. The second is the DetectArena leaderboard, where Winston had a 90.9% win rate across 44 ranked battles at the 24 August 2026 check. DetectArena measures which result voters prefer, not accuracy against known labels, and 44 battles is a thin sample.

Both of those are chosen and presented by Winston. A third-party review on fast.io, last reviewed 18 June 2026, collects the less flattering numbers, most of them taken from competing vendors' write-ups. They are secondhand, so I would read them as warnings rather than measurements.

The spread is the story. Depending on who ran the test and what they fed it, Winston looks like a near-perfect detector or like one that misses a quarter of AI text and flags one human text in ten. I would not trust either end without testing it on your own material.

Pricing and credits

Pricing comes from Winston's help center, checked on 3 November 2026.

Unused monthly credits do not roll over, and monthly plans can be cancelled at any time. The terms say cancellation takes effect at the end of the current paid term and that "All purchases are non-refundable." The help center's pricing pages say nothing about refunds, so the terms are the only statement. If you buy an annual plan, you are paying for the year up front with no refund route in writing.

The free trial is the sensible way in. Scan three or four texts you know the origin of, including something you wrote yourself, before paying.

Who it is for

Winston makes most sense for publishers and editors who want to scan a lot of copy and accept that a score is a prompt for a closer read. It is a weaker fit for high-stakes decisions about one person's work, because Winston's own terms say the result is probabilistic and can be wrong.

The same applies to you if you are the one being scanned. A flag from a single detector does not prove anything, as explained in our article on false positives in AI detectors. For how detectors score text in the first place, see how AI detectors work.

The short version

Winston AI offers a real free trial, plain pricing and a long list of supported models. Its 99.87% claim comes from Winston, and its own terms put the mislabel rate at about 1 in 10. Test it on your own samples during the 14-day trial, never act on one score alone, and remember that annual plans are non-refundable. If you work with text that gets flagged, the Winston AI bypass page covers how our humanizer handles it, and Are AI detectors accurate? puts the published numbers for other tools in one table.

If you want the longer treatment:

• Originality.ai review
• Is GPTZero accurate?
• AI detector for teachers
• What AI detector do colleges use?