One company, both sides of the test

Undetectable AI sells two tools that are built to defeat each other. One is an AI detector that the vendor's site calls "The Most Accurate AI Checker: 99%+ Accuracy Proven By Independent Tests". The other is a humanizer that rewrites ChatGPT or Claude output until detectors stop flagging it, backed by a promise on the pricing page: "Money-back guarantee if output flagged is not human."

Both come on one subscription. Every paid tier lists "Unlimited AI Detecting" next to the humanizer words, so you rewrite a paragraph and then grade it with the same company's checker. That loop is the product. Whether the grade means anything depends on which detectors sit inside it.

The seller is Undetectable Inc. according to the Terms and Undetectable LLC according to the privacy policy, at a registered-agent address in Sheridan, Wyoming. Christian Perry and Devan Leos are named as co-founders, and the product was in pre-launch in early 2023. The user count moves from page to page: "22M+ individuals" on the about page, "+8M users" in the CEO bio on that same page, "24M+ users" on the detector page.

What the humanizer actually does

Paste text, pick a model and a purpose, press Humanize. The web app offers three models - Basic, Stealth and Undetectable - and purposes running from General Writing and Essay to Article, Marketing Material and Story, with readability targets such as high school, university and journalist. The API shows more of the machinery: a strength setting of Quality, Balanced or More Human, and engines where "v2" covers every language with medium humanization while "v11sr" is slower and billed as "best for English, best humanization".

The free offer is smaller than the page title. The humanizer page is headed "Free AI Humanizer (Unlimited, No Signup)", yet the same page says "For 99% human scores confidence, sign up for the free trial and use the Undetectable model." That trial is 250 words over three days, and it takes a card up front. The help centre is plain about what happens next: you are charged "after your 3-day free trial ends or once you exceed the included trial credits". One 300-word essay can end the trial and start the subscription on the same afternoon.

Does the output pass detection?

Against its own checker, often. The detector bundled with the subscription reports scores from eight third-party tools: GPTZero, OpenAI, Writer, CrossPlag, Copyleaks, Sapling, Content at Scale and ZeroGPT. Two names are missing. Turnitin, which most universities run, is not there, and neither is Originality.ai, which many publishers and SEO agencies use. "Passes AI detectors" on the pricing page is measured against a list that leaves out the two tests most buyers actually face.

Both of those have moved against humanizers. In August 2025 Turnitin added detection aimed at "bypasser" tools, and its chief product officer Annie Chechitelli said "humanizers also leave a statistical trace that can be learned and detected. We have updated our model to identify leading humanizers." On August 10, 2026, Originality.ai published its own test of Undetectable: it flagged both the original and the humanized text with "100% Confidence that the text is Likely AI", GPTZero scored the humanized version "91% AI-Generated", and only Writer let it through, at 3%. Originality.ai sells a competing detector, so weigh that result with its motive in mind. GPTZero, though, is on Undetectable's own list of eight, and it still caught the text.

We did not push a batch through Turnitin, and a single run would prove little if we had: detectors retrain without notice, which is exactly why a "passes" claim decays. The only test that counts is your own text through the checker your reader uses. Our reviews of Originality.ai, Copyleaks and ZeroGPT cover what each one catches.

Pricing and the billing terms

The pricing page names no plans. It shows word allowances on a slider, and for each one a monthly-billing price and a lower per-month price for paying a year up front - the help centre's "save 50% on yearly plans". Checked on the pricing page on 2026-10-26:

The slider runs to 380,000 words, while a help-centre article caps purchases at "up to 300,000 words per month". Prices are in US dollars and VAT may be added. The custom Business tier adds "Non-expiring credits", which implies ordinary plan words expire each month; the page never says so outright.

The real costs sit in the account terms. Unless you live in California or New York, cancelling means an email to [email protected] - the Terms let residents of those two states "cancel directly in their account dashboard" while "All other users may cancel by emailing us." Cancel and "you will immediately lose your remaining words", including words prepaid on a yearly plan. Upgrading charges "the full price of the new plan" at once, though unused words convert into non-expiring ones. Section 6.2 adds that the company "does not guarantee that your subscription will last for a full year".

Reviewers have noticed. Read on 2026-10-26, the Trustpilot page showed a 3.5/5 TrustScore from 1,002 reviews: 51% five-star, 40% one-star, and the one-star pile is mostly about billing rather than output. "I didn't think it would charge me immediately...there was little to zero indicator," one buyer wrote on October 16, 2026. Another, on October 2: "Instead of simply being able to cancel...I was instructed to send an email." The company replies to about 18% of negative reviews.

Check Undetectable AI pricing

The money-back guarantee, read in full

The pricing page says: "If anything we produce is flagged as not human, we will refund the cost of humanization." The Terms shrink that in three moves. Section 6.1 sets the bar: "A refund of credits is only available if a submission provably flags as less than 75% human on a reputable AI detector." The same section sets the clock: "Refunds of credit or payment requests must be made within 30 days of the incident." Section 27 closes the door: "Refunds will not be provided unless required by law."

So the money back is, in practice, words back - credited to an account whose words vanish the day you cancel. A 74% human score qualifies. A 76% score that still gets an essay escalated does not. "Reputable" is never defined, and the proof is on you. Cancelling "does not refund prior billing periods unless required by law" either, so a trial charge you missed falls outside the guarantee entirely.

The API, and a page that disagrees with itself

Developers get humanize and detect endpoints, a default limit of 3 requests per second per IP, and the 250 trial words usable through the API. Detection costs a tenth of what humanizing does. What a word costs is less clear. The developer page lists Starter at $29 a month ($0.40 per 1K words, 72,500 words included) and Growth at $49 ($0.20 per 1K words, 245,000 included), but the FAQ on the same page and the help centre both say API usage draws on your account and "the cost is identical to the web platform". Get the answer from support in writing before you budget a pipeline around either figure.

Your text, your output, their rules

Output rights are clean. The Terms say "You retain full ownership of all of your Contributions" and the company claims none, with one restriction: no using output "for the purpose of training, fine-tuning, or otherwise developing any machine learning models". What happens to the text you paste in is another matter. The privacy policy, last updated 03/18/2023, is silent on whether submissions are stored or used to train the humanizer; it only says personal information is kept "for as long as it is necessary". Account deletion is self-service at undetectable.ai/delete. Keep client contracts and unpublished manuscripts out of a tool whose data policy predates the product you are using.

Verdict: who should pay, and who should not

Buy it if you write marketing or blog copy, your AI drafts read like templates, and you want a quick pass that varies sentence shape with a Journalist or Marketing readability target. The yearly 10,000-word tier at $60 is cheap for that job - provided you are fine with email-only cancellation and words that disappear when you leave.

Skip it if you need a guaranteed pass. Its checker leaves out Turnitin and Originality.ai, both now tuned against humanizers, and its guarantee pays in credits at best. Students most of all: the vendor's ethics page lists "Cheating in school" as unacceptable use and says "Our moderation team investigates .edu email signups", while the humanizer page invites you to "Use AI safely and confidently in work and academic settings" and offers an Essay mode. To polish writing that is already yours, a plain editor such as QuillBot or Wordtune does the job without the pretence. If you want a humanizer regardless, read our AISEO review before giving a card to a three-day trial.

Try Undetectable AI