AI humanizer refund policies: what each vendor's terms state

The table quotes the term as the vendor states it. "Where stated" names the page, and every row was read on 9 November 2026 unless another date is given.

Linked sources for the four rows this page leans on hardest: StealthGPT terms of service, WriteHuman pricing page, HIX Bypass refund policy and Undetectable AI pricing page, all read 9 November 2026. The other rows are quoted from the vendor pages named in the table, read on the same date, and each is cited in full on that tool's review in the reviews hub. The pricing and product pages read for Grammarly, Walter Writes AI and Verva (formerly TwainGPT) carry no refund term, so those three do not appear in the table.

Five vendors frame a refund as a bet on a detector. The terms differ in who judges, what counts as proof, and what is excluded, and the differences decide whether the promise is worth anything to a given buyer.

StealthGPT's terms read 9 November 2026; the remaining rows from each vendor's pricing page, read the same day. Two features recur. The vendor is the judge of its own guarantee in every case, and the detector list is the vendor's choice: StealthGPT names Turnitin as acceptable proof while excluding Originality.ai, and Phrasly names Originality.ai while leaving out Turnitin and ZeroGPT. A buyer who tests on a detector the vendor has excluded has no claim.

A detector-tied refund is a commercial term, not a measurement, and this site scores neither. Detector models change without notice, and a reading is true for its date. No detector score should be the only evidence in a misconduct case, and no refund guarantee is evidence that a tool works.

A refund returns money already taken; a cancellation stops the next charge. Several vendors' terms make the second condition a prerequisite for the first, or make the timing of the first depend on the second.

All routes read 9 November 2026; Undetectable AI's route is set out step by step in how to cancel Undetectable AI. The pattern across the table is that yearly plans are paid up front, cancellation stops the following year, and no vendor in the table states a prorated refund of a yearly charge on a change of mind. StealthGPT's prorated refund applies only inside its detector policy.

The FTC announced a final "click-to-cancel" rule on 16 October 2024 that would prohibit sellers from misrepresenting material facts, require clear disclosure of material terms before taking billing information, require express informed consent to recurring charges, and require a simple mechanism to cancel and immediately halt charges (FTC press release, 16 October 2024, read 9 November 2026). The rule concerned cancellation and disclosure; it did not set a refund window, and no US federal rule read for this page requires a humanizer vendor to refund a change of mind.

The 2024 rule is also not the text in force as read. The FTC's Negative Option Rule page lists a 12 February 2026 Federal Register notice revising the rule to conform to federal court decisions and a 13 March 2026 advance notice of proposed rulemaking seeking comment on new amendments (FTC, Negative Option Rule page, read 9 November 2026). What survives either way is the agency's stated position that cancelling should be no harder than signing up, and the vendor table above shows which routes meet that standard on their face and which run through a support inbox.

Outside the US, a buyer's statutory rights differ by country and are not covered here. WriteHuman's "except where the law requires a refund" and StealthGPT's "waiver of withdrawal right for digital content" clause, both read 9 November 2026, are the two vendors in the table that acknowledge such rights in their text.

Read the refund page, not the pricing card: Humbot's pricing table says 30 days and its refund policy page says 7 with a word cap, both read 9 November 2026, and the policy page is the one a support agent will cite. Note the window in days and the usage cap in words, and test inside both. Put the renewal date in a calendar, because every vendor in the table that states a route says cancellation takes effect at the end of the paid term.

If the question is whether to pay at all, the free allowances and paid trials on the free humanizer page let you see output before a refund term matters, and the guides hub carries the read date on every page so a term quoted here can be checked against the vendor's page on the day you buy. The published method will score what each tool does to meaning and grammar once a run exists; it will never score a refund promise.