Undetectable.ai Humanizer Review: Limits and Refund Terms

Undetectable.ai is a paste-box humanizer with Basic and Stealth modes, a published 10,000-character ceiling per run, and paid allowances measured in words per month. The useful caveat is contractual: its Pricing, Help, and Terms pages describe refunds differently. This review records those boundaries without testing detectors or submitting text.

The public pages let a buyer size the job and understand payment. They cannot forecast a detector score. I did not create an account, paste text, run the Humanizer, start a trial, or ask for a refund. Product-effect claims on the site therefore remain vendor claims.

Short answer: check the exact domain, count the real input in characters, choose the billing view you intend to buy, and read the refund and cancellation terms before entering payment details.

The checked public records say no. The undetectable.ai Terms identify Undetectable Inc. as the company doing business as Undetectable AI. The footer on undetectable.com instead names Verva, LLC. Similar domains are not a safe product identifier.

This distinction does not establish ultimate ownership, investment, or every relationship that might exist behind the sites. It establishes the narrower point a buyer needs: do not copy a price, limit, or refund rule from one domain and attach it to the other.

Identity rule: keep the full domain beside every number. A correct limit attached to the wrong Undetectable product is still a false comparison.

The logged-out AI Humanizer had one text area. Two mode buttons read Basic and Stealth, followed by General Writing, Essay, Article, Marketing Material, Story, and More. Page copy puts the ceiling at 10,000 characters per submission. The empty state counts words, so a reader meets both units on one screen.

The page title includes the word `Unlimited`, but the body still publishes a 10,000-character operation limit. Those statements can coexist only if `Unlimited` refers to something other than an unlimited single input. A heading should not be used to erase a more specific boundary in the page copy.

No detector output was generated for this review. I left the ceiling in characters. A passage's language, word lengths, whitespace, symbols, references, and specialist terms all affect what fits.

Editor snapshot: pasted text, 10,000 characters per operation, word counters in the interface, and no completed run tested.

The pricing page changes both the visible allowances and prices when its Monthly and Yearly controls are switched. The table below records what each rendered state displayed on August 26, 2026. It does not blend the two states into one plan ladder.

The same page labels the yearly state `50% off`, says `6 Months Free` and `Cancel Anytime`, and limits the free trial to 250 words. I did not calculate an annual total for the 50K card because that total was not printed in the checked state. The input box is measured in characters; the plan pool is measured in monthly words. For the broader rule, see four different limits hidden behind unlimited.

Pricing rule: record billing view, allowance, unit, and renewal period together. A monthly-equivalent price on a yearly plan is not the same thing as month-to-month billing.

The Terms page, marked `Last updated 01/01/2025`, authorizes recurring charges until cancellation. Its plain-English summary says California and New York residents can cancel in the dashboard. It tells other users to email `contact@undetectable.ai`, while section 6.3 tells the same non-California and non-New York group to email `support@undetectable.ai` from the address associated with the account. The public document does not resolve that email conflict.

The same Terms say cancellation stops future charges rather than refunding prior billing periods, unless law requires otherwise, and that cancellation immediately removes remaining words. That is a more specific consequence than the pricing-page label `Cancel Anytime` communicates on its own.

These are three current official formulations, not three independent guarantees. I would save the pricing and terms shown at checkout and ask the company to clarify both the applicable refund test and the correct cancellation address before relying on the broader wording. This page comparison is not legal advice.

Contract snapshot: the pages disagree about refund conditions, and the Terms contain two cancellation email addresses. Preserve the conflict instead of choosing the friendlier sentence.

The browser Humanizer offers Basic and Stealth plus six writing-purpose choices. The Help page says English is the primary focus, then lists secondary multilingual support. Counting the codes in that list produced 55 entries, including separate `zh-cn` and `zh-tw` codes.

The useful claim is not that all 55 codes perform equally. It is that the public Help page documents a broad list while explicitly prioritizing English, and the API documentation makes another model-level distinction. A multilingual trial should therefore use the exact language and model intended for production.

Language snapshot: 55 listed codes does not mean 55 independently verified or equally strong language experiences.

The Develop page says new accounts receive 250 word credits usable through the API, that API work draws from the same account word balance at the same cost as the web platform, and that the default rate limit is three requests per second per IP address. It also says there is no maximum usage limit, which describes account usage or throughput on that page, not the size of one request.

The checked Humanization API v2 documentation publishes that 50-character minimum but no maximum request size. Searches for `maximum`, `30,000`, and `30000` returned no match while the same extraction found the minimum and schema fields. The defensible conclusion is `maximum not documented on this page`, not `unlimited per request`.

Undetectable.ai also documents a separate Detector API. A Humanizer submission should not be described as automatically running that separate endpoint, and neither API response should be called a Turnitin result.

API snapshot: shared account credits, three requests per second by default, asynchronous jobs, and no published per-request maximum in the checked v2 page.

Use the public contract, not a promised score, to decide whether the product fits the job:

1. Confirm the address bar says `undetectable.ai`, not the separate `.com` product.
2. Count the exact passage in characters and keep it below the published per-run ceiling.
3. Estimate monthly usage in words without converting the operation limit into the same unit.
4. Choose Monthly or Yearly deliberately and record the full charge shown, not only the monthly-equivalent number.
5. Read what cancellation does to unused words before purchasing a larger allowance.
6. Save the refund terms and ask which detector, threshold, evidence, deadline, and email route apply.
7. For API work, test one disposable input, handle asynchronous polling, and enforce your own request-size guard until a maximum is documented.

Publisher disclosure: HumanPen is not a neutral observer here; it offers its own document humanizer. The product page lists TXT, PPTX, and DOCX as inputs. It says each one keeps its file type, while rewrite scope can come from a manual selection or an iThenticate/Turnitin report. Nothing in this article independently tests that description.

Eligible results can continue lowering AI for free.

Decision rule: use a pasted-text tool when its text, plan, and review workflow fit the job. Use a document workflow when editability, layout, citations, and selective scope are part of the deliverable.

On August 26, 2026, I opened the named first-party pages in Ego Lite and read their rendered state. I fetched HumanPen anonymously as a second route so its disclosure did not depend on a signed-in page.

• Undetectable.ai: AI Humanizer, Pricing, Terms, Help, Develop, Humanization API v2, and Detector API.
• Identity guardrail: undetectable.com.
• Publisher disclosure: HumanPen document humanizer.

No text or file was submitted, no detector or API was run, and no account, trial, subscription, cancellation, or refund flow was started. Page copy, prices, and terms can change, so repeat the check immediately before paying or integrating.

Evidence date: August 26, 2026. Every limit in this article belongs to the page, unit, and billing state named beside it.