Terms of use

Effective date: September 3, 2026

This page sets out the rules for aihumanizerbenchmark.com (the “Site”). By visiting or using the Site, you accept these rules. If you do not accept them, please stop using the Site.

What this site is

AI Humanizer Benchmark measures commercial AI humanizers in recurring cycles. In each cycle, each humanizer being tested rewrites one shared set of test texts. A fixed group of commercial AI detectors then scores every output, and a deterministic formula, published with the results, turns those verdicts into scores. We publish the rankings, the raw data, the scoring code, and the methodology.

This site publishes benchmark results. It does not offer an AI humanization service. Using the Site does not create an account, a service agreement, or a vendor relationship with us.

Operator and conflicts

The people who operate the Site also run UndetectedGPT, and UndetectedGPT is one of the humanizers ranked here. The fairness page explains that conflict and how we guard against it. In short, UndetectedGPT is run, scored and penalized exactly like every other tool. The test set is locked with a cryptographic commitment before a run begins, and all raw data is released, so any favorable treatment would show up in the detector verdicts that anyone can inspect. Decisions about the benchmark itself (which tools are tested, the scoring rules, and what gets published) are made separately from UndetectedGPT's commercial product team. Rankings are not for sale, and the Site has no affiliate links.

Use of the benchmark data

Anyone may quote, cite, and discuss the leaderboard and published results for any purpose, including reporting, study, research, product comparison, and commercial work. Licensing is set out in the public repository: CC BY 4.0 for the cycle data and MIT for the scoring code. When you reuse the data, credit “AI Humanizer Benchmark” and include a link. We encourage you to download the raw cycle bundles and run the published scoring code on the published inputs to check our numbers.

Any reuse must not give a materially false picture of how the methodology works, how scoring is done, or how any tool performed.

What the results mean

Each result is a snapshot. It shows what third-party humanizers produced, and how third-party detectors rated that output, when a given cycle ran. These vendors update their products often, so a tool may perform differently today than its published result suggests.

A detector score shows what a commercial detector concluded during that cycle. It does not prove who wrote a text, whether a text is original, or whether anyone broke academic integrity rules. Treat the leaderboard as one input among several. Do not rely on it alone when making a purchase or an academic, legal, or business decision.

Trademarks and third-party names

Every product name, logo, and trademark mentioned on the Site is the property of its respective owner. We use these names only to identify the tools and detectors we test. Their appearance here does not mean that any owner endorses, sponsors, or is affiliated with the Site, or that we endorse them.

The newsletter form collects your email address. We keep it and use it only to send occasional benchmark news. Every email includes an unsubscribe link. The privacy policy explains how the address is handled.

Acceptable use

Please do not interfere with the Site's operation, crawl or scrape it so heavily that it puts an unreasonable load on it, sign up someone else's email address for updates, or misstate what the methodology or results say.

Vendor concerns and corrections

Vendors of humanizers or detectors who think our methodology was applied incorrectly to their product can file a dispute. We check each dispute against the methodology and scoring code we have published, and we publish the outcome. A published cycle is never changed. If a correction is justified, it appears openly in a later cycle. Disagreeing with the methodology is not, on its own, a reason for us to change results that have already been published.

No warranties

We provide the Site, the leaderboard, the data, and all scoring output “as is” and “as available.” We make no warranty of any kind, express or implied. This includes, without limitation, any warranty of merchantability, of fitness for a particular purpose, or of accuracy.

Limitation of liability

As far as the law allows, neither the operator of the Site nor its officers, employees, or agents will be liable for punitive, special, incidental, indirect, or consequential damages, or for lost profits, lost revenue, lost data, or harm to goodwill. This covers any claim that arises from, or relates to, your use of the published data, the leaderboard, or the Site, including times when you could not use them. For direct damages, total liability is capped at US$100 (one hundred United States dollars).

Changes to these terms

We may update these terms. The “effective” date above shows when the current version started to apply. If you keep using the Site after an update, you accept the updated version.

Contact

Send questions about these terms to contact@aihumanizerbenchmark.com.