Editorial policy
The editorial policy of Best AI Humanizer rests on five rules: every tool on the same fields, every number from a named source with a date, every price checked on the vendor's own page, every score from a published run, and every change to a fact logged. This page sets out each rule and how it is applied.
How tools are chosen
A tool joins the test set when it meets three tests. It has a public web interface that accepts pasted text. It has a stated free tier or paid plan that accepts an input of 500 words, so that the test's essay-length inputs can run on it. And there is enough search demand for the tool that readers are weighing it against others. Twelve tools met those tests at launch; the list is on the AI humanizer rankings page.
A tool leaves the set only when its service closes or stops accepting inputs. A tool that changes its name keeps its row and its review under both names. Every change to the set is dated on the method page. A vendor cannot pay to be added, and a vendor cannot ask to be removed because of how its tool reads.
How tests run
Tests run under the protocol published on how we test AI humanizers: 24 AI-written inputs and 6 human-written controls, 12 tools each on its default mode and its strongest documented mode, five named detectors read on the run date with a screenshot, and two editors rating meaning kept and counting errors without knowing which tool produced the output. The weights are published before a run and fixed for a series: meaning kept 35, errors introduced 20, detector pass rate 30, limits and access 15.
One pass per input, no retries unless the tool errors, and the first output is the one rated. A run is never edited after it publishes.
| Rule | What it requires |
|---|---|
| Same fields | Every tool is described on verdict, best for, main limitation, free tier and cheapest plan, in the same order, on every page it appears on |
| One number per fact | A figure appears once in the site's data and is read from there by every page, every table and every download, so the same fact never has two values |
| Named source, same sentence | Every number names who measured it and when, in the sentence it appears in, with a link |
| Checked-on date | Every price and limit carries the day it was read on the vendor's own page |
| Logged change | Every change to a published fact or score is recorded in the corrections log with the date and the reason |
How vendor claims are treated
A vendor's claim about its own tool is quoted, attributed and dated. It is never adopted as the site's finding. When a vendor's page states that its tool passes a named detector, the review reports that the vendor states it, names the page and the date, and leaves the detector reading to the run. When a detector's maker states an accuracy or false positive rate, the site records it as the maker's own figure with the date it was read.
A figure the vendor's page does not state does not appear on the site. Nothing is estimated, averaged from third-party write-ups, or carried over from an earlier year without a fresh check.
The same-fields rule and the one-number rule
Every tool is described on the same fields, in the same order, with the same standard for the main limitation: the single most important thing the tool will not do, or will charge for, stated with a number. A main limitation that could apply to any tool is not specific enough to publish.
One fact has one value. Every price, limit, reading and score lives in one data file, and every page, table, structured data block and download reads from it. When a fact changes, it changes in one place, and the change is logged.
How prices are checked and dated
Prices and limits are read on the vendor's own pricing page or, where there is none, on the vendor's homepage. Several pricing pages default to a yearly toggle and show a monthly rate that is charged annually; the site states the rate and the billing period together, and states the monthly list price only where the page shows one. A price that did not render on the day it was checked is left out until it has been read. Every price carries the phrase "checked on" with the date.
Price is reported, never scored. No page names a cheapest or best-value tool.
How Walter Writes AI is treated
Walter Writes AI is one of the twelve tools in the test set. It is described on the same fields, held to the same standard for a specific main limitation, run on the same inputs, rated by the same editors, and linked in the same style and the same place as every other tool. It receives no extra placement, no badge and no different wording. Any reader who finds a page on which that is not true can report it through the corrections process below.
How corrections work
Anyone, reader or vendor, can report an error or dispute a reading through the contact page. The editors check the report against the source and the raw files. If a fact is wrong, the page is changed and the change is entered in the corrections log with the date, the page, what changed and why. If a dispute concerns a reading from a run, the dispute and the response are recorded as a dated note on the run page. If a number changes as a result, the change is a new run ID, never a silent edit.
Reports on how the policy itself is applied go to the same address. The about page describes who the editors are and how the site is paid for.
More about how this site works: About Best AI Humanizer, Corrections log, Contact the editors, How we test AI humanizers.
