How We Test
A ranking is only as good as the test behind it. This page sets out how we test humanizers and rewrite tools before we compare them, so you can judge any comparison on this site for yourself and spot where it falls short.
Updated
How we test humanizers: every tool gets the same three ChatGPT drafts (a short essay paragraph, a work email and a product description). We score meaning kept, readability, word limits, tones, the free tier and price as published. Each ranking shows its test date, and we retest when a tool changes.
How comparisons are ranked today: our comparisons rank tools on the facts each vendor publishes (free access, limits, tones, languages, price), each with the date we checked it. The three-draft method below is how output is judged when results are added to a page.
The three sample drafts
Every tool gets the same three drafts, word for word. Using the same text is the only way to make results comparable.
- A short essay paragraph. A single body paragraph of the kind a student might draft with ChatGPT's help, with one claim and one supporting point. It tests whether a tool can keep an argument intact in a formal register.
- A work email. A follow up email with a specific ask and a date. It tests whether a tool keeps the ask and the date.
- A product description. A short description with a product name, two features and a number. It tests whether names and figures come back unchanged.
Each draft is generated with ChatGPT from a fixed prompt, saved once, and reused for every tool. We don't edit the drafts between tools, and we don't regenerate them to suit any one product.
What we score
We score each tool on the same fields, in the same order, every time:
| Field | What we record |
|---|---|
| Meaning kept | Whether every point, name and number in the draft survives the rewrite (see the next section) |
| Readability | Whether the output reads naturally aloud, judged by the editor against the original |
| Word limits | The words a tool accepts per run, as the tool states it |
| Tones | Which tones or modes the tool offers |
| Free tier | What a visitor can do without paying, as the tool states it |
| Price | The lowest paid plan, with currency and billing period, as published on the tool's own pricing page on the test date |
We test as an ordinary visitor would, on the free tier first and a paid plan only where the free tier can't run the samples. Prices are copied from the vendor's pricing page and dated, never estimated.
We do not rank tools on detector scores. No rewrite tool can honestly promise how a detector will score text, and a score from one detector on one day says little about the next.
How we check meaning
Meaning is the field people care about most, so it gets its own check. Before testing, the editor lists the facts in each draft: every name, number, date and claim. After each run, the output is read against that list. A fact that is missing, changed or added counts against the tool, and the page says which one. A rewrite that reads well but loses the date in an email has failed.
Dates and retests
Every comparison page, starting with the best ChatGPT humanizer comparison, shows the date its tests were run, and that date matches the page's last updated date. If we rerun the tests, both change together. How we handle sources and corrections is set out in our editorial policy, and who runs the tests is on the about page.
We retest a tool when it changes in a way that could affect results: a new version, new limits, new tones or a new price. Readers can tell us about a change through the contact page, and we'll check it.
Limits of this method
Three drafts can't show everything a tool does. Some tools may do better on long documents, other languages or kinds of writing we don't test. Readability is a judgement, made by one editor, and we say so. And a tool can change the day after we test it, which is why every result carries a date.
The same method applies to our ChatGPT rewrite box whenever it appears in a comparison.
FAQ
Do you test your own box the same way?
Yes. When our own box appears in a comparison, it gets the same drafts, the same fields and the same date rule as every other tool.
Why don't you score detector results?
Because no tool can honestly promise a detector score, and detectors differ in how they score the same text. We score what a reader can check: meaning, readability, limits and price.
Can a company pay to be ranked higher?
No. No tool pays to be listed or ranked on this site.
