AI Humanizer BenchmarkUpdated October 1, 2026
The open benchmark for AI humanizers.
Every month we run 11 AI humanizers through 7 AI detectors, including GPTZero, Originality.ai and Copyleaks, to find out which ones get past them without ruining the text. Each tool rewrites the same 33 texts and is scored on bypass rate, meaning preservation and readability. Every prompt, output and score is public on GitHub.
Currently leading
October 2026AI Humanizer Benchmark: October 2026 Rankings
Each tool gets one overall score, based on 7 detectors and 7 writing categories. How scores are computed
- Last tested
- Sample size
- 33
- Methodology
- v1.1.0
| Rank | Humanizer | Penalties | Trend | Raw data | ||||
|---|---|---|---|---|---|---|---|---|
| 1– |  UndetectedGPTundetectedgpt.ai | 88.12 | 96.0 | 77.4 | 83.6 | none | .jsonl | |
| 2▲4 | 83.87 | 93.5 | 77.9 | 64.3 | none | .jsonl | ||
| 3▲1 | 80.74 | 88.7 | 68.3 | 77.7 | none | .jsonl | ||
| 4▼2 | 80.20 | 78.2 | 82.9 | 77.5 | none | .jsonl | ||
| 5– | 78.33 | 89.5 | 66.4 | 72.1 | -1.0 | .jsonl | ||
| 6▲2 | 78.23 | 83.9 | 75.2 | 71.0 | -1.0 | .jsonl | ||
| 7▼4 | 75.06 | 97.6 | 53.2 | 71.4 | -4.0 | .jsonl | ||
| 8▲1 | 68.06 | 85.9 | 65.5 | 76.8 | -9.0 | .jsonl | ||
| 9▼2 | 64.52 | 62.8 | 67.7 | 80.6 | -4.0 | .jsonl | ||
| 10– | 41.60 | 1.7 | 68.7 | 82.7 | -4.0 | .jsonl | ||
| 11– | 33.60 | 1.8 | 72.9 | 73.2 | -12.0 | .jsonl |
11 humanizers · 7 detectors · scores out of 100, higher is better. Click a column to sort; hover a penalty for the breakdown.
View cycle dataHow scores are computed
See the full methodologyOverall score formula
How often the rewritten text is classified as human by the detector panel.
How closely the rewrite preserves the original text's meaning.
Whether the output is clear, fluent, natural writing.
Whether performance holds across all seven writing types or varies widely between them.
Penalties (right) subtract from the composite, and the result never goes below zero.
Possible penalties
- Identical to input
The text came back essentially unchanged; no meaningful rewriting occurred.
-2.0-10.0 - Refusal
The tool declined to process the text, typically due to its own content filter.
-1.0-10.0 - Meaning drift
The rewrite departed substantially from the original's meaning.
-1.0-10.0 - Length inflation
The output ran far longer than the input, diluting the AI signal with added words rather than removing it.
-1.0-10.0 - Length deflation
The output ran far shorter than the input; content was dropped rather than rephrased.
-1.0-10.0
Maximum possible total penalty: -50.0 (off 100).
How it works
- 01
Generate the test texts
Each cycle starts with new AI-written texts in 7 writing categories, from essays and emails to blog posts and marketing copy. The texts are locked before any humanizer runs.
- 02
Run every humanizer on them
Every humanizer rewrites the same 33 texts. Each one is run through API calls on its default settings, not a setup tuned for the test.
- 03
Check with 7 detectors
Every rewrite is scored by GPTZero, Originality.ai, Copyleaks, Winston AI, ZeroGPT, QuillBot and Grammarly. Each detector's result is only converted to a common 0–1 scale. Nothing else is adjusted.
- 04
Grade meaning and readability
Semantic similarity to the input text measures how much meaning each rewrite kept. A language model rates how readable it is.
- 05
Publish all the data
Each monthly cycle is released in full on GitHub. Inputs, outputs, each detector's verdicts and the scoring code all stay there permanently.
The detector panel
Each text's bypass result is the median across all 7 commercial AI detectors, so no single detector outweighs the others. Getting past one detector proves little on its own. Open any detector below to see the humanizers ranked by that detector alone.
GPTZero
Widely deployed in education; often the first check a text faces
Originality.ai
Paid tool that publishers and SEO teams run before content goes live
Copyleaks
Used by universities and businesses, where a flag can have formal consequences
Winston AI
Commercial detector issuing document-level verdicts
ZeroGPT
Free browser detector with very large consumer reach
QuillBot
Detection built into the QuillBot writing suite
Grammarly
Detection inside Grammarly's writing products
+Suggest a detectorKnow one we should test against?Why this exists
Much of what is published about AI humanizers is paid for or written by the vendors, often without saying how anything was tested. We wanted real numbers. So we run a monthly test in which every tool gets the same texts, detectors and formula, and every figure links back to raw inputs and outputs that anyone can inspect.
Our team also builds UndetectedGPT, which is included in the rankings. Because of that, the inputs, outputs, AI detector verdicts and scoring script are public, so anyone can check and recompute the results.
Read the full storyBy use case
The main leaderboard uses one formula for every tool. Some readers want a narrower answer, like the best AI humanizer for essays or the one that does best against a single detector. These views reweight or filter the results to answer that.
Best for Quality
Ranks tools on the quality of the rewrite alone: meaning kept, readability, and penalties. Detector results are left out.
Best for Meaning preservation
Only meaning preservation counts here: how closely each tool's rewrites keep what the original text said.
Best for Readability
Only readability counts here: how clear and fluent each tool's rewrites are.
Best for Bypass rate
Only bypass rate counts here: how human each tool's rewrites scored across every AI detector we test.
Best for Students
For coursework that will be run through an AI detector. Academic essay results count double.
Best for Essays
Ranks tools on essay writing, where the argument has to survive the rewrite. Both essay categories count double.
Best for GPTZero
Only GPTZero counts here. Tools are ranked by the scores GPTZero gave their outputs.
Best for Originality.ai
Only Originality.ai counts here. It is one of the strictest detectors in our panel.
Best for Turnitin
Turnitin is not open to automated testing, so Originality.ai stands in for it. It is one of the strictest detectors in our panel.
Best for SEO
For published content that has to get past editors as well as detectors. Blog post and marketing copy results count double.
Best for Academic writing
For theses, papers and applications. Academic essay and application essay results count double.
Best for Marketing
For landing pages, product descriptions and other promotional text. Marketing copy and blog post results count double.
Best for Free
Just the humanizers that have a free tier, ranked by their overall score.
Frequently asked questions
What is an AI humanizer?+
A tool that rewrites AI-generated text so AI detectors label it as human-written. It changes sentence structure, rhythm and word choice while trying to keep the meaning the same. Many of these tools make the writing worse, so this benchmark scores meaning preservation and readability as well as bypass rate.
How are the tools tested?+
Each month, every humanizer rewrites the same set of newly generated texts on its default settings. The texts cover essays, emails, blog posts and other kinds of writing. Then 7 commercial AI detectors score each rewrite, and every rewrite is also measured for meaning preservation and readability.
What does the score out of 100 mean?+
It combines four measurements: bypass rate against the detectors (42%), meaning preservation (32%), readability (16%) and consistency across writing types (10%). Penalty points are subtracted for quality problems such as drifting from the original meaning, output that is much longer or shorter than the input, and refusals. A tool that beats detectors but wrecks the text does not rank high.
Which AI detectors do you test against?+
We test against 7 detectors: GPTZero, Originality.ai, Copyleaks, Winston AI, ZeroGPT, QuillBot and Grammarly. They are used in education, publishing and business. Each humanizer's page shows its results for every detector, and each detector has its own page that ranks the humanizers by bypass rate against that detector.
How often do the rankings change?+
A new cycle runs each month on newly generated test texts. The draw that picks them stays secret until the cycle closes, so vendors cannot prepare for them. Rankings can shift between cycles. Past cycles stay published and unchanged, so you can see how any tool has moved over time.
Do tools pay to be listed or ranked?+
No. The site carries no sponsored content and no affiliate links, and no vendor can pay for a position. A tool's rank depends only on its test results.
UndetectedGPT operates this benchmark. How is bias handled?+
AI Humanizer Benchmark is run by the team that makes UndetectedGPT, and UndetectedGPT is ranked here. That is a potential conflict of interest, so we publish each monthly cycle in full: inputs, outputs, detector verdicts, scoring rules and the audit bundle. UndetectedGPT goes through the same pipeline as every other tool, with the same prompts, settings, detectors and scoring code. Anyone can review our results and recalculate them from the public data.
Can I check the results myself?+
Yes, in two ways. For a quick check, take any published output, run it through the detector yourself and compare the verdict with ours. For a full check, clone the public data repository and run one command. It rebuilds the whole leaderboard from the raw inputs, outputs and detector scores.