AI Humanizer BenchmarkDetectors / GPTZero
Best for GPTZero
In the October 2026 cycle, GPTZero scored 363 outputs from the 11 humanizers in the benchmark and caught 203 (55.9%), the most of the seven detectors. The test set has no human-written text, so false positives are not measured here.
Highest average GPTZero human-likelihood this cycle: WriteHuman (100.0), CleverHumanizer (78.3) and UndetectedGPT (72.0).
One of the earliest AI text detectors, widely used in education to check whether student writing was generated with AI.
How we read its score: Reports a 0-1 AI probability; inverted to human = 1 - ai.
11 humanizers × 33 prompts = 363 outputs · run October 1, 2026 · methodology v1.1.0
Humanizer ranking against GPTZero
Sorted by the average human-likelihood GPTZero gave each tool's 33 outputs: its bypass rate against GPTZero. Only one of the 11 tools had 80% or more of its outputs passed by GPTZero. The top three had 33, 26 and 25 of 33 passed. In September 2026, CleverHumanizer ranked first.
| Rank | Humanizer | Penalties | Last tested | ||
|---|---|---|---|---|---|
| 1 | 100.0 | 75.06 | -4.0 | ||
| 2 | 78.3 | 68.06 | -9.0 | ||
| 3 |  UndetectedGPTundetectedgpt.ai | 72.0 | 88.12 | none | |
| 4 | 55.1 | 80.74 | none | ||
| 5 | 47.6 | 78.23 | -1.0 | ||
| 6 | 41.5 | 64.52 | -4.0 | ||
| 7 | 40.9 | 78.33 | -1.0 | ||
| 8 | 28.3 | 80.20 | none | ||
| 9 | 12.0 | 33.60 | -12.0 | ||
| 10 | 4.0 | 83.87 | none | ||
| 11 | 0.0 | 41.60 | -4.0 |
11 humanizers · 7 detectors · scores out of 100, higher is better. Click a column to sort; hover a penalty for the breakdown.
Pass rate on GPTZero vs the other detectors
Five humanizers passed GPTZero at least 30 percentage points less often than the other six detectors in October 2026: HIX Bypass, SmartHumanizer, AI Humanize, SuperHumanizer and GPTinf. They do well elsewhere but poorly against GPTZero specifically.
| Humanizer | Passed GPTZero | Passed the other 6 | Difference | Cycles 30+ pts lower |
|---|---|---|---|---|
| 3% 1/33 | 92% 183/198 | −89.4 | 2 of 2 | |
| 27% 9/33 | 80% 158/198 | −52.5 | 1 of 2 | |
| 45% 15/33 | 89% 176/198 | −43.4 | 1 of 2 | |
| 45% 15/33 | 81% 161/198 | −35.9 | 1 of 2 | |
| 55% 18/33 | 87% 172/198 | −32.3 | 1 of 2 | |
| 42% 14/33 | 65% 128/198 | −22.2 | 0 of 2 | |
| 0% 0/33 | 20% 40/198 | −20.2 | 0 of 2 | |
| UndetectedGPT | 76% 25/33 | 93% 184/198 | −17.2 | 0 of 2 |
| 79% 26/33 | 83% 165/198 | −4.5 | 0 of 2 | |
| 12% 4/33 | 9% 18/198 | +3.0 | 0 of 2 | |
| 100% 33/33 | 96% 191/198 | +3.5 | 0 of 2 |
Passed means a human-likelihood of 0.50 or higher. The other six detectors' verdicts on the same outputs are pooled. Difference is the pass rate on GPTZero minus the pooled rate, in percentage points; negative means GPTZero was harder for that tool.
GPTZero compared with the other AI detectors
All seven detectors scored the same 363 outputs in October 2026. GPTZero gave the same verdict as Copyleaks most often (68.0% of outputs) and as Grammarly least often (57.6%). Despite the similar names, GPTZero and ZeroGPT are separate detectors: GPTZero caught 55.9% and ZeroGPT 19.6%.
| Detector | Caught | Catch rate | Change since Sep 2026 | Unique catches | Same verdict as GPTZero |
|---|---|---|---|---|---|
| 203 of 363 | 55.9% | +11.3 | 64 | – | |
| 145 of 363 | 39.9% | −19.6 | 13 | 68.0% | |
| 125 of 363 | 34.4% | −12.4 | 8 | 65.3% | |
| 99 of 363 | 27.3% | −30.3 | 0 | 64.2% | |
| 95 of 363 | 26.2% | +1.9 | 7 | 60.3% | |
| 71 of 363 | 19.6% | −1.7 | 1 | 57.6% | |
| 67 of 363 | 18.5% | +1.4 | 1 | 57.6% |
Change is in percentage points of catch rate. Unique catches are outputs a detector caught while every other detector passed them. Same verdict is the share of outputs where the detector and GPTZero both caught or both passed the output.
Outputs only GPTZero caught
In October 2026, GPTZero caught 64 outputs that all six other detectors passed: 36.8% of the 174 outputs the others let through. Largest score gap first; the top 10 are shown.
| Prompt | Humanizer | GPTZero score | Lowest other score |
|---|---|---|---|
| Argumentative essay · Gemini 2.5 Flash | GPTinf | 0.00 | 1.00 |
| Discussion post · Claude Sonnet 5 | SuperHumanizer | 0.00 | 1.00 |
| Argumentative essay · Gemini 2.5 Flash | HIX Bypass | 0.00 | 1.00 |
| Argumentative essay · GPT-5.6 Terra | HIX Bypass | 0.00 | 1.00 |
| Product description · Claude Sonnet 5 | AI Humanize | 0.00 | 1.00 |
| News article · GPT-5.6 Luna | CleverHumanizer | 0.00 | 1.00 |
| Discussion post · Gemini 3.5 Flash | UndetectedGPT | 0.00 | 1.00 |
| Personal statement · GPT-5.6 Luna | AI Humanize | 0.00 | 1.00 |
| Lit review · Claude Sonnet 5 | HIX Bypass | 0.00 | 1.00 |
| News article · Gemini 3.5 Flash | HIX Bypass | 0.00 | 0.99 |
Scores are human-likelihood from 0 to 1. Every other detector scored these outputs 0.50 or higher; the last column is the lowest of those scores. Each prompt opens its evidence page at that humanizer's output.
GPTZero catch rate by writing type
In October 2026, GPTZero caught business email outputs most often (82%) and academic essay outputs least often (47%).
| Writing type | GPTZero | Other 6 detectors |
|---|---|---|
| Academic essay | 47% 31/66 | 18% 72/396 |
| Application essay | 59% 39/66 | 38% 150/396 |
| Blog post | 50% 33/66 | 30% 117/396 |
| Business email | 82% 27/33 | 43% 85/198 |
| Marketing copy | 52% 34/66 | 26% 103/396 |
| Discussion board | 61% 20/33 | 21% 41/198 |
| News article | 58% 19/33 | 17% 34/198 |
The last column pools the other six detectors' verdicts on the same outputs. By prompt, GPTZero caught the most tools on Business email · GPT-5.6 Terra (10 of 11) and the fewest on How-to blog · Gemini 3.5 Flash (3 of 11).
GPTZero by month
| Month | Humanizers | Caught | Catch rate | Avg human-likelihood | Rank by catches |
|---|---|---|---|---|---|
| October 2026 | 11 | 203 of 363 | 55.9% | 43.6 | 1 of 7 |
| September 2026 | 11 | 162 of 363 | 44.6% | 53.9 | 4 of 7 |
Rank 1 is the detector that caught the most outputs that month.
Using these results to pick a humanizer
The ranking on this page uses GPTZero only. It does not account for the other six detectors, for how well a rewrite keeps the original meaning, or for readability. A tool can pass GPTZero while changing facts or reading poorly. The Overall column above and the leaderboard cover those.
Results can shift between cycles as detectors and humanizers update, so check which cycle a figure comes from, and each tool's history on its own page.
Before relying on a tool, run a sample of your own text through it, check the output in GPTZero, and compare it with your original for changed facts or dropped citations.
Frequently asked questions about GPTZero
Can GPTZero detect humanized AI text?
Some of it. In October 2026, GPTZero caught 203 of the 363 humanized outputs it scored (55.9%), the most of the seven detectors in the benchmark. The test set has no human-written text, so how often GPTZero flags human writing is not measured here.
Can GPTZero be bypassed?
Often. In October 2026, 160 of the 363 humanized outputs (44.1%) passed GPTZero as human. Only one humanizer had at least 80% of its outputs passed.
What is the best AI humanizer for GPTZero?
By GPTZero alone, WriteHuman: it had the highest average human-likelihood from GPTZero in October 2026 (100.0 out of 100), and GPTZero caught 0 of its 33 outputs. The leaderboard ranks tools on all seven detectors plus meaning and readability.
Which writing types does GPTZero catch most often?
Business email outputs, at 82% in October 2026. Academic essay outputs were caught least often, at 47%.
Are GPTZero and ZeroGPT the same detector?
No. They are separate products despite the similar names. On the same 363 outputs in October 2026, GPTZero caught 55.9% and ZeroGPT caught 19.6%, and they gave the same verdict on 57.6%.
How do you test GPTZero?
Each month, every humanizer rewrites the same 33 AI-written texts on its default settings, and GPTZero scores every rewrite. Each result is converted to a human-likelihood from 0 to 1, as described above. A score below 0.50 counts as caught.
Raw test data
A JSONL file with one row per test from the October 2026 cycle. Each row has the text GPTZero scored and the score it returned. To check a result, paste a row's text into GPTZero and compare.
Other detectors
Related
How 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).