AI Humanizer BenchmarkDetectors / Grammarly
Best for Grammarly
In the October 2026 cycle, Grammarly scored 363 outputs from the 11 humanizers in the benchmark and caught 67 (18.5%), the fewest of the seven detectors. The test set has no human-written text, so false positives are not measured here.
Highest average Grammarly human-likelihood this cycle: UndetectedGPT (96.8), WriteHuman (95.9) and AI Humanize (91.1).
The AI detector built into Grammarly's writing assistant, where much drafting happens.
How we read its score: Reports AI percentage; inverted and divided by 100.
11 humanizers × 33 prompts = 363 outputs · run October 1, 2026 · methodology v1.1.0
Humanizer ranking against Grammarly
Sorted by the average human-likelihood Grammarly gave each tool's 33 outputs: its bypass rate against Grammarly. Eight of the 11 tools had 80% or more of their outputs passed by Grammarly. The top three had 33, 32 and 32 of 33 passed. In September 2026, CleverHumanizer ranked first.
| Rank | Humanizer | Penalties | Last tested | ||
|---|---|---|---|---|---|
| 1 |  UndetectedGPTundetectedgpt.ai | 96.8 | 88.12 | none | |
| 2 | 95.9 | 75.06 | -4.0 | ||
| 3 | 91.1 | 78.33 | -1.0 | ||
| 4 | 88.9 | 83.87 | none | ||
| 5 | 88.4 | 80.74 | none | ||
| 6 | 88.2 | 68.06 | -9.0 | ||
| 7 | 83.5 | 78.23 | -1.0 | ||
| 8 | 81.6 | 80.20 | none | ||
| 9 | 69.0 | 64.52 | -4.0 | ||
| 10 | 50.2 | 41.60 | -4.0 | ||
| 11 | 20.6 | 33.60 | -12.0 |
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 Grammarly vs the other detectors
No humanizer passed Grammarly 30 or more points less often than the other detectors in October 2026. One passed Grammarly at least 30 points more often.
| Humanizer | Passed Grammarly | Passed the other 6 | Difference | Cycles 30+ pts lower |
|---|---|---|---|---|
| 97% 32/33 | 97% 192/198 | 0.0 | 0 of 2 | |
| 12% 4/33 | 9% 18/198 | +3.0 | 0 of 2 | |
| 91% 30/33 | 81% 161/198 | +9.6 | 0 of 2 | |
| 70% 23/33 | 60% 119/198 | +9.6 | 0 of 2 | |
| UndetectedGPT | 100% 33/33 | 89% 176/198 | +11.1 | 0 of 2 |
| 97% 32/33 | 80% 159/198 | +16.7 | 0 of 2 | |
| 97% 32/33 | 80% 158/198 | +17.2 | 0 of 2 | |
| 88% 29/33 | 70% 138/198 | +18.2 | 0 of 2 | |
| 94% 31/33 | 73% 145/198 | +20.7 | 0 of 2 | |
| 100% 33/33 | 76% 151/198 | +23.7 | 0 of 2 | |
| 52% 17/33 | 12% 23/198 | +39.9 | 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 Grammarly minus the pooled rate, in percentage points; negative means Grammarly was harder for that tool.
Grammarly compared with the other AI detectors
All seven detectors scored the same 363 outputs in October 2026. Grammarly gave the same verdict as ZeroGPT most often (86.8% of outputs) and as GPTZero least often (57.6%).
| Detector | Caught | Catch rate | Change since Sep 2026 | Unique catches | Same verdict as Grammarly |
|---|---|---|---|---|---|
| 203 of 363 | 55.9% | +11.3 | 64 | 57.6% | |
| 145 of 363 | 39.9% | −19.6 | 13 | 76.3% | |
| 125 of 363 | 34.4% | −12.4 | 8 | 80.2% | |
| 99 of 363 | 27.3% | −30.3 | 0 | 85.1% | |
| 95 of 363 | 26.2% | +1.9 | 7 | 85.7% | |
| 71 of 363 | 19.6% | −1.7 | 1 | 86.8% | |
| 67 of 363 | 18.5% | +1.4 | 1 | – |
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 Grammarly both caught or both passed the output.
Outputs only Grammarly caught
In October 2026, Grammarly caught 1 output that all six other detectors passed: 0.9% of the 111 outputs the others let through. Largest score gap first.
| Prompt | Humanizer | Grammarly score | Lowest other score |
|---|---|---|---|
| Listicle blog · Gemini 3.5 Flash | StealthGPT | 0.48 | 0.62 |
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.
Grammarly catch rate by writing type
In October 2026, Grammarly caught business email outputs most often (36%) and news article outputs least often (6%). News article was its least-caught type in both cycles.
| Writing type | Grammarly | Other 6 detectors |
|---|---|---|
| Academic essay | 11% 7/66 | 24% 96/396 |
| Application essay | 26% 17/66 | 43% 172/396 |
| Blog post | 21% 14/66 | 34% 136/396 |
| Business email | 36% 12/33 | 51% 100/198 |
| Marketing copy | 14% 9/66 | 32% 128/396 |
| Discussion board | 18% 6/33 | 28% 55/198 |
| News article | 6% 2/33 | 26% 51/198 |
The last column pools the other six detectors' verdicts on the same outputs. By prompt, Grammarly caught the most tools on Cover letter · GPT-5.6 Luna (5 of 11) and the fewest on Argumentative essay · Claude Haiku 4.5 (0 of 11).
Grammarly by month
| Month | Humanizers | Caught | Catch rate | Avg human-likelihood | Rank by catches |
|---|---|---|---|---|---|
| October 2026 | 11 | 67 of 363 | 18.5% | 77.7 | 7 of 7 |
| September 2026 | 11 | 62 of 363 | 17.1% | 78.9 | 7 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 Grammarly 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 Grammarly 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 Grammarly, and compare it with your original for changed facts or dropped citations.
Frequently asked questions about Grammarly
Can Grammarly detect humanized AI text?
Some of it. In October 2026, Grammarly caught 67 of the 363 humanized outputs it scored (18.5%), the fewest of the seven detectors in the benchmark. The test set has no human-written text, so how often Grammarly flags human writing is not measured here.
Can Grammarly be bypassed?
Often. In October 2026, 296 of the 363 humanized outputs (81.5%) passed Grammarly as human. Eight of the 11 humanizers had at least 80% of their outputs passed.
What is the best AI humanizer for Grammarly?
By Grammarly alone, UndetectedGPT: it had the highest average human-likelihood from Grammarly in October 2026 (96.8 out of 100), and Grammarly caught 0 of its 33 outputs. The leaderboard ranks tools on all seven detectors plus meaning and readability.
Which writing types does Grammarly catch most often?
Business email outputs, at 36% in October 2026. News article outputs were caught least often, at 6%.
How do you test Grammarly?
Each month, every humanizer rewrites the same 33 AI-written texts on its default settings, and Grammarly 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 Grammarly scored and the score it returned. To check a result, paste a row's text into Grammarly 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).