Best AI Humanizer for Copyleaks
Copyleaks (copyleaks.com) is an enterprise AI checker and plagiarism-detection platform used by universities, businesses, and content-moderation teams. In October 2026, Copyleaks caught 58 of the 420 outputs from the 14 AI humanizers we tested (13.8%), the third lowest catch rate of the six detectors. That is down from 32.9% in September 2026, and ten of the 12 humanizers tested in both months passed it more often. We only test humanized AI text, so this says nothing about how Copyleaks treats human writing.
As of October 2026, StealthGPT and Phrasly are tied as the best AI humanizers for Copyleaks on HumanizerBench, each with the highest average Copyleaks score (100.0), followed by Clever AI Humanizer (96.7).
14 humanizers · 30 prompts each · Last tested · Methodology v1.3.0
Copyleaks catch rate
| Month | Rate Catch rate | Caught | Outputs | Tools Humanizers | Avg score Avg human score | Rank |
|---|---|---|---|---|---|---|
| 13.8% | 58 | 420 | 14 | 85.9 | 4 of 6 | |
| 32.9% | 146 | 444 | 14 | 67.3 | 3 of 5 | |
| 42.7% | 169 | 396 | 12 | 57.2 | 3 of 5 | |
| 41.8% | 179 | 428 | 13 | 58.1 | 3 of 5 | |
| 40.0% | 156 | 390 | 13 | 60.1 | 3 of 5 |
Swipe the table sideways to see every column.
Caught means Copyleaks gave a human-likelihood score below 0.50 on a 0 to 1 scale. Avg human score is the mean score on a 0 to 100 scale. Rank 1 caught the most that month.
Humanizers that pass Copyleaks
Every AI humanizer we tested against Copyleaks in October 2026, ordered by its average Copyleaks score. 10 of 14 passed Copyleaks on at least 80% of their outputs. The top three passed 30, 30 and 29 of their 30 outputs, and first place changed hands in each of the last 4 monthly updates.
| Rank Position by average Copyleaks score this cycle. Learn more → | Humanizer | Average of Copyleaks's human-likelihood scores on this tool's outputs, where 100 means every output was scored as human. The table is ordered by this column. Learn more → ↓ | Outputs Copyleaks scored as human (a human-likelihood of 0.50 or higher), out of the tool's outputs it scored. Ranked by share passed. Click a count to read this tool's outputs. Learn more → | This tool's Copyleaks score in September 2026. "New" means it was not tested that month. Learn more → | Overall score across all detectors, after penalties. Learn more → |
|---|---|---|---|---|---|
| 1 | 100.0 | 30/30 | 81.8 | 74.50 | |
| 2 | 100.0 | 30/30 | 81.8 | 61.79 | |
| 3 | 96.7 | 29/30 | new | 69.02 | |
| 4 | 96.7 | 29/30 | 81.8 | 65.98 | |
| 5 | 93.3 | 28/30 | 86.7 | 74.89 | |
| 6 | 93.3 | 28/30 | new | 61.74 | |
| 7 | 89.4 | 26/30 | 59.1 | 63.53 | |
| 8 | 87.9 | 27/30 | 74.3 | 71.18 | |
| 9 | 83.3 | 25/30 | 63.7 | 67.58 | |
| 10 | 81.7 | 25/30 | 84.9 | 65.14 | |
| 11 | 78.2 | 23/30 | 85.3 | 62.75 | |
| 12 | 73.6 | 23/30 | 60.6 | 68.24 | |
| 13 | 70.5 | 21/30 | 53.5 | 68.87 | |
| 14 | 58.2 | 18/30 | 60.3 | 61.29 |
Swipe the table sideways to see every column.
Gold, silver and bronze mark the top three in this month's columns, ties sharing a color; Passed is ranked by share of outputs passed.
Copyleaks compared with the other detectors
Copyleaks gave the same verdict as Winston AI most often (84.3% of outputs) and as Pangram least often (31.0%).
| Detector | Caught | Rate Catch rate | Avg score Avg human score | vs Sep 2026 Change since Sep 2026 | Only it Only it caught | Agrees Same verdict as Copyleaks |
|---|---|---|---|---|---|---|
| 348 of 420 | 82.9% | 16.9 | n/a | 147 | 31.0% | |
| 151 of 420 | 36.0% | 61.3 | −11.2 | 1 | 66.9% | |
| 62 of 420 | 14.8% | 76.5 | −3.9 | 0 | 82.4% | |
 Copyleaks | 58 of 420 | 13.8% | 85.9 | −19.1 | 0 | |
| 48 of 420 | 11.4% | 87.2 | −12.1 | 1 | 81.4% | |
| 30 of 420 | 7.1% | 91.8 | −28.3 | 1 | 84.3% |
Swipe the table sideways to see every column.
Change is in points of catch rate. Only it caught counts outputs no other detector caught, of the 420 all six scored.
What Copyleaks catches
By writing type, Copyleaks caught humanized academic essays most often (21%, 18 of 84) and marketing copy least often (5%, 4 of 84).
| Caught by | Academic Essay | Application Essay | Blog Post | Discussion Board | News Article | Marketing Copy |
|---|---|---|---|---|---|---|
| Copyleaks | 21% 18/84 | 19% 16/84 | 17% 14/84 | 7% 3/42 | 7% 3/42 | 5% 4/84 |
| Other five detectors | 28% 118/420 | 31% 132/420 | 32% 133/420 | 31% 66/210 | 22% 46/210 | 34% 144/420 |
Swipe the table sideways to see every column.
The second row pools the other detectors' verdicts on the same outputs. Each type had 3 to 6 prompts this month.
Outputs only Copyleaks caught
No output was caught by Copyleaks alone in October 2026.
Of the 30 prompts, Copyleaks caught the largest share of tools on two prompts, including Argumentative Essay · claude-sonnet-5-5 (7 of 14) and no tool on 11 prompts, including Product Desc · gemini-3-8-flash (0 of 14). All 30 October 2026 prompts
Questions about Copyleaks
Is Copyleaks accurate?
On humanized AI text, Copyleaks has caught between 13.8% (October 2026) and 42.7% (August 2026) of outputs in the five months we have tested it. We only test AI-written passages after a humanizer rewrites them, so we cannot say how often Copyleaks flags human writing by mistake.
Can Copyleaks be bypassed?
In October 2026, 362 of the 420 humanized outputs (86.2%) passed Copyleaks as human-written. It depended on the tool: the 14 humanizers passed it on between 60% and 100% of their outputs.
What is the best AI humanizer for Copyleaks?
As of October 2026, StealthGPT and Phrasly are tied as the best AI humanizers for Copyleaks on HumanizerBench, each with the highest average Copyleaks score (100.0), followed by Clever AI Humanizer (96.7).
How does Copyleaks compare with other AI detectors?
In October 2026, Copyleaks had the third lowest catch rate of the six detectors we test: Pangram 82.9%, GPTZero 36.0%, ZeroGPT 14.8%, Copyleaks 13.8%, Originality.ai 11.4% and Winston AI 7.1%.
What kind of writing does Copyleaks flag most?
Over the five months we have tested it, Copyleaks caught humanized business emails most often (103 of 115, 90%) and discussion board posts least often (34 of 198, 17%).
- Other detectors
- All detectors
GPTZero
Originality.ai
Winston AI
ZeroGPT
Pangram
- Picking a tool
- LeaderboardCompare toolsRankings by use case
- Data
- October 2026 evidence detector-scores.json Methodology