September 2026 prompt evidence

Each of these 33 prompts was written by a frontier model, then handed to all 14 ranked humanizers, and every rewrite was submitted to 5 commercial AI detectors. Every run was screen-recorded. Open a prompt to read the input, every tool's rewrite, each detector's verdict, and to watch the recording of that run.

Cycles: September 2026 August 2026 July 2026 June 2026

Prompts

Tools ranked

462

Humanizer outputs

2268

Detector verdicts

How to read these scores

Each detector returns a human-likelihood on a common 0 to 1 scale, where 1 means it judged the text human-written and 0 means it flagged it as AI. On these pages a verdict counts as passed when that score is at least 0.50, the midpoint of the detector's own scale. That threshold exists only to draw the chips: the bypass rate on the leaderboard is the mean of each test's median score across the 5 detectors, a continuous number, so a tool's pass count and its bypass rate will not be the same figure.

A few runs come back with fewer than 5 verdicts because a detector did not return a usable score for them. A run still counts as passing at least 4 detectors only if that many detectors actually scored it and passed it. A partially scored run is counted as caught if any detector that did score it flagged the output, and is otherwise listed separately as partly scored.

Meaning is the input↔output embedding cosine and readability is a language-model writing-quality rating, both published per test in tests.json. Words is the output's length as a multiple of the input's; the scoring code penalizes ratios above 1.40 or below 0.60. Full definitions live in the methodology.

Prompt	Category	Written by	Input words	Passed	Failed	Hardest	Open
01 Argumentative Essay	Academic Essay	GPT-5.5	434	10	9	GPTZero	→
02 Discussion Post	Discussion Board	Gemini 3.5 Flash	253	11	10	GPTZero	→
03 Personal Statement	Application Essay	GPT-5.5	364	11	9	GPTZero	→
04 Discussion Post	Discussion Board	Claude Sonnet 5	246	11	8	GPTZero	→
05 News Article	News Article	GPT-5.5	452	11	6	ZeroGPT	→
06 Product Desc	Marketing Copy	Claude Sonnet 5	192	9	10	GPTZero	→
07 Cover Letter	Application Essay	Claude Sonnet 5	405	1	14	Winston AI	→
08 Personal Statement	Application Essay	Claude Sonnet 5	361	10	8	GPTZero	→
09 Business Email	Business Email	Gemini 3.5 Flash	269	1	12	Winston AI	→
10 Lit Review	Academic Essay	Gemini 3.5 Flash	401	12	5	GPTZero	→
11 Discussion Post	Discussion Board	GPT-5.5	260	11	10	GPTZero	→
12 Argumentative Essay	Academic Essay	Gemini 3.5 Flash	374	12	10	GPTZero	→
13 Landing Copy	Marketing Copy	Gemini 3.5 Flash	296	7	9	Copyleaks	→
14 Argumentative Essay	Academic Essay	Claude Sonnet 5	425	8	11	GPTZero	→
15 Lit Review	Academic Essay	GPT-5.5	431	11	6	GPTZero	→
16 Howto Blog	Blog Post	GPT-5.5	442	10	6	ZeroGPT	→
17 Product Desc	Marketing Copy	Gemini 3.5 Flash	186	9	7	GPTZero	→
18 Lit Review	Academic Essay	Claude Sonnet 5	384	12	6	GPTZero	→
19 Personal Statement	Application Essay	Gemini 3.5 Flash	316	11	8	GPTZero	→
20 Landing Copy	Marketing Copy	GPT-5.5	315	9	11	GPTZero	→
21 Cover Letter	Application Essay	Gemini 3.5 Flash	354	4	13	Winston AI	→
22 Cover Letter	Application Essay	GPT-5.5	411	1	13	Winston AI	→
23 Landing Copy	Marketing Copy	Claude Sonnet 5	313	10	8	GPTZero	→
24 News Article	News Article	Gemini 3.5 Flash	366	8	12	GPTZero	→
25 Listicle Blog	Blog Post	GPT-5.5	469	7	7	ZeroGPT	→
26 Business Email	Business Email	GPT-5.5	248	1	14	GPTZero	→
27 Listicle Blog	Blog Post	Gemini 3.5 Flash	476	13	6	ZeroGPT	→
28 Product Desc	Marketing Copy	GPT-5.5	194	6	12	ZeroGPT	→
29 Listicle Blog	Blog Post	Claude Sonnet 5	404	8	6	GPTZero	→
30 Howto Blog	Blog Post	Claude Sonnet 5	415	11	7	Copyleaks	→
31 Business Email	Business Email	Claude Sonnet 5	278	1	12	Winston AI	→
32 Howto Blog	Blog Post	Gemini 3.5 Flash	454	9	8	Originality.ai	→
33 News Article	News Article	Claude Sonnet 5	405	4	13	Winston AI	→

"Passed" counts the ranked tools whose rewrite of that prompt at least 4 of the 5 detectors scored as human-written (0.50 or above); "Failed" counts those flagged by at least one detector that scored them; "Hardest" is the single detector that flagged the most rewrites of that prompt.