# Ran the same paragraph through six detectors. The disagreement was not small **URL:** \ **Category:** Experiments & Case Studies\ **Created:** [September 17, 2026, 2:38am UTC](https://humanizeaiforum.com/t/ran-the-same-paragraph-through-six-detectors-the-disagreement-was-not-small/216 "2026-09-17T02:38:05Z")\ **Posts on this page:** 6\ **Page:** 1
**Author:** ![tallest.tower.2016](https://avatars.discourse-cdn.com/v4/letter/t/c89c15/32.png) [@tallest.tower.2016](https://humanizeaiforum.com/u/tallest.tower.2016)\ **Post date:** [September 17, 2026, 2:38am UTC](https://humanizeaiforum.com/t/ran-the-same-paragraph-through-six-detectors-the-disagreement-was-not-small/216/1 "2026-09-17T02:38:05Z")
ran the same paragraph through six AI detectors and the disagreement was not small. sharing the results because I couldn’t find a systematic comparison when I was looking for one. the paragraph: a methodology section from a published human-written research paper. chosen because it’s the writing style most likely to trigger false positives. results below. ---
**Author:** ![stormhaze71](https://avatars.discourse-cdn.com/v4/letter/s/e47774/32.png) [@stormhaze71](https://humanizeaiforum.com/u/stormhaze71)\ **Post date:** [September 17, 2026, 3:09pm UTC](https://humanizeaiforum.com/t/ran-the-same-paragraph-through-six-detectors-the-disagreement-was-not-small/216/2 "2026-09-17T15:09:07Z")
methodology sections are the worst-case scenario for detectors precisely because they’re supposed to be formulaic. passive voice, sequential steps, hedging language, consistent sentence structure. that pattern is what AI text looks like to models trained on general content. ---
**Author:** ![codewrath58](https://avatars.discourse-cdn.com/v4/letter/c/ecccb3/32.png) [@codewrath58](https://humanizeaiforum.com/u/codewrath58)\ **Post date:** [September 18, 2026, 11:18am UTC](https://humanizeaiforum.com/t/ran-the-same-paragraph-through-six-detectors-the-disagreement-was-not-small/216/3 "2026-09-18T11:18:07Z")
the spread you’re seeing is what the research predicts. on general content these tools agree more often. on academic and formal writing the disagreement rate is documented at 30-40% across tools. that’s not a bug, it’s the tools being accurate about genuine uncertainty in the signal. ---
**Author:** ![ErnestPrice](https://yyz1.discourse-cdn.com/flex033/user_avatar/humanizeaiforum.com/ernestprice/32/24_2.png) [@ErnestPrice](https://humanizeaiforum.com/u/ErnestPrice)\ **Post date:** [September 18, 2026, 9:16pm UTC](https://humanizeaiforum.com/t/ran-the-same-paragraph-through-six-detectors-the-disagreement-was-not-small/216/4 "2026-09-18T21:16:07Z")
share the numbers if you can. not asking for the paragraph itself, just the scores across tools. ---
**Author:** ![Appartement](https://yyz1.discourse-cdn.com/flex033/user_avatar/humanizeaiforum.com/appartement/32/34_2.png) [@Appartement](https://humanizeaiforum.com/u/Appartement)\ **Post date:** [September 19, 2026, 3:15pm UTC](https://humanizeaiforum.com/t/ran-the-same-paragraph-through-six-detectors-the-disagreement-was-not-small/216/5 "2026-09-19T15:15:07Z")
scores: GPTZero 71%, Originality.ai 58%, aichecker.tech 42%, aitextdetector.ai 38%, Proofademic 12%, Walter Writes detector 31%. [Proofademic](https://proofademic.ai/) returned the lowest score and the most useful output. the sentence-level breakdown showed which specific sentences drove the score on the tools that flagged high. every flagged sentence was either a topic sentence, a transition, or a passive-voice procedural description. none of the analytical content flagged. that sentence-level detail is what makes the high-flagging scores look bad. the tools flagging 58-71% were apparently flagging the most conventional, formulaic sentences in academic writing. not the parts that would indicate AI involvement. the practical lesson: a high score on a methodology section is almost meaningless without knowing which sentences drove it. the score alone tells you nothing. ---
**Author:** ![penhillcurrant](https://yyz1.discourse-cdn.com/flex033/user_avatar/humanizeaiforum.com/penhillcurrant/32/23_2.png) [@penhillcurrant](https://humanizeaiforum.com/u/penhillcurrant)\ **Post date:** [September 20, 2026, 3:12pm UTC](https://humanizeaiforum.com/t/ran-the-same-paragraph-through-six-detectors-the-disagreement-was-not-small/216/6 "2026-09-20T15:12:30Z")
the spread from 12% to 71% on the same paragraph from a published paper is the number I’m going to use the next time someone tells me detector scores are reliable enough to base decisions on.