isGenAI research · Weekly tracker
How much of Hacker News looks AI-written?
8 November 2026: 2% of the 183 long front-page comments we scored read as AI-written (4 comments at or above a score of 0.5, under 30 top stories). The story with the highest share was Writing as a thinking tool at 7% of 14 comments. Every number below comes from the open run file; a score is a lead, not proof.
Front-page stories, 8 November 2026
Usernames in this run
177 Hacker News usernames had at least one comment scored. Usernames are public on Hacker News; the text of each comment stays there, so open the profile to read what was scored. A high mean from one comment is weak evidence; look at the count.
Showing the 50 highest mean scores of 177 usernames. Type to search all of them.
| Username | Comments scored | Mean score | Highest score | Profile |
|---|---|---|---|---|
| mentalgear | 1 | 0.90 (AI-looking) | 0.90 | HN profile |
| patryk_builds | 1 | 0.75 (AI-looking) | 0.75 | HN profile |
| mohsen1 | 1 | 0.72 (AI-looking) | 0.72 | HN profile |
| sigttou | 1 | 0.50 (AI-looking) | 0.50 | HN profile |
| emidoots | 1 | 0.35 | 0.35 | HN profile |
| pteetor | 1 | 0.30 | 0.30 | HN profile |
| xmddmx | 1 | 0.30 | 0.30 | HN profile |
| LaurensBER | 2 | 0.25 | 0.40 | HN profile |
| carschno | 1 | 0.25 | 0.25 | HN profile |
| Jamesbeam | 1 | 0.25 | 0.25 | HN profile |
| le-mark | 1 | 0.25 | 0.25 | HN profile |
| theapiartist | 1 | 0.25 | 0.25 | HN profile |
| fancyfredbot | 1 | 0.22 | 0.22 | HN profile |
| cco | 1 | 0.20 | 0.20 | HN profile |
| drewbeck | 1 | 0.20 | 0.20 | HN profile |
| Ericson2314 | 1 | 0.20 | 0.20 | HN profile |
| hnlmorg | 1 | 0.20 | 0.20 | HN profile |
| vgalin | 1 | 0.20 | 0.20 | HN profile |
| w10-1 | 1 | 0.20 | 0.20 | HN profile |
| armanckeser | 2 | 0.15 | 0.15 | HN profile |
| anuptalwalkar | 1 | 0.15 | 0.15 | HN profile |
| baxtr | 1 | 0.15 | 0.15 | HN profile |
| billypilgrim | 1 | 0.15 | 0.15 | HN profile |
| bilsbie | 1 | 0.15 | 0.15 | HN profile |
| hemkeshr | 1 | 0.15 | 0.15 | HN profile |
| imgyuri | 1 | 0.15 | 0.15 | HN profile |
| ivan_gammel | 1 | 0.15 | 0.15 | HN profile |
| kkotak | 1 | 0.15 | 0.15 | HN profile |
| Kuyawa | 1 | 0.15 | 0.15 | HN profile |
| mococa | 1 | 0.15 | 0.15 | HN profile |
| ohyes | 1 | 0.15 | 0.15 | HN profile |
| poopingatwork | 1 | 0.15 | 0.15 | HN profile |
| raincole | 1 | 0.15 | 0.15 | HN profile |
| thenoblesunfish | 1 | 0.15 | 0.15 | HN profile |
| therealmarv | 1 | 0.15 | 0.15 | HN profile |
| tim-projects | 1 | 0.15 | 0.15 | HN profile |
| toolbrewlab | 1 | 0.15 | 0.15 | HN profile |
| tsergiu | 1 | 0.15 | 0.15 | HN profile |
| microflash | 2 | 0.13 | 0.15 | HN profile |
| Animats | 1 | 0.12 | 0.12 | HN profile |
| armcat | 1 | 0.12 | 0.12 | HN profile |
| armchairhacker | 1 | 0.12 | 0.12 | HN profile |
| bbor | 1 | 0.12 | 0.12 | HN profile |
| bottlepalm | 1 | 0.12 | 0.12 | HN profile |
| brabel | 1 | 0.12 | 0.12 | HN profile |
| donatj | 1 | 0.12 | 0.12 | HN profile |
| jjallen | 1 | 0.12 | 0.12 | HN profile |
| jonathanstrange | 1 | 0.12 | 0.12 | HN profile |
| mvdwoord | 1 | 0.12 | 0.12 | HN profile |
| saidnooneever | 1 | 0.12 | 0.12 | HN profile |
Tag @isgenai on X or paste text at /ai-text-detector to check your own writing.
Method and limits
- A score is a lead, not proof. The detector returns an AI probability from 0 to 1 for one comment. A comment at or above 0.5 counts as AI-looking here. Polished, formal or list-heavy human writing can score high; edited or short AI output can score low.
- Detector: isGenAI text detector (DeepSeek reasoning scorer, policy 2aa4be4aa9be…). The identity string in the run file (deepseek:2aa4be4aa9beb07b9cfb8af42d6692369fb46f58b2ddcd34fc300e55ce1f9173) changes whenever the prompt, model or endpoint changes, so runs can be compared only when it matches.
- Sample: the top 30 stories from the official Hacker News API at run time, up to 15 top-level comments per story, top-level comments only (no replies), at most 300 comments per run. Comments under 300 characters are not scored because short text gives unreliable verdicts; this run skipped 233 short comments and 8 comments the detector could not score.
- Bias: long comments are a minority of Hacker News comments and lean towards explanatory, essay-like writing, which is also the register the detector flags most. The share is a share of long front-page comments, not of Hacker News.
- Privacy: only the comment id, the public username, the length, the time and the score are stored. Nothing is kept about comment text, and nothing is written back to Hacker News.
- Repeat it: every comment id links to its Hacker News item; paste the text into /ai-text-detector to see the same score with highlights. Corrections: /corrections.
Data: latest.json (this run) and dated copies at /research/hn-ai-tracker/runs/<date>.json, for example runs/2026-11-08.json. CC0-1.0.