DropSlop research · AI has found its way into the public comment box

public comments to federal agencies since ChatGPT are written by AI.

Key findings

Exploratory Organized campaigns send the same letter many times. We split comments that match another comment on the same rule (campaign copies) from comments that match none (standalone).

Check your own text

Paste a paragraph and see which phrases read machine-made. Free, no account. To read a whole site page by page, run a site audit.

Exploratory Each dot is one rule open for comment, placed by the share of its comments the detector flagged (this includes false alarms). Salmon dots drew unusually many flagged comments.

Comments grouped by the year they were posted, in pairs. Before ChatGPT, the detector’s hits are false alarms; they set the yardstick every later year is measured against.

Exploratory Comments of 150 words or more, split at 250 words. Each group is measured against its own comments from before ChatGPT.

Rules grouped into three kinds of subject. Few rules sit behind each group, so read the whole range, not the point.

Exploratory Known chatbot words, per million words of comment, before ChatGPT and since. A word counts only with 20 uses before ChatGPT across 10 rules. We never show comment text.

How we did it

Detector flags, with its false alarms taken out and its misses put back. The full method is on its own page.

Use these findings freely with a link back to this page.Every figure links to the data file it comes from.

Want to check a comment before you send it?

Paste your comment and see which phrases read machine-made.