Humanize Lab

How we test AI humanizers

Short version: fixed passages, three detectors, same day, monthly re-runs. Published results are per tool version, and we say so.

The protocol

  1. Three fixed passages: 300-word essay paragraph, 200-word product description, 150-word email.
  2. Baseline AI-probability from three independent detectors, same day, same account.
  3. One pass through the tool, default settings, no manual edits.
  4. Re-score with the same three detectors. Record meaning drift separately.
  5. Re-run monthly. Detector models change; a six-month-old score is history.

What the numbers mean

Detector output is a probability, not a verdict. Moving 98% to 40% means the text changed bucket for one model on one day. It does not mean the text is human, and it does not mean a different detector will agree.

Independence

This site is independent and not affiliated with, endorsed by, or sponsored by QuillBot, Grammarly, ZeroGPT or any tool mentioned. Product names and logos are trademarks of their respective owners and are used here for identification and comparison only.

We do not accept payment for a placement or a score. If a tool does not help, the page says so.