Transparency and Testing
We want to be transparent about how Humanize AI performs and how we test it. Our Humanizer offers two re-write modes:
-
Balanced — preserves your natural tone and style while reducing AI‑detector scores. In our internal tests, we aim to keep scores under ~30% on commonly used detectors.
-
Ultra — stronger transformation for the lowest detector scores, while keeping the meaning intact.
Detectors we monitor
We regularly check results against four widely used detectors:
-
GPTZero — gptzero.me
-
QuillBot AI Detector — quillbot.com
-
ZeroGPT — zerogpt.com
-
Grammarly — grammarly.com
Humanizer(Paraphraser) Latest results (as of 1 Sep 2025)
Arithmetic means across the 20 topics provided; detectors: ZeroGPT, QuillBot, Grammarly, GPTZero.
| Detector | Raw AI (baseline) | Humanize — Ultra | Humanize — Balanced |
|---|---|---|---|
| ZeroGPT | 99.76% | 8.51% | 18.60% |
| QuillBot | 95.85% | 5.50% | 24.65% |
| Grammarly | 73.60% | 5.00% | 9.10% |
| GPTZero | 92.85% | 4.75% | 26.85% |
Notes
-
These are simple averages over 20 topics; your mileage may vary by text, topic, and detector version.
-
Detector vendors update models and thresholds; scores can change over time.
Source data
https://docs.google.com/spreadsheets/d/1kc8bPs6R6rX2mbKoL2kDcGGRepebG7iABxbYAv7lWFY/edit?gid=0#gid=0