Check any paper for AI-generated text, with no limits and no sign-up. Verdict reads LaTeX and PDF, marks the passages that look machine-written, and publishes its own false-positive rate. A screening signal, not proof.
| Verdict | GPTZero · Pangram · Quillbot | |
|---|---|---|
| Checks | Unlimited, no sign-up | Metered or paywalled |
| LaTeX and .tex papers | Read and analysed natively | Not built for it |
| PDF and DOCX upload | Yes | Varies |
| False-positive rate | Published, held-out | Usually undisclosed |
| Calibrated for | Scientific writing | General web text |
| Price | Free | Freemium or paid |
A general comparison; all three meter usage, which is the point. We would rather you check the numbers below than trust a table.
| 94.5% | correctly distinguishes machine from human text, shown one of each (held-out) |
| 99.6% | genuine human papers correctly cleared, not falsely flagged |
| 0.4% | false-positive rate on real papers, the number we optimise |
Model re-optimised on GPT-5 Sol and Claude Fable output, and updated as new models ship. English only for now.