Protecting honest writing from false positives
Scan your text sentence-by-sentence against the same neural patterns detectors use. Fix only what needs fixing.
Two modes built on the same engine
Estimate for free, verify with Pro
Free Tier
Estimate
Paste up to 500 words and get an instant heuristic score. No account required.
Pro · $18/mo
Verify
Full neural model scoring on every sentence. 2,000 words, unlimited suggestions, scan history.
Scanner
See what detectors see before you submit
Sentence-level scanning powered by a fine-tuned neural model that catches the same token-level patterns commercial detectors use.
Every sentence scored across five signals
Detectors don't just read your text. They measure burstiness, vocabulary density, perplexity, structural patterns, and coherence. GPTypo runs the same checks so you can see which sentences trip which signals before you submit.
Neural model
Fine-tuned on detector patterns — catches what surface-level tools miss.
98.2% accuracy
Independently verified on the RAID benchmark (ACL 2024).
Signal grid
Five signal categories per sentence so you see why text is flagged.
Rescan & delta
See score changes instantly after each edit. Repeat until clean.
Scan history
Track every scan and see your progress over time on Pro.
Editing
Fix only what needs fixing, keep your voice
Human-in-the-loop editing — not a one-click humanizer. You control every change.
AI-powered rewrite suggestions you control
Click any flagged sentence for alternatives. Accept, edit manually, or skip. Then rescan to verify your new score.
Flagged sentence
Suggestions
Claude-powered
Suggestions preserve your writing style and tone.
Score preview
See the projected score before accepting a rewrite.
Manual editing
Prefer to rewrite yourself? Edit directly and rescan.
Not a bypass
Built for honest writers — never a one-click humanizer.
Detection Accuracy
We publish our numbers so you don't have to trust our marketing
Every number comes from automated testing against a labeled corpus of known human and AI-generated text. Updated regularly.
AI text correctly flagged
99.9%
2863 of 2865 samples
Human text correctly passed
97.1%
1231 of 1268 samples
False positive rate
2.9%
37 human samples flagged
Human avg score
97.4
0–100 scale
| Source | Samples | Pro (neural) |
|---|---|---|
| Casual online writing | 447 | 96.9% |
| Academic papers | 226 | 98.7% |
| Non-native English | 196 | 94.4% |
| Clinical / medical | 178 | 97.2% |
| Creative fiction | 160 | 98.1% |
| Articles / essays | 61 | 98.4% |
The neural model struggles with formal academic and clinical text, where human prose shares structural features with AI output. We re-run this benchmark periodically and retrain the model as needed. Read the full methodology →
Latest updates
