Free document tools•POST /v1/detect-ai-text
Detect AI-written text in submissions
Supported Formats: Prose Submissions, Cover Letters, Manuscripts, Reviews
Core Verification Mechanisms: Linguistic Tell Breakdown · Non-Prose Abstention · Probability Confidence
Need copy-paste code with sandbox testing?Open Developer Playbook (cURL / Python / MCP) →
The problem
Cover letters, applications, and reviews — when human voice is the signal, undisclosed AI generation distorts the evaluation.
Most detectors return opaque percentages that false-alarm heavily on non-native or templated prose.
How Stipple helps
Stipple highlights the exact linguistic tells in the text, reports probability bounds, and explicitly abstains on tables, code, and structured forms where style detection is mathematically unreliable.
The call
View Full Recipe →bash
curl -X POST https://www.stipple.sh/v1/detect-ai-text \
-H "Content-Type: application/json" \
-d '{"text": "The submitted essay text..."}'
# → { "probability": 0.92, "lean": "ai", "tells": [...], "applicable": true }Evaluate writing submissions with transparent evidence rather than a black-box percentage score.
How the corpus was built, what was caught, and what the method does not prove.
Questions
- What is the difference between this and watermark detection?
- Style analysis estimates how model-like the prose reads. Watermark verification is key-based provenance testing that only the model’s provider can perform, and those APIs are not public yet. This tool does the first; the watermark guide explains the second.
- Does a low score prove a human wrote it?
- No. A low score means no AI-style signals fired in the prose that could be read. Paraphrasing lowers any style score, and most AI text carries no provider mark either — so absence of evidence is not evidence of absence, in both instruments.