How ProofLens works

ProofLens is a free AI detector for text, essays, PDFs, Word documents, images, audio and video. Every file type gets its own forensic examination — all of it on your device, nothing uploaded.

Detecting AI-written text

Style analysis measures sentence-length “burstiness” (humans mix short and long sentences; AI keeps a steady rhythm), the density of AI-typical wording, leftover chatbot Markdown, invisible zero-width characters, and human quirks like typos and slang. Optional deep analysis runs a small language model on your own device to measure how statistically predictable the text is — section by section, so a half-human, half-AI essay shows exactly which paragraphs look generated.

Checking PDFs and Word documents for AI

The metadata tells a story: which application produced the file, how many minutes of editing it records, how many revision-save IDs it accumulated. A document “written” in one shot with zero edit time reads very differently from one worked on across many saves. The extracted text then goes through the full text analysis with inline highlighting.

Spotting AI-generated images

ProofLens cryptographically verifies C2PA Content Credentials (DALL·E, Adobe Firefly and others sign their images — the strongest provenance evidence that exists), reads Stable Diffusion prompt chunks embedded in PNG files, checks camera EXIF data, and spots generator-default dimensions.

Audio and video

Provenance metadata only — encoder signatures, recording-device fields, Content Credentials — and the report clearly says the analysis is limited. A clean result does not mean a recording is real.

Scoring: evidence, not verdicts

Each signal points toward AI or human with a weight; together they produce a score from 0 to 100 and a band: Likely human-made · Inconclusive · Possibly AI-made · Likely AI-made. A single weak clue never moves the verdict on its own, and every signal is shown with its reasoning.