Deterministic measurements
Surface patterns, compression, structure, and frequency—not black-box scores.
Free public beta · No account needed
Check writing, images, websites, audio, and video. Get an explanation of what stands out, what could explain it, and what to check next.
Analyze privately
Paste or open a document, then analyze. You can optionally select its type for a more specific comparison.
Paste it below or choose one local file. A selected file is extracted in browser memory for your review before analysis.
Your evidence report
Add writing on the left, choose its document type, and select Analyze privately.
Patterns are clues, not proof of AI use. CouldItBe does not provide a verified authorship verdict or AI percentage. How to read a result →
Surface patterns, compression, structure, and frequency—not black-box scores.
Every finding keeps supporting observations beside contradictions and counterexamples.
Provenance checks, missing evidence, and limitations are never hidden behind a verdict.
Five ways to inspect
Choose a format to begin. Each report explains what was measured, what stands out, and what the evidence can support.
All five analyzers are operational. Maturity labels describe research depth—not whether a tool works. Research maturity requires independent-source, counterexample, frozen-candidate, untouched prospective, and independent-review checks.
Paste, open, or locally scan writing. The bounded long-form pattern rule has source-grouped research and counterexample testing; it still cannot determine authorship.
Analyze textMeasure compression, repetition, texture, color, frequency, edges, metadata, and Content Credentials locally. Development and transformation testing is underway; this is not an origin detector.
Inspect an imageReview one public page’s visible-text patterns, declared metadata, and structure through bounded retrieval. Retrieval security is extensively tested; the discarded page body is never treated as authorship proof.
Inspect a URLInspect WAV, MP3, and M4A level, dynamics, silence, clipping, frequency, and container measurements locally. Research still needs broader independent human, generated, mixed, and transformed audio.
Inspect audioInspect MP4 and WebM frame texture, luminance, clipping, repetition, transitions, and container evidence locally. Secure validation is operationally mature, but origin-relevant research remains early.
Inspect videoHelp test CouldItBe
Run material you are authorized to use. Review the result. Optionally contribute only derived measurements and an unverified source label. Analyzer source text or media is never included.
Write a short assigned paper in the browser and capture privacy-minimized drafting checkpoints.
Participate in the writing studyContribute an authorized human, AI-generated, or mixed-production recording for private research review.
Participate in the audio studyAdults can check an authorized MP4 or WebM locally while private volunteer intake remains closed. Videos will never be published or publicly playable.
Review the Video studyStudy records are voluntary and privately reviewed. Each study page explains its current submission status. Research review states are evidence, not authorship or origin verdicts.
Source-grouped records that satisfy a frozen study contract.
Known fixtures used to test mechanics, privacy, and failure handling.
An optional expectation that remains unverified and is never treated as ground truth.
Content-free reports that improve reliability, access, and clarity.