About Could This Be True?

Why this exists

Every major AI-detection product outputs a single number — “87% AI” — and asks you to trust it. We don't.

Real forensic work shows its evidence. A pathologist doesn't hand you a verdict; they walk you through the slides. A photo forensics expert doesn't say “edited” — they show you the splice, the compression mismatch, the noise inconsistency. That's the model we built around: every signal is visualized, every decision is auditable.

What we run

• Text: burstiness, lexical tics, n-gram repetition, sentence-start variety, punctuation patterns.
• Image: Error Level Analysis, FFT magnitude spectrum, RGB channel decomposition, noise residual (PRNU proxy), C2PA signature verification.
• Audio: spectrogram analysis, spectral flatness, harmonic stability, spectral-centroid dynamics.
• Video: per-frame noise consistency, lighting plausibility, and motion plausibility across sampled frames.

What we don't do

• Sell the bypass. We will never sell a tool that helps content evade detection. The conflict of interest would destroy the entire product.
• Pretend to be a polygraph. Detection is evidence, not proof. False positives have caused real harm — students wrongly accused, photographers wrongly flagged. Treat every verdict as the start of an investigation, not the end.
• Upload your data unnecessarily. Most checks run in your browser. Nothing leaves your machine unless you explicitly choose a server-side option.

What's next

Text, image, audio, and video detection all ship today, each visualizing its signals and reading C2PA provenance where present. On the roadmap:

• A public API and batch processing for fraud / moderation / journalism workflows.
• Deeper video forensics — face-landmark stability and lip-sync drift.

Get in touch

We're actively looking for design partners in fraud investigation, journalism, insurance, and content moderation. If forensic explainability matters to your workflow, we want to talk.