Three approaches. One proves authorship.

Process Evidence
vs. Watermarking vs. AI Detection

Watermarks prove someone signed the text. AI detectors guess what generated it. Only process evidence captures how it was written.

Feature-by-Feature Comparison

DimensionWritersLogicProcess EvidenceWatermarkinge.g., EncypherAI Detectione.g., GPTZero
Provenance
What it provesThe fundamental question each approach answers"I wrote this" — the author's own attested processSomeone signed it (ownership claim)Statistical guess (probability score)
When evidence is createdAt what point in the writing lifecycle proof is generatedDuring writingAfter writingAfter publication
Can AI-generated text pass?Whether someone can sign AI output and claim authorshipNoYesPartial
Robustness
Survives paraphrasingWhether proof remains intact when text is rewordedYesNoN/A
Adversary-resistantHow difficult it is for a motivated adversary to defeatLocked to your device hardwareInvisible characters (easily stripped)Model-dependent (20%+ error rate)
Tamper-evidentWhether altering evidence invalidates the proof chainYesPartialNo
Legal & Standards
Legal evidentiary weightUsefulness as evidence in legal or disciplinary proceedingsConcrete evidence of the writing processOwnership claim onlyProbabilistic opinion
Standards trackBacked by recognized standards bodiesIETF draft, plus 6 algorithms registered on the C2PA Soft Binding Algorithm ListC2PA Section A.7, no registered algorithmNone
Practical
Privacy preservingWhether actual content or keystrokes are recordedYesYesNo
Works offlineCan generate proof without internet connectionYesNoNo
Open sourceLicense and availability of source codeAGPL 3.0 open core + commercial licensingAGPL 3.0 (viral copyleft)Proprietary
Records how you writeCaptures your editing rhythm, revision patterns, and pause timingYesNoNo
Cost to sign and verify at scaleWhat publishing and checking provenance records costsFree up to 10,000/mo — runs on git, so the cap is abuse prevention, not costFree trial caps at 1,000 items/month, then $99+/moN/A

The fundamental difference

Watermarking proves someone claimed the text. AI detection guesses what generated it. Neither captures the act of writing itself.

We include invisible watermarking too — free, unmetered, no account required. It's a binding layer: cheap to run, which is why it's cheap to strip. We don't sell it as your only shield. Process evidence is the layer you reach for when stripping actually matters.

WritersProof records how you write: your rhythm, your pauses, how you revise and edit. This evidence is locked to your device hardware and sealed so that changing any part invalidates the whole record. Anyone can verify it for free.

Someone can strip a watermark with a few lines of code. To fake process evidence, they would need to sit down and type every word of the document with realistic human editing patterns while defeating hardware-level security. That is a fundamentally harder problem.

Six of the entries on the C2PA Soft Binding Algorithm List are ours

A soft binding is what finds a document’s C2PA manifest again after the manifest itself is gone — after the text is copied out of a file, reformatted, quoted in part, or reworded. Text needs a family of them, because no single technique survives every edit. WritersLogic wrote and registered six, more text soft bindings than any other organization on the list, and a manifest can carry several at once.

Every entry is public and independently checkable in the C2PA Soft Binding Algorithm List. The watermarking product compared above has no entry on it, and AI detectors have none. See how the six compose inside a single manifest.

Ready to prove authorship, not just claim it?

See how WritersProof captures process evidence for writers, publishers, and institutions.