Verified Human Thoughts.
AI detectors guess from the finished text. Ezniek records where text came from: an accountable authorship declaration, cryptographic fingerprints, timestamps and—when available—documentary lineage reaching back to earlier drafts or publications.
Check provenance before you accuse.
Choose a text-based PDF. It is processed locally on this device: the manuscript itself is not uploaded to Ezniek. Only cryptographic hashes, fragment fingerprints and minimal metadata are sent to the registry. No match is never treated as evidence of AI authorship.
Local processing: the selected PDF is not uploaded. Only fingerprints are queried against the registry.
Check whether an internet page has an independently archived capture from before 2020.
Write anywhere. Register the finished work.
Use Word, LibreOffice, Scrivener or any editor you like. Export a PDF. Your browser fingerprints the exact file, normalized text and overlapping fragments on your own device. The manuscript itself is not sent to Ezniek; only the resulting fingerprints and your accountable authorship declaration are registered.
Large manuscripts are supported. Processing time depends on PDF size and your device; no server upload-size limit applies because the PDF bytes never leave your browser.
Published before 2020? That matters.
Archived web publication
A Wayback or equivalent independent capture before 2020 establishes that a web resource existed by that date. A verified textual match can then support human provenance.
Print & scholarly publication
ISBN, DOI, publisher records, library catalogues and dated editions can corroborate publication chronology. Crossref exposes open bibliographic metadata and publication dates for scholarly works.
Fingerprint corpus
Ezniek can index fingerprints—not necessarily full copyrighted text—from pre-2020 corpora. A later submission can match a prior publication without republishing the underlying work.
Example output: Registry: n/a · Human provenance: very likely · Basis: substantial textual match with independently dated pre-2020 publication. This supports the matched material's chronology; it does not establish the identity of a particular author unless authorship metadata is independently corroborated.
A public-good registry with a sustainable certification model.
One text · lifetime
For students, writers, applicants, freelancers and anyone who wants one durable registration for one specific text.
Planned after pilot. Registration lifetime; optional paid upgrades for identity/provenance review.
Publishers
Unlimited text registrations/verification for one publishing organisation, plus publisher dashboard and machine-readable records.
Ezniek Mark
A recognisable provenance mark, comparable in function—not endorsement scope—to a trusted quality seal: clearly defined criteria, public verification and revocation.
The mark certifies the stated Ezniek provenance status, not literary quality or truth.
Different evidence answers different questions.
AI watermark
Can provide positive evidence that supported content was generated by a participating AI system.
AI detector
Infers likely origin from statistical patterns. A score is not documentary provenance.
Ezniek
Records positive human-authorship claims and documentary chronology. It can coexist with C2PA, SynthID and detector scores.
“If provenance exists, do not replace evidence with a guess.”