How Certification Works
True Human Mark documents how a written work was created. We review evidence of the creative process and issue one of three classifications. We do not use AI detectors to decide who wrote something, and we do not claim absolute proof of authorship.
1. The three classifications
Human Made (green): The evidence shows the work was written by people, with no material contribution from generative AI. Ordinary spelling and grammar tools are fine.
AI Assisted (blue): A person led the creative work and made the key decisions, while generative AI provided disclosed supporting help such as brainstorming, outlining, or draft suggestions that the person reviewed and revised.
AI Made (orange): Generative AI produced or substantially shaped the work, and that use was fully disclosed and supported by evidence. This is a reviewed disclosure label, not a penalty.
Disclosing AI use never disqualifies a work. It determines which classification fits. Hiding material AI use can lead to denial or revocation. Every submission receives a full evidence review.
There is no shortcut tier and no self declaration badge.
2. What we review
What we review, strongest first.
- 1Drafts and version history. Native version history and intermediate drafts showing the work developing over time. This carries the most weight.
- 2Revision records. Tracked changes, editorial correspondence, and marked drafts showing decisions being made.
- 3Process artifacts. Handwritten notes and outlines, notebook pages, marked up printouts, and red ink edits. We encourage these. Their value is in the progression across several pieces, not any single page.
- 4Research notes and sources. The materials the work was built from.
- 5Workflow and editorial records. Assignments, editorial calendars, and correspondence connecting the applicant to the work.
- 6Photo evidence. Photos of the work in progress. Stronger than a declaration, weaker than version history.
- 7Your signed declaration. Required, and never enough on its own.
What carries more weight:
Evidence created as you worked, not assembled afterward. Several types that agree with each other. A trail that shows the work changing over time and matches what your declaration says.
Redactions are welcome.
Black out source names, locations, or private details and include a short note saying what category you removed and why. We keep the redacted copy. Protecting a source does not weaken your submission, and the judgment shown in a careful redaction can strengthen it.
What carries less weight:
A single polished artifact with nothing behind it, records created after the fact to match a finished piece, and a declaration standing alone.
Applicants must disclose everyone who made a material contribution and every generative AI tool that materially shaped the work. Saying “I did not personally use AI” is not enough if an editor, collaborator, or contractor might have.
3. How a review works
Intake: We confirm identity, the work and edition, and that the submission is complete.
Completeness check: We review the evidence package and ask for anything missing.
Evidence review: A reviewer scores the submission against a published rubric and compares the evidence with the requested classification.
Second review: During the pilot, the founder conducts every review. Close calls and integrity concerns receive a separate, documented second pass, and we are working toward adding an outside reviewer.
Decision: We issue a written approval, a request for clarification, or a denial with reasons.
Badge and public record: Approvals receive a badge linked to a public certification record.
4. How decisions are made
Each submission is scored on a 100-point rubric covering identity and edition match, development over time, draft and version evidence, disclosure consistency, supporting records, and attestation integrity. Approval requires at least 75 points with no unresolved integrity concern.
When the evidence could support more than one classification, we assign the more AI-involved one. This boundary rule protects the meaning of the Human Made badge.
Automated AI detection results may be used only as a limited investigative signal. They never decide a classification on their own.
Commercial considerations such as fee size or subscriber status never influence a decision.
5. If you disagree
Every applicant may appeal a denial, a classification, a suspension, or a revocation. Appeals are free. During the pilot, appeals receive a documented second review by the founder, with outside consultation on close cases where feasible. New evidence may be considered.
6. The badge
A badge applies to one specific work and edition, not to a creator in general. It carries a unique certification ID and links to a public certification record. During the pilot, badges expire 24 months after issuance and may be renewed for an unchanged edition. If a work is substantively revised, the new edition needs its own review. Misused, expired, suspended, or revoked badges must be removed.
7. What certification is not
Certification is our professional judgment, based on the evidence available at the time of review, that the evidence supports the classification. It is not a guarantee of copyright ownership, originality, factual accuracy, literary quality, legality, or the absence of all undisclosed AI use.
8. Pilot program
The pilot covers short written works under 5,000 words, such as articles, essays, reviews, newsletter issues, and blog posts. Have a longer work? Email [email protected] to ask about certifying it. Submitted evidence is held in confidence. Certified works appear in the public library only if you opt in.