Compared · Human authorship evidence
What each approach can actually establish
AI detectors, self-declared authorship marks, content credentials and HumanMark answer different questions.
The relevant distinction is not whether an approach uses the word human. It is what evidence the approach creates, what that evidence establishes, and whether another party can verify it independently.
HumanMark does not determine who mentally composed a sentence. It records observable production facts and binds those observations to a verifiable record.
The four approaches
| Approach | What it establishes | Limitation |
|---|---|---|
| AI detectors | A statistical estimate that finished content resembles patterns associated with model-generated material. | A classifier evaluates the finished content. It does not create a contemporaneous record of how the work was produced. |
| Self-declared authorship marks | An assertion that a person or organization describes the work as human-created. | The strength of the evidence depends on the process supporting the assertion. The label alone does not establish how the work was produced. The claim cannot be independently verified by the recipient. |
| Content credentials | Signed provenance information associated with a digital asset, including supported capture and editing history. | Strong provenance evidence does not necessarily establish the nature or extent of human participation in producing text or other work. |
| HumanMark | Observable production facts recorded during creation and cryptographically bound to the resulting record. | HumanMark establishes what was observed. It does not infer mental authorship, determine who composed an individual sentence, or make claims about activity it did not observe. |
Professional work is rarely binary
A simple human / AI label does not adequately describe many modern production workflows.
The relevant facts may be materially different:
- 1Human production without generative AI. The work is produced directly by a person without generative AI contributing to the content.
- 2Human production with AI-assisted research. AI supports research, retrieval or analysis while the substantive work is produced by the person.
- 3AI-assisted drafting with substantial human revision. AI contributes an initial draft or material that is subsequently reviewed, rewritten and verified by a person.
- 4AI-generated material with human editing. AI produces substantial content and a person reviews or modifies it before use.
- 5AI-generated material used with limited modification. AI-generated content is incorporated substantially as produced.
Those are materially different production histories. A binary label cannot represent them with sufficient precision.
What HumanMark records
HumanMark preserves observable evidence from the production process rather than inferring provenance from a finished document.
Depending on the workflow, the record can include observable production facts such as:
- Typed inputInput entered directly during the recorded session.
- Pasted inputInput introduced into the recorded environment through paste operations.
- Record associationThe production evidence is associated with the corresponding artifact or record.
- Time and sequenceRelevant observations are recorded as part of the production history.
- Cryptographic integrityThe resulting evidence record can be checked for subsequent alteration.
HumanMark reports the observations. It does not convert those observations into a broader claim that the evidence cannot support.
What HumanMark does not claim
No AI detection
HumanMark does not examine finished text and estimate whether an AI system produced it.
No sentence-level authorship determination
HumanMark does not claim to determine who mentally composed an individual sentence.
No inference beyond the observed record
Activity outside the recorded production environment is outside the evidence HumanMark can establish.
No legal conclusion
The record provides evidence. It does not determine admissibility, legal authorship, compliance or the legal significance of the evidence in a particular matter.
Why observation and detection are different
Detection works backward from the finished artifact.
A classifier examines completed content and produces an estimate about how that content may have been created.
HumanMark records forward from the production process.
It preserves observable facts while the work is being produced and binds those facts to a verifiable record.
The distinction is fundamental:
Detection produces an inference. HumanMark produces a record of observed production facts.
HumanMark and content credentials answer different questions
These approaches can be complementary.
Content credentials can provide strong provenance information about an asset, including supported capture and editing history.
HumanMark focuses on observable human-production activity associated with the work.
One addresses provenance of the asset. The other records observable production facts.
Neither should be described as proving more than its evidence establishes.
Evidence should state exactly what it supports
A useful evidence system should make its limits visible.
HumanMark therefore distinguishes between:
- what was observed
- what was cryptographically bound to the record
- what can be independently verified
- what the evidence does not establish
That distinction is part of the evidence model, not a disclaimer added afterward.
Content remains separate from the evidence record
HumanMark records production facts without the underlying work becoming part of the evidence record.
The resulting record can therefore establish relevant production observations while maintaining separation from the substantive content.
Independently verifiable
A recipient verifies the integrity of a HumanMark record without relying on any statement from Regulayer.
Verification does not require the recipient to accept an unsupported authorship claim.
The evidence states what was recorded and allows the recipient to verify that record independently.
Create a verifiable production record
HumanMark is included in the Regulayer attorney workflow.
Seven days free, then $349 per month.
Create the production record while the work is being made, preserve the evidence with the resulting artifact, and provide an independently verifiable record when the production history matters.
Start the free weekRegulayer for enterprise →
See also: The Receipt · Verify free · HumanMark · AI rules by jurisdiction
Common questions
Can HumanMark prove that a human wrote a document?
No. HumanMark does not claim to determine mental authorship.
It records observable production facts associated with the work and preserves those observations in a verifiable record.
Does HumanMark detect AI-generated text?
No.
HumanMark does not classify completed text as human-generated or AI-generated. It records observable production activity rather than making a statistical inference from the finished content.
What is the difference between detection and attestation?
Detection examines completed material and estimates how it may have been produced.
Attestation preserves evidence about facts recorded during the relevant process and binds those facts to a verifiable record.
Why record typed and pasted input?
Typed and pasted input are observable production facts.
Recording that distinction provides more specific evidence about the production process without converting the observation into an unsupported authorship conclusion.
What happens if work was created outside HumanMark?
HumanMark can establish only what was observed within the recorded workflow. It does not make claims about activity outside that record.
Is HumanMark a compliance certification?
No.
HumanMark creates evidence. The legal, regulatory or contractual significance of that evidence depends on the applicable requirements and the context in which it is used.
Information, not legal advice.