What a Humanums certificate proves — and what it doesn't

Every trust product eventually faces the same question: what exactly does your seal mean? For passports, the answer is identity. For organic labels, it is farming practice. For a Humanums certificate, the answer is precise, and it is narrower than some people assume — which is exactly why it holds up.

This post draws the boundary clearly. If you are deciding whether to trust a certificate someone showed you, or deciding whether to certify your own work, this is the honest version of what the certificate covers.

What the certificate proves

1. The text was typed, not pasted or generated in place

The core of every certificate is a behavioral record captured while the writing happened: keystroke cadence, pause timing, revision activity, paste events, and session distribution. AI-generated text arrives in a document either by pasting or by programmatic insertion. Both leave an unmistakable signature — or rather, the absence of a human one. A certified document with a high score is a document where the words accumulated keystroke by keystroke, at human speed, with human rhythm.

2. The writing had human texture

Humans do not type like machines replaying a script. We pause to think — long pauses at sentence boundaries, short ones mid-phrase. We go back. We delete a clause, rewrite it, move on, and return to it twenty minutes later. The six signals we score are six independent views of that texture. A transcription session, where someone retypes prepared text, looks measurably different: steady cadence, almost no revision, pauses in the wrong places. The score reflects that difference.

3. The content has not changed since certification

At certification time we hash the document content and sign the hash, the score, and the session statistics with a server-held secret. The verification page shows what was certified. If a single character of the certified text changes, the hash no longer matches. That makes a certificate tamper-evident: it is bound to one specific version of one specific document.

4. The process happened inside a monitored session

The certificate covers writing that happened in the Humanums editor or through our Chrome extension, where the telemetry capture was active. That scope is stated on every verification page. We certify what we observed — nothing more.

What the certificate does not prove

It does not prove identity

A certificate says a human typed the document under a given account. It does not biometrically establish which human. If a writer hands their keyboard to someone else, the behavioral record is still real — it just belongs to a different person. For most use cases this does not matter, because the question being asked is “did a human write this?” rather than “did this specific human write this?” Where identity matters, the certificate works alongside the account's public profile, not as a substitute for it.

It does not prove that no AI was consulted

A writer can read an AI-generated outline on a second screen, think about it, and then write their own prose. The typing is genuinely human; the influence is invisible to us, just as a conversation with an editor or a stack of research notes would be. We consider this the correct boundary. Writing has always been informed by outside input. What the certificate rules out is the failure mode that actually erodes trust: text generated by a machine and presented as typed work.

The hardest adversarial case — manually retyping AI output word for word — is the one our scoring is specifically designed to catch. Transcription produces flat cadence, near-zero structural revision, and pause patterns uncorrelated with sentence structure. It scores poorly. It is also self-limiting: retyping a thousand words of AI text convincingly, with fake hesitations and fake revisions, approaches the effort of just writing.

It does not prove originality or quality

A human can type plagiarized ideas in their own words, and a human can type something mediocre. The certificate is silent on both. Plagiarism detection and editorial judgment are different problems with different tools, and pretending one seal covers everything is how trust products lose credibility.

Why a narrow claim is a strong claim

AI detectors fail because they make a broad claim — “this text is AI” — from weak evidence. The claim sounds decisive and collapses under a single false positive. We made the opposite trade: a narrow claim from strong evidence. “A human typed this, with human writing behavior, and it has not been altered since” is a statement we can back with a signed behavioral record that anyone can inspect on a public page.

When you see a Humanums badge on an article, that is what it means. Not that the author is a saint, not that the prose is brilliant — that a human did the writing. In a feed full of generated text, that narrow fact turns out to be the one readers most want to know.

If you want the full technical detail behind the six signals, read how Humanums works. If you want to see a certificate in the wild, this post — like every post on this blog — carries its own badge below.