Disclosure text

A free, public generator needs no login: pick a duty, a delivery format, and a language, and copy the result. It writes nothing and remembers nothing.

A system’s own Disclosure page only offers the duties that actually apply to that system, worked out from its intake answers by the same applicability engine that computes its coverage score — not a generic dropdown. Using it also keeps a dated record of exactly what text was generated, for that exact version of your system.

Pressing Generate & record on a system’s Disclosure page writes one evidence item: a dated, chained record that this exact sentence was produced, for this system, at this version. It carries no link to any requirement your coverage score is built from, so generating text can never move that score — by construction, not by convention.

Once you have actually published the text somewhere real, a second action — I have deployed this — asks where, and records an ordinary statement against the requirement it satisfies, carrying a fingerprint of the exact text you attested to. That is the one that moves your score. Skipping it is the correct choice if you have not actually deployed the text yet — a coverage score built from generated-but-unpublished text would be recording a hope, not a fact.

Every duty is written in all 24 official EU languages. Each language carries its own review status: today, every one of them — including English — is machine-assisted and has not been checked by a lawyer. You will see that caveat wherever it is useful to you: on the generator itself, here, and as a code comment in the HTML/React/Vue output. It is never inside the plain-text sentence itself or the visible part of any other format, because that sentence is what you paste in front of your own users — adding a caveat there would put words in front of the people Article 50 exists to protect, not in front of you.

A language’s caveat disappears the moment that language is reviewed by counsel — a change to the underlying data, not to the software.

Article 50(2) — marking AI-generated content so machines, not just people, can detect it — is not satisfied by a caption alone. A visible label is not machine-readable marking; the regulation asks for the output itself to be marked (for example a C2PA / Content Credentials manifest, an embedded watermark, or standardised metadata), with the generated wording as the human-readable declaration that goes alongside it. The generator says this plainly wherever that duty is offered, rather than implying a caption is enough.

Generating disclosure text costs nothing, on the free plan or any paid one, and needs no login on the public generator. What stays behind the paid tiers is unchanged: a signed, downloadable evidence pack.

If you change your system in a way that opens a new version — a new capability, a new risk classification, anything that matters enough to record — any disclosure text you generated for the earlier version may no longer be accurate. Clause50 raises an alert when that happens, prompting you to regenerate and redeploy rather than leaving an out-of-date banner live. See Alerts.