For agencies that put AI chat on client sites
Article 50(1) of the EU AI Act has applied since 2 August 2026. A person talking to an AI system has to be told so, at the latest at the first interaction. If you install or tune AI agents and chat widgets for clients, sooner or later a client asks whether that notice is actually showing on their site. DisclosureProof opens the site the way a first-time visitor does and reads the chat's first message. It keeps a dated, hashed record of what it saw.
What the Business plan covers
- Client sites
- Up to ten verified domains, re-scanned daily or weekly.
- Depth
- Up to 500 pages per scan on a verified site.
- Alerts
- An email with the before and after when a disclosure changes between two scans, for example when a widget update drops the AI label from the greeting.
- Deliverable
- Evidence Pack PDFs are included. The cover carries your agency's name, with a provenance line in the footer saying the record was prepared using DisclosureProof. See a specimen with an agency's name on the cover (PDF): fictional scan data and example seal values, so it cannot be verified as signed evidence.
- Records
- Kept for as long as the subscription runs.
- Price
- €200 a month, or €2,000 a year. Across ten client sites that is €20 a month per site, a line item that fits inside a maintenance retainer. A first subscription starts with a 7-day trial.
Set it up in the same visit as the widget
-
Install or tune the chat as usual
Nothing about your build changes.
-
Add the verification tag in the same edit
One meta tag in the client's page head, or a DNS TXT record if you manage their DNS. It shows the domain is under your management. Monitoring and 500-page scans need it, and so does the direct question described below. If you already edit the page head to install the widget, this needs no further access.
Meta tag or DNS TXT · once per domain -
Pick a cadence
Daily or weekly. You can also re-scan from the dashboard whenever you ship a change.
The direct question, and what it costs your client
On a verified site the scanner can type one question into the chat: whether the visitor is talking to a person or an AI. AI agents are often billed per resolution, so a question asked on every daily run could add to a client's bill and land in their support metrics. Monitors therefore skip the question whenever the first message already says the assistant is an AI. It is asked only when the greeting leaves the point open, which is the case the question exists to settle.
Before there is a contract
The free scan takes any public homepage, needs no account, and finishes within a 90-second capture limit. Run it on a prospect's site before a pitch to see whether there is a chat widget and what its first message says. Where one client needs a record and not a subscription, an Evidence Pack (€79 once) seals that scan as a PDF and adds a separate crawl of up to 100 pages, with no verification step.
What the record says
Findings read detected, not detected or could not verify, graded against a dated rule pack for the four Article 50 duties. Chat-panel screenshots are attached when the panel opens. Some widgets cannot be opened by an automated visitor, and when that happens the report says which widget it saw and why it could not read it, instead of guessing. The word “compliant” never appears in a report; the reasons are here. The hashes in each manifest can be re-checked by anyone at /verify/.
DisclosureProof provides automated technical checks for informational purposes only. It is not legal advice and does not certify compliance. Firms advising clients on Article 50 may prefer the counsel workflow.