Your clients’ chatbots now have to say they’re AI. See which ones do.

Since 2 August 2026 the EU AI Act requires that people are told when they talk to an AI system. The law puts that on the provider, which for an off-the-shelf widget is the chat vendor; a vendor’s terms can hand the task to its customers, and the label lives in the widget settings your team configured. Scan any client homepage free; at launch, the agency plan re-reads all of them every month.

Observed facts: found or not found. Never a compliance verdict; findings are inputs for your client’s own legal review.

Article 50 signal scanready

3 free scans per day.

The law · Regulation (EU) 2024/1689, Art. 50(1)EUR-Lex

“Providers shall ensure that AI systems intended to interact directly with natural persons are designed and developed in such a way that the natural persons concerned are informed that they are interacting with an AI system, unless this is obvious from the point of view of a natural person who is reasonably well-informed, observant and circumspect, taking into account the circumstances and the context of use.”

A vendor’s help centre · Intercom, citing its AI terms §5Source

“…our customers are responsible for ensuring that their end users are informed that they are interacting with an AI system.”
The law names the provider. The vendor’s terms name its customer. The switch sits in the widget your team set up.

In Lithuania, 22 of 174 homepages had a chat widget. One showed AI disclosure wording.

This is the readout the agency plan will keep for your client list. The rows below are real, from our own corpus run; only the domains are withheld.

Client sites · readout8 example rows · July 2026
SiteWidget foundPlatformDisclosure wording
.ltVoiceflowAI bot builderDisclosure:Not found
.ltChatbaseAI bot builderDisclosure:Not found
.ltIntercomChat / helpdeskDisclosure:Not found
.ltTidioChat / helpdeskDisclosure:Not found
.deGorgiasChat / helpdeskDisclosure:Not found
.itZendeskChat / helpdeskDisclosure:Not found
.ltCrispChat / helpdeskDisclosure:Found
.ltTawk.toChat / helpdeskDisclosure:Not found

+16 more widget sites in this run21 of 22 Lithuanian widget sites without disclosure wording in the page source

Our corpus run, July 2026: 248 mostly-SME homepages readable out of 302, 174 of them Lithuanian. The figures above are for Lithuania only: the other 74 homepages (Germany 21, Poland 17, Italy 8 and others) are too few per country for a rate, so their sites appear only as rows. “Not found” means not in the static page source: a widget can still disclose inside the chat window, which only the rendered audit scan sees. Several vendors here are live-chat tools that may be staffed by people, which Article 50 does not cover; the platform column marks the tools built as AI bots.

What the scan reads in a client’s page source

Three signal families. Every finding in the paid audit comes with the exact fragment it was matched on, so you can show your client the exact line we matched.

  1. Chat & support widgets

    25 vendor fingerprints (Intercom, Crisp, Tidio, Smartsupp, Botpress and others) matched in the page source.

    free scan · static

  2. AI disclosure wording

    Phrases that tell a visitor they are talking to a bot. Matched in 7 languages.

    free scan · static

  3. AI-image credentials

    C2PA / IPTC provenance markers read from image bytes in the deep audit scan. CDNs often strip them, so “not found” never means “not AI”.

    audit · rendered

Known limits, stated up front. The free scan is static: it does not execute scripts or click consent banners, so widgets loaded after consent stay invisible to it; the audit's rendered scan covers those. We respect robots.txt and never bypass access controls. To tell blocking from downtime we send up to two diagnostic requests with other user agents (a bot-style one, and a browser-style one only if the site refuses our scanner); both carry our name and contact link, and only their status codes are read, so we never read page content through them. And no scan can read facts that live outside the page: who operates the bot, what your images depict, whether a human reviewed a text.

One plan for every client site, open for pre-order

Launch pricing: what you see is what early customers pay. All prices exclude VAT and are for business customers. The free scan works today; paid plans open at launch, and reserving one costs nothing.

Agency planAt launch

€149–249/ month · by number of sites

  • Every client homepage in one list, re-read monthly
  • Change alerts: a new widget, a disclosure that disappeared
  • A client-ready readout per site, with the matched fragment
  • Free scan

    €0

    One homepage, on-screen readout, 3 per day. No account needed, just your email. Available now

  • Full audit, one site

    €99one-time

    Deep rendered scan (after-consent widgets), AI-image credentials, page-by-page findings with evidence and a fix list. Ships after launch; pre-order customers are notified first. Pre-order

    one-time launch price, excl. VAT · for business customers

  • Monitoring, one site

    €29–49/ month

    Monthly re-scan with change alerts for a single site. At launch

Reserve a place

Tell us what you need. We confirm by email and invoice only when it ships; you will know the date before anything is billed.

Plan
Client sites with a chat widget

You pay nothing today.

Straight answers

Who is responsible for the chatbot disclosure: the vendor, our client or us?

Article 50(1) puts it on the provider of the AI system. For a widget used as shipped, that is the vendor; whoever builds a bot, or has one built and runs it under its own name, becomes a provider. A vendor’s terms can hand the task to its customers. Deep fakes and AI-written text published on matters of public interest are a separate duty for whoever publishes them (Article 50(4)). Which applies to a given setup is a judgement for your client and their advisers.

Can we scan a client’s site?

The scan reads a public homepage the way a crawler does and respects robots.txt. It does not log in, click consent banners or bypass access controls.

When do these rules start to apply?

Article 50 transparency obligations apply from 2 August 2026. Breaching them can cost up to €15 million or 3 % of worldwide annual turnover, whichever is higher (Article 99(4)); for SMEs the lower of the two amounts applies (Article 99(6)).

Is this legal advice?

No. The scan reports what is technically visible on a page: found or not found. Whether a specific setup meets Article 50 is a legal judgement for your client and their advisers.

Does a clean readout mean the site passes?

No scan can tell you that. Some obligations depend on facts no crawler can see: who operates the bot, what the images depict, whether a human reviewed a published text. We show the signals; your client and their advisers decide.

What does the free scan see, and what does it miss?

It reads the homepage’s static source: widget fingerprints and disclosure wording. It cannot execute scripts or click consent banners, so after-consent widgets stay invisible to it; the audit's deep rendered scan covers those, plus AI-image credentials.

Who is behind Klaroscan?

Klaroscan is built and operated by Švaros Banga, MB (trading as Aivikta), Lithuania (EU), on the same engine that powers our AI-visibility audits. EU-based, GDPR-first: scans store a hashed IP, never the raw address.