Is your Intercom chatbot EU AI Act ready?

Intercom and Article 50, briefly

For most companies Intercom started life as live chat with human agents and picked up an AI agent along the way: Fin. That distinction is exactly what makes Article 50 interesting here. The Messenger itself is not an AI system, but the moment Fin or a custom bot gives the first answer your visitor is talking to an AI, and they are meant to know that before they start typing.

What our scans found on Intercom sites

We have not yet scanned enough sites where Intercom was the first widget we detected to say anything meaningful here. As soon as the sample is large enough it will appear, counts included. We would rather publish nothing than a percentage based on two sites.

How we detect Intercom

Our scanner looks for these signals on your public page. Two independent signals are required for a match, so a blog post that merely names the vendor does not count.

Script hosts

widget.intercom.iojs.intercomcdn.comapi-iam.intercom.io

JS globals

window.Intercomwindow.intercomSettings

DOM selectors

#intercom-container.intercom-lightweight-app#intercom-frame

Where the AI notice goes in Intercom

Messenger, Content or Settings, Greeting. Running Fin or another AI agent? Put it in that agent's opening message too, because that is what visitors actually read first.

1. 1

   Open Intercom and go to Messenger in the left column.

2. 2

   Pick the Messenger profile that runs on your website (web, not mobile).

3. 3

   Go to Content or Settings and find the greeting.

4. 4

   Running Fin? Open Fin, the Behaviour tab, and edit the opening message too.

5. 5

   Publish, then test in an incognito window with an empty profile.

Intercom documentation: www.intercom.com

What goes wrong most often with Intercom

Three mistakes we run into on sites that did set the notice. Each one costs points in the scan, and each one takes minutes to fix.

• The notice is in the greeting, but Fin overwrites it

  With Fin enabled the conversation often opens with Fin's message rather than the Messenger greeting. We regularly see a tidy AI sentence in the settings while the visitor reads something else entirely. Set the sentence in both places.

• A human name and photo with no mention of AI

  An avatar reading "Sanne from support" above an answer written by Fin reads as a person. Our scanner flags that as a human persona. Add "AI assistant" to it, either in the agent name or in the first line.

• Only the English text filled in

  Intercom supports a separate greeting per language. If you serve Dutch visitors, the notice belongs in Dutch too. We often find a polished English sentence on an otherwise fully Dutch site.

Checking that it works in Intercom

Test in an incognito window with a clean profile: Intercom remembers visitors, so in your own browser you usually get a continued conversation instead of the greeting. Open the Messenger, read the first message that appears and hold it against what you set in the settings. Running Fin, click through to the first generated answer and check that the notice was already there before it. Repeat on mobile, because Intercom shows a more compact greeting there in which a long sentence gets cut off.

After that you can scan again. Our scan looks at the same public page a visitor does, so what you see in an incognito window is roughly what we see.

Frequently asked questions

Running another widget as well?

• Chatwoot and Article 50
• Freshchat and Article 50
• Landbot and Article 50