A responsible-gambling checklist for outbound AI calls

Why do AI calls need their own responsible-gambling checks?

A call is a conversation, and conversations reveal things that deposit data doesn't. In Great Britain, remote operators must build customer interaction around three steps: identify, act and evaluate. The warning signs they must watch include contact the customer starts themselves, and a player saying "I want to stop" on a call is exactly that. The rules also require those signs to be flagged promptly for a person to deal with, and marketing and new bonus offers to stop where strong signs of harm appear.

An AI agent makes thousands of these conversations possible. That's the opportunity and the risk: every safeguard has to work at the same scale as the calling.

What should be checked before the first call?

• Marketing consent for phone calls is confirmed for that product. Consent for casino emails is not consent for sportsbook calls.
• Self-excluded players are removed using the right register for the market. In Great Britain, that means GAMSTOP.
• Players already showing strong signs of harm are removed from the list, because offers must not reach them.
• Calling hours are set as a hard limit, not a guideline.
• The script and the offer are approved by compliance, and by the regulator where required.
• One named person owns the harm-flag queue, with an agreed response time.

What must the agent do during the call?

• Tell the player at the start that they're speaking with a virtual assistant.
• Make only the offer that was approved. Never improvise a bonus.
• Listen for harm in the player's own words: losing too much, wanting to stop, not being able to afford it, distress.
• The moment harm appears, stop the offer, never return to it, and tell the player a person will follow up.
• Treat "stop calling me" as final, immediately.
• Never discuss balances, transactions or account details. Those belong with the operator's support team.
• End the call if the person sounds under 18.

What happens after a flag?

Detection is not action. Spotting a sign of harm protects nobody unless a person acts on it.

• Run more than one check. A strong setup has three: the agent flags harm during the call, the outcome is classified when the call ends, and a separate scan reads the player's own words afterwards for quieter signs the first two missed.
• Send every flag to a queue that a named person works through, and keep that queue as a record that can't be edited or deleted.
• Block flagged numbers from future marketing calls at the database level, so no later campaign can call them by mistake.
• Let the operator's trained team take the action: self-exclusion, account closure or support.

How do you prove the safeguards work?

• Before every campaign, make test calls that disclose harm, and check each one reaches the queue and blocks the number.
• Review flagged transcripts every week, including any the scan caught that the agent didn't.
• Track how long flags wait before a person acts.
• Keep responsible-gambling flags separate from simple account closures, so the figure your compliance team reports means what it says.

Can you test this yourself?

Yes. The demo agent on our iGaming page runs a welcome call for a fictional brand. Tell it you've been losing too much and want to stop, and listen to how it responds.

Related service: AI player reactivation for iGaming operators