How to disclose AI on calls without losing the conversation

You should disclose AI on calls: say early, in plain words, that the caller is an AI agent, who it is calling for, and why. Laws and regulator guidance are moving toward requiring this, and even where they do not, a person who feels tricked into talking to a machine is unlikely to buy from you. Done well, disclosure takes five seconds and barely changes how calls go.

This is general information, not legal advice. Disclosure rules are developing quickly and differ by country, state and industry. Check current requirements for the places you call.

Is disclosure legally required?

The honest answer is "increasingly, and in a patchwork". A few points worth knowing:

• EU. The EU AI Act includes transparency obligations for AI systems that interact directly with people, so they are informed they are interacting with an AI, unless it is obvious from the circumstances. Application dates are phased, so check the current timetable and any guidance on implementation.
• US. There is no single federal rule requiring AI disclosure on every sales call. The FCC treats AI-generated voices as artificial voices under the TCPA, which affects consent (see TCPA and AI voice calls). Some states have passed or proposed laws on bots and AI disclosure, and the FTC has warned that deceiving consumers about whether they are speaking to a person can be an unfair or deceptive practice.
• UK. PECR and data protection rules require you to identify yourself and be transparent. Regulators expect honesty about who and what is calling. See PECR and AI calls in the UK.
• Industry rules. Finance, insurance and healthcare often have their own conduct rules that apply regardless of how the call is made.

Even if no law applies to your exact situation, a platform, carrier or customer contract might. And if the call gets recorded and posted online, "the sales call that pretended to be human" is not a story anyone wants.

Why disclosure is also good practice

Disclosure is a trust issue before it is a legal one. People tolerate automation that is honest about itself. They dislike being manipulated by it. Reasons to disclose early:

1. It sets expectations. People speak differently to an AI: shorter, clearer, and often more direct about what they need.
2. It makes the first ten seconds honest. The goal of an opener is to earn permission to continue, which is harder after a deception.
3. It protects you in complaints. A clear opener captured in the recording is evidence of good faith.
4. It filters in better conversations. People who dislike talking to an AI hang up quickly, saving everyone time.

What good disclosure sounds like

A good opener does four things quickly: who, what, why, and a way out.

"Hi Maya, this is Alex, an AI voice assistant calling for Greenfield Solar. You asked us about a quote on our website this morning. Is now an okay time for two quick questions?"

Variations for different situations:

Inbound call

"Thanks for calling Hartley Plumbing. I'm an AI assistant and I can book a visit or take a message. How can I help?"

Returning a missed call

"Hello, this is a call from Brightside Dental's AI assistant, returning your enquiry about a check-up. Is this a good time?"

Reminder call

"This is an automated call from Eastgate Clinic, an AI assistant, to confirm your appointment tomorrow at ten. Are you still able to make it?"

Wording to avoid

• "I'm just a virtual member of the team" is vague and often reads as evasive.
• Opening with a human name and a human line, then disclosing only if challenged.
• Burying "this call is from an AI" in a long legal paragraph.
• Phrases like "I'm a real person" or "I'm not a robot", which are false and may be unlawful.
• Disclosure that arrives only after the person has shared sensitive information.

Handling "are you a robot?" in the middle of a call

People will test the agent. The prompt should handle the question without drama:

"Yes, I'm an AI assistant. I can still help with your booking, or I can arrange for a person to call you back. Which would you prefer?"

Then the agent should do exactly what it offered. For hand-offs to people, see warm transfer from AI to human. Purple Voice can transfer hot calls to a human, so a "can I speak to a person?" request should lead to a real option, not a loop.

Good rules for the prompt:

• Never deny being an AI, even when asked repeatedly or jokingly.
• Answer once, briefly, and move on.
• Offer a human option whenever someone sounds uncomfortable.
• If the person says they do not want to talk to an AI, end politely and log it. Consider marking the contact for human follow-up only.

See how to write an AI voice agent prompt for where this fits in a prompt structure, and the opening lines that work for opener structures to test.

Disclosure and recording together

If you record or transcribe the call, mention that in the same breath. A single line covers both:

"This is an AI assistant calling for Northside Roofing, and the call is recorded so we can keep an accurate note."

Details on lawful basis, notices and retention are in GDPR and call recording. In some regions you may need explicit consent to record, not just a notice.

How to test your disclosure

Do not guess whether your opener works. Test it.

1. Call your own phone from the agent and listen to the first 20 seconds on a bad line.
2. Check that the AI statement, the company name and the purpose are all intelligible.
3. Read 20 real transcripts for the point where people hang up. If most drop immediately after the first sentence, adjust the framing, not the honesty.
4. Try two openers for a week each and compare how many people stay past 30 seconds.

A common pattern: openers that start with the person's reason for being called ("you asked about a quote") keep more people on the line than openers that start with the AI statement. You can still say both, in the first sentence.

A short disclosure checklist

• The opener states that the caller is an AI within the first two sentences.
• The opener names the business and the reason for the call.
• Recording is mentioned if applicable.
• The prompt forbids claiming to be human.
• A human option exists: callback or transfer.
• Opt-out requests are logged and honoured.
• Wording is reviewed against current local rules.

Try it

Purple Voice lets you write the agent's opener yourself and test it with a call to your own phone before any lead hears it. The Free plan has one agent, one number and 100 minutes a month with no card required. Start at voice.purplerocket.co and read is AI cold calling legal for the wider picture.