How to disclose AI to customers without killing the conversation

How to disclose AI to customers without killing the conversation

Everyone’s hesitant to admit they use AI, and many customer service teams don’t want to outright disclose the AI agent to customers. They think if the customer learns it’s a bot, they’ll bail.

Sure, that’s a fair concern, but it's also mostly wrong.

76% of consumers think an AI agent should identify itself at the very start of a conversation. They want to be told. Not because they want to abandon the conversation—because they want to know if they’re talking to a human or a robot. That’s a fair question to ask, especially depending on their question or issue.

Honesty isn’t hurting your engagement. What tanks engagement is a disclosure that sounds like a disclaimer, arrives late, or reads as an apology for the experience the customer is about to have.

There’s a right and wrong way to go about disclosing your AI agent to customers. Here's how to say it so the conversation keeps going.

The drop-off fear (warranted or not)

The concern is that unmasking the bot gives people permission to leave. Yes, some will. And some definitely should. Those are largely customers whose issue needed a human anyway, and you'd rather lose them in turn one than after four minutes of failed containment.

Consider the alternative, though. A customer who doesn't know spends the conversation trying to work out what they're talking to. 75% of consumers think they can spot a text-based AI. However, 90% get it wrong. So they're guessing, guessing badly, and eventually finding out.

And that doesn’t make them happy. Not one bit.

That's the moment that costs you. Not the disclosure.

Simple, short disclosures that work

Keep your AI disclosures short, specific, and paired with the exit.

1. Web chat

The greeting does three jobs: name what it is, say what it's good at, and show the door.

Hi, I'm an AI assistant. I can help with orders, returns, and account questions. Type "agent" any time and I'll bring in a person.

No apology, no hedging, and no paragraph about how the AI works. The customer gets what they need to decide how to proceed, then gets on with it.

2. SMS

Tighter constraints but the same rules. Start with the disclosure.

This is an AI assistant from [Brand]. I can check your order status or start a return. Reply HUMAN to reach an agent.

A customer who moved from web chat to SMS thinks they started a new conversation. Disclose again. Just in case. It doesn’t hurt.

3. Voice

Voice is where hedging does the most damage. That’s because there's no transcript to scan through. Say it in the first sentence before the caller starts explaining.

Hi, this is an automated AI agent for [Brand]. I can help with billing and appointments, or say "agent" and I'll transfer you.

Then leave a beat. Voice interfaces reward a pause where a person would take one, and it gives the caller time to decide what they need. Give them a few seconds.

5 things that kill the conversation

The disclosure itself isn’t usually the problem. No, but these are:

1. Burying it: Disclosure in a pre-chat notice, a footer, or a settings page doesn't count. If it isn't in the first message, assume the customer didn't receive it. That’s why only 22% of consumers say AI agents identify themselves as AI immediately.

2. Dodging the word: Call a bot a bot. Virtual assistant, digital concierge, and smart helper are intentionally ambiguous. Say AI. People know what it means, and softening it produces the confusion you were trying to avoid.

3. Apologizing for it: Lines like "unfortunately you've reached our automated system" tell the customer this is the worse option before they've tried it. You've pre-loaded the complaint.

4. Over-explaining: Nobody wants a paragraph on your model architecture. Two sentences, then get to work.

5. Disclosing without an exit: This is the one that reliably backfires. Telling someone they're talking to a machine and giving them no way out produces the bypass behavior you were worried about. 63% of consumers want easy escalation at any point. 47% of brands provide it. And that’s exactly why 78% of customers have tried to bypass AI agents to get straight to a human.

How to know if your disclosure worked

81% of brands say their AI identifies itself immediately. Only 22% of consumers say it happens. They can’t both be right.

Check the transcripts. Twilio Conversation Intelligence runs language operators across conversations, so you can count how often disclosures fired, on which channels, and in which turn. Look at that next to your containment and escalation rates before and after a wording change and you'll know within a week whether the drop-off fear was real.

Most teams find it wasn't.

Good disclosure is short, early, and honest

Say it first, say the word AI, say what it can do, and say how to leave. New Twilio research shows customers are asking you to do this. Statistics aside, it’s the least you can do (and it usually ends up more favorably for you anyways).

The teams that get this wrong aren't the ones being blunt. They’re the ones keeping customers guessing until they’re so unsure they demand a human.

We have numbers to back all of this up, too. Check out the Customer Insights Series report: How to Scale Automation Without Losing Customer Confidence. Or start building with Twilio for free.

Frequently asked questions

How should a chatbot introduce itself as AI? 

Twilio recommends a first-turn greeting that discloses the AI, states what it can help with, and provides a path to a human.

Does disclosing AI reduce engagement? 

Twilio research suggests the opposite risk matters more. 76% of consumers want AI to identify itself upfront, and 78% have tried to bypass an AI agent, which usually points to a missing exit.

Should I say AI or virtual assistant? 

Twilio recommends saying AI. Terms like virtual assistant and digital concierge are ambiguous. This just reproduces the confusion that your disclosure is meant to prevent.

When should an AI agent disclose itself? 

Twilio recommends disclosing in the first message before the customer describes their issue. 76% of consumers expect disclosure at the start of a conversation. Late disclosure feels more like an admission or correction.

Do I need to disclose AI again on a different channel? 

Twilio recommends disclosing on every channel. A customer moving from web chat to SMS or voice experiences a new conversation (even when your system treats it as one continuous thread).