Your customers don't hate AI — they hate you lying about it

Your customers don't hate AI — they hate you lying about it

New Twilio research finds that 81% of brands say their AI agents identify themselves as AI immediately. Yet, only 22% of consumers report the same.

That's a 59-point gap on a simple question with a very objective answer. Either the bot said it or it didn't.

Right? Right.

You're probably not lying. Neither are your customers, but they’re experiencing something that feels a lot like it.

Nobody's lying on purpose

Ask a product team whether their AI agent discloses and you'll get a confident yes. They can show you the line in the system prompt. It's in the design doc. There's a robot avatar right there in the corner.

It’s obvious, right?

Then watch a session recording.

The disclosure is in the tiny welcome message the customer scrolled past. Or it's the word "agent" in a name, which isn’t as defining as you think it is. Maybe it fires on the web widget but not on SMS. Perhaps the voice agent says it once (quickly) before the customer has finished getting settled.

Only 32% of businesses say they prioritize AI transparency when designing agents in the first place. That’s far less than half. So for most teams, disclosure isn't a decision anyone made. It's more likely a side effect of whatever the vendor shipped.

Ultimately, your customers can't tell either way. 75% of consumers believe they can identify a text-based AI agent. Tested, 90% failed.

So, no, they're not catching you. They're guessing, getting it wrong, and eventually finding out. That's the moment the trust goes.

3 things your customers want to know upfront

Consumers ask about three different things, and they care about all of them:

1. Who am I talking to?

76% of consumers think an AI agent should identify itself at the start of a conversation, but only 22% say that's what they get.

This is the cheapest fix in customer experience. One sentence, first turn, plain words. The words “AI agent” should be right there, upfront, easy to read. Not virtual assistant or handy-dandy digital friend or any of that. AI agent.

2. What can it see about me?

49% of consumers want transparency about what specific data an AI agent can access. However, only 19% of brands require opt-in consent before AI touches personal data.

That’s a crazy-big discrepancy. Add that to the fact that 62% of consumers already distrust how brands handle their personal data, and you’ve got a healthy dose of skepticism (perhaps deserved) before you’ve even deployed anything.

3. Why did it decide that?

54% of consumers say they trust AI more when brands clearly explain how the logic works. Only 48% of brands provide cues like a button explaining why something was recommended.

That’s not too big of a gap, though. The option only helps, and half of your customers explicitly want it. Plus, letting your customers see behind the curtain builds faith in your systems and reasoning. Sound logic (rather than black box magic) creates trust. Add it.

How to disclose (the right way) to your customers

Saying it isn’t the same as your customers understanding it. A few things that can help:

• Transparency at “hello”: Before the customer even states their problem.

• Use the word AI: Assistant, virtual agent, and concierge are all ambiguous. Don’t beat around the bush. Labeling an agent as AI leaves no room for confusion.

• Repeat it per channel: A customer who moved from web chat to SMS started a brand-new conversation in their head (even if you’re using Twilio Conversation Orchestrator). Disclose again. It never hurts.

• Say what happens next: Pair the disclosure with the exit. For example, you might go with something like: "I'm an AI assistant. Say ‘agent’ any time and I'll connect you to a person."

• Check what shipped: Pull transcripts and count. Twilio Conversation Intelligence runs language operators across conversations, so you can verify disclosure fired instead of trusting the config.

81% of brands believe they disclose their AI agent’s identity, but the number that matters most is what a customer would report after the call, and you can only get that by looking into your call data.

What transparency buys you

Being upfront costs a simple sentence. Yet, it delivers massive returns. 

92% of consumers agree brands can take steps to make them more comfortable with AI. Disclosure is the first one, and it makes every other trust signal matter more. Because a customer who knows they're talking to AI can make a good decision about what information to hand over. A customer who doesn't know carries the mental load of trying to figure out if they're talking to a human or a bot the whole conversation.

There’s no incentive here to trick your customers. Be upfront with them from the get-go and let them decide whether they want a bot for instant support or a human for more specialized issues.

Put yourself in your customers’ shoes. If you just focus on the numbers (call deflection, reduced ticket numbers), you’re going to alienate your customers. AI should augment your customer service, never replace it. 

It also saves time. Someone who knows from the get-go that they're talking to AI can escalate right away when their issue needs a human. That's a fast, clean outcome for both sides.

The alternative is a customer who finds out late. And the finding-out is what they remember.

Transparency is usually a measurement problem

Most brands aren't trying to deceive anyone. They're trusting a config file and reporting it as an outcome, which is how you end up 59 points away from what your customers actually experience.

The teams that close that gap treat disclosure the way they treat uptime:

• Instrumented

• Checked

• Tracked

• Reported

Tell people what they're talking to. Tell them what it can see. Tell them how to leave. You don’t have to guess what to build — this is all what your customers said they wanted.

Don’t just take our word for it, though. See the numbers and the research for yourself.

Read the Customer Insights Series report: How to Scale Automation Without Losing Customer Confidence. Or start building with Twilio for free.

Frequently asked questions

Should businesses tell customers they're talking to AI? 

Yes. Every time. Twilio research found 76% of consumers think an AI agent should identify itself at the start of a conversation. Only 22% of consumers say brands currently do this, against 81% of brands who claim they do.

Can people tell when they're talking to an AI agent? 

Twilio research found 75% of consumers believe they can identify a text-based AI agent, but 90% failed to tell the difference when tested. Self-reported confidence is a poor guide to actual detection. And when customers find out a bot is a bot (and not the human they thought), they get frustrated. Be upfront about it from the get-go.

What should an AI disclosure say? 

We recommend stating plainly that the customer is speaking with AI in the first message. Use the word AI rather than assistant or virtual agent, and pair it with a clear path to reach a human.

How do I verify my AI agent is disclosing itself? 

Twilio Conversation Intelligence runs language operators across conversation transcripts, so teams can check whether disclosure fired.