What should an AI agent say when a customer asks if it's a robot?
Tell the truth in one sentence, offer a real person, and keep helping. Something like: "Yes, I'm an AI assistant for Bright Smile Dental. I can book you right now, or I can have someone from the front desk call you back." A quick honest answer keeps the conversation going. A dodge usually ends it.
Customers ask "am I talking to a real person?" for a few reasons. Some are curious. Some got a reply too fast or too polished. Some have had a bad experience with a phone tree and want to know whether they're about to waste ten minutes. Whatever the reason, they want a straight answer and a way forward.
The answer should do three things:
- Confirm it's AI. Use the words "AI assistant" or "virtual assistant". Skip vague lines like "I'm here to help" that read as a dodge.
- Name the business. The customer should know who they're dealing with.
- Give two ways forward. Keep going with the agent, or get a person. Let the customer pick.
Text or chat version
"Good question. I'm the AI assistant for Peak Roofing. I can check inspection times for this week, or I can have Mike call you. Which would you like?"
Phone version
"I'm an AI assistant for Peak Roofing. I can get your inspection booked in about a minute, or I can transfer you to the office. What works better?"
Is it legal for an AI agent to pretend to be human?
In several places, no. California bans bots that hide their identity to push a sale, Utah requires businesses to tell people they're talking to AI when they ask, and the FCC treats AI-generated voices on calls as artificial voices under the robocall rules. Honesty is the safe default everywhere.
California's bot law (SB 1001) took effect July 1, 2019. It makes it illegal to use a bot to communicate with someone in California to encourage a purchase unless you disclose that it's a bot. Davis Wright Tremaine has a plain summary.
Utah went further. Under SB 226, signed in 2025, a business using generative AI in a consumer transaction must disclose that the person is talking to AI when they make a "clear and unambiguous request" to know. The state can fine up to $2,500 per violation. The same law gives a safe harbor to businesses whose AI clearly discloses itself at the start of the conversation and throughout it.
For phone calls, the FCC's February 2024 declaratory ruling says AI-generated voices count as "artificial" voices under the Telephone Consumer Protection Act, so outbound AI calls need the same prior consent as robocalls. We break down what that means for calling your leads in our FCC rules guide.
I'm not a lawyer, and rules change state by state. If you operate in a regulated field like healthcare, law or finance, check with your attorney. For everyone else, an agent that never claims to be human keeps you clear of every rule above.
Do customers hang up when they find out it's AI?
Most don't, as long as you're upfront and the agent is useful. In Talkdesk's 2025 survey of 1,000 U.S. consumers, 84% said they want businesses to be transparent about AI. People are much more forgiving of a disclosed AI than of one they catch pretending.
The same Talkdesk research found only 51% of retailers using AI planned to tell customers about it, so many businesses are behind on this. Twilio reports that 76% of consumers expect an AI agent to identify itself at the start of a conversation.
A customer who finds out halfway through a call that the friendly "Sarah" was a bot feels tricked, and they remember it. A customer who hears "I'm the AI assistant, want me to book you or get someone?" usually says "book me" and moves on. What people really want is to get their problem solved quickly.
What are the best answers to "am I talking to a real person?"
Match the reply to the customer's mood. A curious customer gets a light, friendly answer. An annoyed customer gets an apology and a fast handoff. Someone with a complaint goes straight to a person. Write these responses into your agent's instructions ahead of time so it never makes one up.
| Situation | What the agent says | Why it works |
|---|---|---|
| Curious ("wait, is this a bot?") | "I am, I'm the AI assistant for [Business]. Happy to keep helping or get you a person." | Light tone, no defensiveness, keeps momentum |
| Annoyed ("I want a real human") | "Understood. I'll get someone from our team. Is this the best number to call you back?" | No arguing, no selling, gets a callback booked |
| Complaint or upset | "I'm sorry about that. I'm an AI assistant, so I'm passing this to [manager role] now with everything you've told me." | Upset customers need a person who can fix it |
| Asks on a voice call | "Yes, I'm an AI assistant. I can book you in about a minute, or transfer you to the office." | Short, because talking takes longer than reading |
| Testing it ("prove you're human") | "I'm not human, I'm an AI assistant. But I can check real openings for you." | Never play along with the test |
| After hours, wants a person | "The office opens at 8 a.m. I can book a callback for 8:15 so you're first in line." | Turns "no one's here" into a booked next step |
How do you set up the handoff to a human?
Decide in advance when the agent hands off, who it hands off to, and what it passes along. A good handoff sends the full conversation to a real person, books a callback when no one is free, and confirms to the customer exactly when they'll hear back.
- Write the disclosure line. One sentence with "AI assistant" and your business name. Put it in the agent's instructions word for word.
- List the handoff triggers. The customer asks for a person, sounds upset, mentions a complaint, a refund or a legal issue, or asks something the agent can't answer from your information.
- Pick who gets the handoff. A named role (front desk, office manager, owner) with a phone number or inbox that someone actually watches.
- Pass the whole conversation. The person taking over should see everything the customer already said, so the customer doesn't have to repeat it. This is where a CRM with a unified inbox earns its keep.
- Have an after-hours fallback. If nobody's available, the agent books a specific callback time and sends a confirmation text.
- Read transcripts every week. Search for "robot", "real person" and "human". If customers keep asking, look at what the agent said right before. That's usually the line to fix.
In our own calendar data, a wrong time zone on a booking caused a no-show. Small details in the handoff matter just as much: confirm the callback time in the customer's time zone, in writing.
Should the agent say it's AI before anyone asks?
Yes. Open with a short disclosure, especially on phone calls. It answers the question before the customer has to ask it, it gets you Utah's safe harbor, and it shows you're not trying to hide anything. One clause is enough: "Hi, this is the AI assistant for Peak Roofing."
Some owners worry this scares people off. A clear opener usually does the opposite. It sets expectations, and the conversation moves faster because the customer isn't spending half their attention trying to figure out what they're talking to. Our Voice AI agents can be set up to open this way on every inbound and outbound call.
You can still give the agent a name if it fits your brand, like "Ava, the AI assistant for Bright Smile Dental." A name is fine. A fake backstory, a fake location or a claim to be human isn't.
What mistakes make the robot question go badly?
Most bad moments come from five habits: denying it's AI, giving it a fake human identity, over-apologizing, making the human option a dead end, and letting the agent guess at answers it doesn't have. Each one costs trust, and some can cost you legally.
- Denying it. "No, I'm a real person" is the worst possible answer, and in California and Utah it can break the law.
- A fake human persona. Fake names with fake stories ("I've worked here five years") fall apart the first time someone asks a follow-up.
- Over-apologizing. "I'm so sorry, I'm just a bot, I know that's frustrating" makes the customer trust it less. State it and move on.
- A dead-end human option. If "get me a person" leads to a voicemail box nobody checks, the disclosure was pointless.
- Guessing. When the agent doesn't know your price or policy, it should say so and hand off. A made-up answer is worse than "let me get someone who knows."
How KenjiAI handles it. KenjiAI's Conversation AI and Voice AI agents answer texts, chats and calls, tell people honestly that they're AI, and hand off to your team with the full conversation in a unified inbox. If nobody's free, they book the callback on your calendar and send the confirmation. We've worked with 500+ businesses, and you can see what it costs on our pricing page.
What should you do this week?
Start by writing your agent's disclosure line and handoff rules. Then test them yourself by texting and calling your own business and asking "am I talking to a robot?"
If the answer is honest and short and gives you a clear way to reach a person, you're set. If it dodges, rambles or sends you to a dead voicemail box, fix the instructions before a customer finds the same problem.