Telling customers they’re talking to an AI
Key takeaways
- Say it’s an AI in the first message, in plain words, along with what it can help with.
- If a customer asks whether they’re talking to a bot, the answer is always a straight yes. Test for it before going live.
- Make reaching a person one easy step, with honest timings and the conversation passed along.
Short version: tell customers they’re talking to an AI in its very first message, in plain words, and tell them how to reach a person in the same breath. If anyone asks “am I talking to a bot?”, it answers yes, straight away. Don’t give it a human name and a smiling photo. People forgive a helpful bot far more readily than one that pretended to be Sarah.
Why say anything at all?
Because customers are already wondering. Lots of people now start a chat by trying to work out whether a real person is on the other end, and the guessing itself is irritating.
The question isn’t really about technology. It’s about what they can expect. Can this thing actually sort out the problem, or will they end up explaining it twice? Tell them up front and you’ve answered that before they had to ask.
What should AI chatbot disclosure look like?
Short, early and honest. One or two sentences in the first message does it. Three things belong in there:
- That it’s an AI. Say so in ordinary words. “I’m an AI assistant” is clear. “I’m your virtual concierge” is not.
- What it can help with. Order tracking, delivery questions, starting a return: whatever you’ve actually let it do, and nothing you haven’t.
- How to reach a person. A word to type, a button to press or an email address, and roughly when a person will pick it up.
Say you sell garden planters online. The opening line might be: “Hi, I’m the shop’s AI assistant. I can track an order, answer delivery questions or start a return. Type ‘person’ at any time and someone from the team will reply during working hours.”
That’s it. No essay on how the AI works, no apology. Customers want to know what they’re dealing with, not read a manual.
Am I talking to a bot or a human? Answer it straight
Some customers will ask anyway, sometimes halfway through a conversation. The only good answer is a plain yes: “Yes, I’m an AI. Shall I pass this to a person?”
Check this before it goes live. Put the question into your test conversations in a few different wordings, and make sure the answer is honest every time. An AI told to sound warm and personal can drift into dodging the question, which is the worst of both worlds. There’s more on writing those tests in Test your AI before it talks to customers.
Skip the human name and the stock photo
It’s tempting to call the assistant Sarah and give it a friendly face. It makes the chat window look warmer. It also sets up a small fib that gets found out the first time the AI misunderstands something a person never would.
When that happens, the customer doesn’t just think the bot got it wrong. They start to wonder what else you weren’t straight about. A plain label such as “Shop assistant (AI)” costs nothing and avoids the whole problem. If you want some personality, put it in the tone of the replies, not in pretending to be someone.
Make reaching a person easy
Telling customers it’s an AI only helps if there’s a way past it. A handover that’s hidden, slow or goes round in circles undoes all the goodwill the honest opening bought you.
- One step, not a maze. A single word or button. Don’t make people ask three times, or prove the AI has failed first.
- Honest timings. If nobody’s around until Monday morning, say so. “A person will reply on Monday” beats a cheerful promise nobody keeps.
- Pass the conversation along. Whoever picks it up should see what has already been said, so the customer doesn’t have to start again.
- Some things go straight to a person. Complaints, anything about money and anyone who is clearly upset. The AI’s job there is to say a person will handle it, then hand over.
This fits neatly with sorting jobs into act, draft or never. Anything in the draft pile already has a person deciding, so the customer can be told a person is looking at it.
How can customers tell if customer service is AI?
Often they can, whether you say so or not. Instant replies at three in the morning, the same phrases repeated, polite answers to a slightly different question from the one asked. Customers are getting good at spotting it.
Which is the strongest argument for saying it yourself. If they’re going to work it out anyway, it’s far better they hear it from you in the first line than guess it on the fifth and feel misled.
What about replies a person has approved?
Not every AI reply goes out on its own. If the AI drafts an email and a person reads it, edits it if needed and presses send, a person has answered. The customer is dealing with your team, with some help on the typing.
Where the AI sends on its own, say so. Where a person has checked and approved every word, there’s no need to label each email. The line to hold is simple: never let a customer believe a person read their message when nobody did.
Want to know where AI could help your customer service without putting anyone off? Grafto’s AI Tools Report works it out from a short questionnaire.
Frequently asked questions
Should my chatbot tell customers it’s an AI?
Yes, in its first message and in plain words. Add what it can help with and how to reach a person. Customers who know what they’re dealing with get less frustrated and know when to ask for someone.
What should a chatbot say if asked “am I talking to a bot or a human?”
A straight yes, followed by an offer to pass the conversation to a person. Test this with several wordings before going live, because an AI told to sound friendly can drift into dodging the question.
How can you tell if customer service is AI?
Instant replies at any hour, repeated phrases and polite answers to a slightly different question are common signs. Customers spot these quickly, which is why it’s better to tell them yourself in the first line.
Questions, or want a tool we don't have yet? Email hello@grafto.co.uk — a real person replies.