On this page
It’s tempting to give an AI assistant a human name and a stock photo and let customers assume they’re chatting with a person. It feels warmer. It also tends to backfire — and in more and more places, it’s becoming something you’re required to avoid.
Why hiding the AI backfires
Customers notice. The answers come too fast, the phrasing is too even, it can’t follow a joke. The moment someone realises they were misled, everything else the business said becomes suspect — including the answers that were correct.
People are generally comfortable talking to AI for simple questions. What they dislike is being deceived about it.
The rules are moving in one direction
Regulators are increasingly requiring that people be told when they’re interacting with an AI system — the EU AI Act is one prominent example. Rules differ by country and change over time, so check what applies to you. But designing for clear disclosure now saves rework later.
What good disclosure looks like
- The bot pretends to be a person
- The way to a human is buried
- Answers give no source
- A label on every AI message
- A notice before the first AI reply
- A person is always one click away
- Answers cite their source
A clear label on every AI message
Every message the AI writes should be visibly marked as AI. In AskCub, Cubby’s messages always carry an “AI” label.
A notice before the first AI reply
Before the AI answers for the first time, tell the visitor they’re chatting with an AI assistant. Keep it short and friendly. In AskCub this notice is always shown and can’t be turned off.
An easy way to reach a person
Disclosure without an exit isn’t enough. Make “talk to a person” easy to find, and honour it immediately — or offer a message form if nobody is online.
A name is fine; pretending is not
Giving your assistant a name (Cubby is ours) makes it easier to talk to. Just don’t give it a human biography or claim to be a team member.
Explain what the AI uses
Customers increasingly want to know where AI answers come from. Two habits help:
- Show sources. An answer that says where it came from (“From: Returns policy”) is easier to trust and easier to check.
- Publish an AI policy. A short page explaining what the AI does, what data it uses and how to reach a person. Ours is at /ai-policy.
Does disclosure hurt satisfaction?
In practice, the opposite is usually true. When customers know they’re talking to AI, they phrase questions more simply, expect a quick factual answer — and are pleasantly surprised when it’s good. When the AI hands over to a person, the human feels like an upgrade rather than a correction.
The takeaway
Label AI messages, tell people before the first reply, show sources and make a person easy to reach. It’s the right thing to do, it’s increasingly the required thing to do, and it makes AI support work better.
Related articles
Trust & privacy2 min read
What to do when customers paste card numbers into chat
Sooner or later someone sends a full card number in a support chat. How to keep it out of your records — and what your team should say.
Read article
Trust & privacy2 min read
How long should you keep chat transcripts?
Keeping every conversation forever feels safe but adds risk. A practical way to choose a retention period — and what to do before data is deleted.
Read article
AI customer service4 min read
When should an AI agent hand a conversation to a person?
Five hand-off triggers that keep customers from getting stuck with a bot — and the three mistakes that make AI support feel worse than no support.
Read articleTry it on your own site
A free plan with no time limit, AI agent included. One line of code.