Customers trust agents that say what they are

Customers are more willing to rely on an agent that tells them plainly it is an AI agent, what it can do and when a person will step in. Disclosure sets fair expectations. A customer who knows the limits accepts a handover or a refusal; one who thought they were talking to a person feels misled when they find out.

Why does disclosure build trust rather than lose it?

The fear behind hiding an agent is that customers will trust it less if they know. Some will, at first. But trust built on a misunderstanding does not survive contact with the agent's limits. Every agent eventually refuses something, hands over, or gives an answer that reads as automated. At that point a customer who was told the truth thinks "that's what it said it would do". A customer who was not told thinks "they tried to fool me".

The second reaction is far more damaging, and it spreads beyond the conversation. A customer who feels deceived about who they were talking to starts to doubt what they were told.

Disclosure also helps the agent do its job. A customer who knows they are talking to an AI agent phrases requests more clearly, accepts structured questions more readily and asks for a person sooner when they need one. That makes escalation work better, which is where support agents are most often judged.

What should the agent say about itself?

A short, plain introduction at the start of each conversation, and honest answers if asked later. It should cover:

1. That it is an AI agent, in those words or close to them.
2. What it can help with: the kinds of request it handles.
3. What it cannot do: the main things that need a person, such as exceptions to policy or decisions on complaints.
4. How to reach a person: what to say or press, and roughly how long it takes.
5. That the conversation is recorded, where that is the case.

The introduction should be brief. A paragraph of caveats before every chat is its own kind of poor service. The detail can come when it is relevant: the agent explains it cannot make an exception at the moment the customer asks for one.

What the agent should never do is claim or imply that it is a person. The list of what an AI support agent should never say to a customer puts that near the top.

Is disclosure a legal requirement?

In some places, yes. The EU AI Act's Article 50 sets transparency obligations, including that people must be told when they are interacting with an AI system, unless that is obvious from the circumstances. Other markets have consumer protection and fairness rules that make misleading customers about the nature of a service risky, even without an AI-specific law. In the UK, for example, the FCA's Consumer Duty expects firms to communicate in a way that supports good customer understanding.

Firms operating in several markets usually find it simpler to disclose everywhere than to manage different rules by channel and country.

This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.

Should the agent have a human name?

This is a real design choice with a trade-off. A name can make conversations feel friendlier and give customers something to refer to. But a human name, a photo of a person or a backstory blurs the line the disclosure is meant to draw. If the agent is called "Sarah" and has a profile picture of a smiling woman, a single line saying it is an AI agent is working against the rest of the design.

A reasonable middle path is a name that is clearly a product name, or a human-sounding name paired with persistent, visible labelling as an AI agent throughout the conversation, not only at the start. Whatever the choice, test it with real customers: if a meaningful number later say they thought they were talking to a person, the design is not disclosing clearly enough.

How does disclosure work across channels?

Each channel needs its own approach, because the cues differ.

Messaging apps need particular care, since conversations can be long-running and customers return to them days later. The approach is covered in how to answer customers on messaging apps with an AI agent.

Does disclosure change how customers treat the agent?

Sometimes, and not always well. Some customers are blunter or ruder to an agent than they would be to a person. That is not a reason to hide what it is. It is a reason to design for it: the agent should stay calm, set limits clearly and hand over when a conversation becomes abusive. The guidance in how to handle abusive customers with an AI agent applies regardless of whether the customer knows they are talking to software.

How do you know whether disclosure is working?

Ask, and watch behaviour. Post-conversation surveys can include a direct question about whether the customer knew they were talking to an AI agent. Complaints that say "I didn't realise" are a direct signal. Behavioural measures help too: requests for a person, repeat contacts and abandonment rates before and after a change in disclosure wording. The broader question of how to know if customers trust an AI support agent gives more methods.

What should you do differently?

Write the agent's introduction as if a regulator, a journalist and your most sceptical customer were all reading it. Make it short, plain and true. Label the agent throughout the conversation, not just at the start. Then measure whether customers understood, and change the design if they did not. Disclosure is part of the stance behind the AI support agent that knows when to say no: an agent that is honest about its limits from the first line.