Quick answerAnswer immediately, plainly, and without qualifying language. State that the customer is talking to an AI agent, in one sentence, before continuing whatever task was in progress. This is a distinct moment from the upfront disclosure requirements many jurisdictions already impose at the start of a conversation, because the question means the customer's confidence about who or what they are talking to has already broken down mid-task, and any hedge or deflection at that exact moment reads as evasive even if the agent disclosed correctly at the start. A direct answer, followed by an offer to continue or to hand off to a person, protects trust far better than a clever non-answer does.
Why this moment is different from the disclosure requirement you already have
Most companies already have a policy for the opening seconds of a conversation, driven by disclosure and transparency requirements that specify what must be said before substantive help begins. That policy does not automatically cover what happens when a customer, three or ten turns into a conversation, stops and asks directly. By that point the customer has formed an impression, sometimes a wrong one, and the question is really asking whether that impression should be corrected. Treating the reactive question as already handled by the opening disclosure is the most common mistake here, and it is the one that produces the evasive non-answers customers screenshot and complain about.
The failure mode is a technically true but evasive answer
A surprising number of agents are tuned to respond to this question with something like "I'm here to help you with your request," which is not a lie, but it is also not an answer, and customers notice the dodge immediately. This pattern usually gets introduced by a well-meaning attempt to keep the conversation on task, or a worry that confirming AI status will make the customer disengage. Both concerns are secondary to the trust cost of an evasive answer. If the honest answer would ever damage the relationship, that is a signal about the underlying product experience, not a reason to obscure the answer to this specific question. This connects to the broader discipline of how an agent should express genuine uncertainty without losing trust: a direct question deserves a direct answer, stated with the same confidence as any other fact the agent is sure of.
From the team
We build production AI systems for startups.
LLM pipelines, RAG, and agent workflows that hold up under real traffic — not just in the demo.
Give the customer a real choice after answering
A good response to this question does more than confirm AI status. It should also make clear what happens next: the customer can continue with the agent, or ask to be moved to a person, and that choice should be honored without friction or a guilt-inducing script. This is where the reactive-question policy connects to your trustworthy AI agent UI design, since a visible AI indicator in the interface reduces how often this question gets asked defensively in the first place, but never eliminates it, because plenty of customers simply prefer to ask out loud and hear it confirmed in the conversation itself rather than trust an icon.
FAQ
What if the agent is uncertain whether the customer is testing it or genuinely confused? It does not matter. Answer directly either way. A customer testing the system deserves the same honest answer as one who is genuinely unsure, and treating the question differently based on suspected intent introduces exactly the kind of evasiveness that damages trust.
Should the agent explain why it did not disclose this earlier in the conversation? Only if asked. Volunteering a defensive explanation turns a simple, honest answer into something that sounds like an excuse. Confirm the AI status, offer the human handoff option, and move on.
Does the answer need to change for a returning customer who already knows? No. Answer the same way every time, regardless of how many times a given customer has asked before or how well established the relationship is. Consistency here is worth more than perceived redundancy.