Should a chatbot say it is AI? The honest case for disclosure

A customer who believes they are writing to a person and discovers otherwise loses trust in the whole company; a customer who knew calibrates and moves on.

Somewhere in the market right now, a customer is typing a question to a company and wondering whether a person will read it. Many companies prefer that the question never gets answered: the assistant is given a human name, its sentences are sanded until nothing machine-like remains, and the disclosure is left out because honesty is assumed to cost sales. We think the assumption is backwards, and this essay argues the disclosure question the way we actually run our own assistant.

The honest position is that a chatbot should say it is AI, and the reason is mechanics before it is ethics. A customer who believes they are writing to a person and discovers otherwise loses trust in the whole company, because the deception retroactively poisons everything the assistant said. A customer who knows they are talking to an assistant calibrates naturally: they ask plainer questions, they forgive the seams, and when the moment comes they route to a human without resentment, because nothing was hidden from them.

The false version buys a few extra minutes and pays for them with the relationship. Sanding the edges off an assistant to impersonate a person works until it does not, and it always does not: the customer asks directly, or the assistant breaks a promise no person would have made, or the reveal comes from a colleague's offhand comment. When the truth surfaces, the customer does not just distrust the chat. They reread every message they sent and wonder what else was staged.

The failure modes are familiar. The assistant that dodged the question about being a bot until the third ask, at which point the sale was already dead, because the dodge itself was the answer. The company whose disclosure existed but was buried in a footer where nobody reads, which is disclosure as decoration. And the team that never told customers anything, not because they decided against it, but because nobody ever sat down and decided what the honest answer was. The last one is the most common, and it is the easiest to fix.

The practice starts where the conversation starts: a plain disclosure, in ordinary language, that the other side is an AI assistant, placed where a reader actually sees it rather than in a footer. When asked directly, the answer is direct, and this one is ours to own: our assistant says plainly that it is AI and names the human who takes over afterward. And the handoff that follows the disclosure is one prepared handoff, a single defined route to a person, not a scramble invented on the day a customer demands one.

Guidance, not guarantees: we built and run our own assistant on these terms, and we would not run it any other way, because the trust we keep is worth more than the minutes we would gain. What any given sector requires in the way of disclosure, and where regulatory duties begin and end, are questions for practitioners in that sector, and we make no claims about them. What we can say flatly is that the truth is easier to operate than the costume.