The argument is over
Disclosure has moved from an ethical preference to an operational necessity: customers increasingly assume it, regulators increasingly require it in some markets, and the cost of being caught concealing it is a trust problem rather than a service one.
Concealment fails on its own terms
Undisclosed AI is discovered — through phrasing, through speed, through the moment it doesn’t understand something a person would. The discovery converts a routine limitation into a story about your integrity, which is why hiding how an answer was produced costs more than the answer.
Customers forgive an AI that doesn’t know something. They don’t forgive being misled about what they were talking to.
Disclosure lowers the bar in your favour
People adjust expectations to what they’re dealing with. Told plainly, they ask clearer questions, accept a handoff more readily, and judge the interaction against a different standard. Concealment holds you to a human standard you cannot meet.
Say it once, plainly, at the start
One clear line at the opening beats a footnote or a disclaimer nobody reads. Name it, say what it can do, and say how to reach a person — which should be a real, working path, not a phrase that loops.
Don’t give it a fake biography
A name is fine and useful. A fabricated photograph, a claimed location, or a personal history is not, because it converts an assistant into a deception. The line worth holding is that nothing should imply a person exists who doesn’t.
Escalation is the other half of disclosure
Telling someone they are talking to an AI is only honest if a human is genuinely reachable. Pair the disclosure with escalation that triggers before the customer has to ask, and the policy holds up under pressure.