Corporate AI already quotes our prices, takes our data, and decides our cases.
But when the chat closes, we are left with nothing. No record of what was promised. No proof of what was said. Nothing to point at when tomorrow's answer contradicts today's.
I noticed this properly during a mystery shopping test of an airline's chatbot. The bot was, genuinely, good — honest about the rules, resistant to manipulation, helpful beyond what I asked. Then I requested a confirmation of our conversation and got: "I cannot provide a transcript or confirmation of this conversation." Everything it had told me about my flight conditions evaporated the moment I closed the tab.
Think about what that means in any other commercial context. You buy bread and you get a receipt. You call a bank and the call is recorded "for quality purposes" — theirs, notably, not yours. You sign a contract and both sides keep a copy. The entire architecture of consumer trust rests on the boring principle that interactions leave traces, and that the weaker party keeps one.
AI conversations are the first mass commercial interaction in modern history designed to leave the customer with nothing.
That is not a conspiracy. It is an oversight, of the kind that happens when a technology is deployed by engineers optimising for containment rates and by executives optimising for cost per contact. Nobody in that room is the person standing at the check-in desk being told something different from what the bot said. The customer's evidentiary position is simply not a metric anyone owns.
So I named the missing thing. I call it the AI receipt: a keepable record of what a company's AI did, said, and promised.
There is precedent, and I like it because it is so unglamorous. In 1879 a saloon owner in Dayton, Ohio named James Ritty patented one of the first cash registers. His motive was not consumer protection at all — he suspected his bartenders were pocketing the takings. But the side effect changed commerce permanently: the transaction now printed a record, and both parties held the same version of what happened. Trust in retail was not built on honest bartenders. It was built on a machine that made dishonesty inconvenient.
We are at the pre-cash-register stage of AI services. The bartender might be honest. You have no way to prove it either way.
What belongs on the receipt is not complicated: what I asked, what the AI answered, what it promised or refused, what it decided about me and on what basis, when it happened, and how to escalate. Any competent engineering team could ship that in a sprint. The reason it does not exist is not technical difficulty. It is that nobody has asked for it.
That is the part consumers actually control, and it is why I wrote the brochure — this is not a problem waiting on regulators. Every time one customer says "please send me a confirmation of this conversation," a company logs a request it cannot fulfil. A thousand of those and it becomes a product requirement. The rules for how AI treats customers are being written right now, in ordinary support chats, by whoever bothers to ask.
For business owners reading this, there is a self-interested version too. The receipt protects you as much as the customer. When someone claims your bot promised them a refund it never promised, the transcript ends the argument in thirty seconds instead of turning into a public complaint you cannot disprove. The company with records wins disputes. The company without them settles them.
Ask for the receipt. Give the receipt. Same document, both sides better off — which is usually how you can tell a fix is the right one.