The decision is made before you ever see them.
Last summer my wife and I were driving through Italy. Rome, then Montepulciano, Viterbo, Sorrento, Matera. Somewhere on one of those roads I picked up my phone and started planning next year's trip. Not on Google. I opened ChatGPT and told it what I wanted: somewhere in Latin America, flying Aeromexico so we could use the credit card points we cash in every November, the way we'd done for Europe. We love food, architecture, a bit of shopping. Find me two places that fit.
It gave me Lima and a couple of Mexican options. So I kept going. We'd already done Mérida, so I asked for a second pick outside Mexico. It came back with Cartagena. Then I narrowed again: our real shortlist is Peru and Colombia, we want to split it into two trips across the year, here's roughly how. Then I asked the question that changed the whole thing. I told it what I was spending right there in Italy, 180 euros a day on food, drinks and museums, and that our flights to Rome had cost 5,500 pesos for the two of us on points. Compare that, I said. Tell me what Peru and Colombia actually cost against this.
No search box takes that. There is no keyword for "compare this to what I'm spending today, on the trip I'm on, at the budget I actually have." I wasn't searching. I was thinking out loud with something that thought back.
The decision formed in the conversation, not in an answer
By the end of that exchange I had a plan. Two countries chosen, a rough sense of the cost, the year split into two trips. The whole thing took maybe fifteen minutes across a dozen turns, and every turn fed it something no search engine ever had: our travel history, our points, our budget, the trip we were physically on.
Here's the part that matters. Fourteen months later, this September, we're flying to Colombia. Cabo to Mexico City to Medellín to Bogotá to Cartagena and back. We booked it. Real money, real flights, the trip that started as a conversation in a car.
And when we booked, I didn't reopen that chat. I didn't need to. It had already done its work months earlier. No tool will ever record that it did. In any analytics dashboard anywhere, that booking is "direct traffic, July 2026." The thing that actually shaped the decision left no fingerprint at all.
That's the shift, and it's bigger than travel. People have moved the entire middle of the decision, the part where you weigh, narrow, and talk yourself into something, into a conversation with an AI. It happens when you're choosing a country, a laptop, a pair of running shoes, a place to live. You reason it out in private, you come out the other side already leaning, and by the time anyone in the business sees you, the deciding is mostly done.
Everyone is watching the wrong moment
Right now the whole industry is shouting about one thing: how to show up in AI search. How to get cited. How to be the answer ChatGPT reads back.
That work is real, and I'll come back to it, because it matters more than it looks. But most of the noise is aimed at the last inch of a mile-long process. Being the cited answer is one moment near the very end. It is not where the decision gets made. The decision got made across all those turns before, in the reasoning, the narrowing, the back-and-forth where the model helped me think. No optimization tactic touches that part, because there's nothing there to rank for. There's just a person and a conversation.
This is why attribution is quietly breaking. We spent years building tools to trace the path: the search term, the click, the journey you could follow. That path is going dark. The moment that tips a decision now happens somewhere you cannot see, cannot rank for, and cannot measure. You can't A/B test your way into someone's private conversation with a model.
Now make the conversation about buying a home
My trip is a clean example because it stays vague. Nobody is the promoted answer to "should I visit Peru or Colombia." There's no name to surface.
But watch what happens as a conversation like that narrows. Change the subject from travel to property. Someone opens ChatGPT and says they're thinking about buying in Los Cabos, here's roughly their budget, what should they be looking at. The model talks them through San José del Cabo versus Cabo San Lucas, the trade-offs, the neighborhoods. Then comes the turn that changes everything: "who's a good agent I should actually talk to?"
Now there's an answer. A specific, nameable one. At the top of that funnel, "which country," there was nothing to cite. At the bottom, "who do I call," there absolutely is, and it's either your name or it isn't.
That's the whole thing. The deeper the conversation goes, the more specific it gets, and the more it needs a real, trustworthy, citable answer. Being present in AI isn't about winning the vague top of the funnel where nothing can be won. It's about being the answer that exists when the conversation finally gets specific enough to name someone.
What you actually do about it
So both things are true at once, and they don't cancel out.
You cannot trace the decision. It forms in a conversation you'll never see, and no dashboard will tell you which turn did the work. Stop expecting a clean line from a tactic to a sale. That line is gone.
And you can still decide whether you're the name at the end of it. You do the work not because you can prove which piece converted, but precisely because you can't, so the only safe move is to be genuinely present and genuinely worth referencing at every point that conversation might touch. Have the site that earns trust when someone checks. Be the source that's solid enough to get cited when the question gets specific. Be the answer that holds up when a buyer, mid-conversation, asks the model who they should call.
The businesses that win here aren't the ones with the cleverest checklist for gaming an AI answer. They're the ones who understand that the decision already happened somewhere they couldn't watch, and who made sure that wherever it happened, they were the obvious thing to land on.
You can't get into the room where your buyer is deciding. You can make very sure that when they walk out of it, they're walking toward you.