The enquiry arrived by email, like dozens do. When we first met in person, I asked what I ask everyone: “How did you find us?” Across the table sat the man whose board had put him in charge of the company’s digitalisation. His answer: ChatGPT had suggested which partners to consider. That is how the loyalty programme overhaul began - Slovenia, Croatia and Serbia, SAP Engagement Cloud and OpenLoyalty - the first major deal a machine has ever brought us.
After that meeting, I kept thinking about when I turn to the same machine for advice. Trail-running shoes are my one real indulgence - a new pair every two months - and I now run every choice past ChatGPT. This year I used it to buy artificial flies as well. I told it which stretch of the Sava I was fishing and when. I pasted in the water level from my angling club’s app. Then I asked which flies would work in those conditions. My fly-fishing mentor never explained his recommendations in that much depth - and I fished with him for two years, twenty-five years ago.
That is when it clicked: the buyer across that table was doing nothing exotic. He was doing exactly what I do - except the thing on the table was not a box of flies. It was our deal.
The buying process has not merely gone digital. It has moved to an adviser you do not know, cannot see, and cannot take to lunch. According to Gartner’s May 2026 survey, 45% of B2B buyers used generative AI in their most recent purchase. Mostly to gather information about providers. And 69% then take the AI’s findings to a salesperson to validate them.
Read that order twice. The machine builds the list; the human confirms it. Your salesperson enters the game only once you are on the list. Google gave the buyer ten blue links and left the choice to them. ChatGPT returns three names - and by then, the choice is mostly made. More and more buyers are not searching.
They are asking.
We now measure how often AI search engines mention FrodX. That also shows me what buyers actually ask these machines. This prompt is from our latest weekly report: “I’m looking for a partner in Slovenia who can implement HubSpot for a financial institution and understands regulation, consent and integrations. Who should I look at?”
That is not a keyword. That is a brief: industry, technology, regulatory context and decision stage in a single sentence. It is exactly how I asked about the flies - river, date, water level. The machine built a recommendation from context. For ten years, we optimised three-word phrases. Buyers, meanwhile, ask in paragraphs and get answers in names.
Now the part this column exists for. In the latest weekly report, our overall visibility fell from just over 20% to 16%. Most of the drop came from a single engine, so I do not read too much into weekly swings. These numbers are young and jumpy. The trend matters, not the week. One data point, though, was not noise. Visibility fell most sharply - by 26% - for this prompt: “We need a loyalty programme overhaul for Slovenia, Croatia and Serbia. Which partners can cover strategy, technology, integrations and campaigns?”
Read it again, then read the first paragraph of this column. It is almost a word-for-word description of the deal we are handing over right now. We are falling exactly where we have just delivered our biggest project of this kind - and where our international references go back years. In this very industry.
The loop is wickedly elegant. You win the deal. Because you take it seriously, your head is in the integrations, not in publishing. Because you do not publish, the machine slowly forgets you. Because it forgets you, the next such buyer gets three other names.
Successful delivery produces invisibility.
And this loss does not hurt, because you never see it. In a classic tender, you at least learn you lost. Here you never learn there was a tender at all. Somewhere, someone wrote down three names, yours was not among them, and the meetings moved on without you. No rejection. No trace in the CRM.
The reflex answer is obvious: publish a case study. And here I have to admit something you will not find in AEO handbooks: we hardly publish references. On purpose.
The first reason is uncomfortable. Every reference that names the client is an invitation. Some competitor will always go after the people named in it, insisting he could do it all cheaper if he just got one chance. The second reason is simpler still: our clients do not want to explain publicly how they fight for market share. A loyalty programme is a competitive weapon, not PR material.
So we use references in sales, not in marketing. Most of our clients will take a call from a prospective client of ours and talk to them one on one. Without us in the room. A client willing to do that is not someone you ask to pose for adverts.
That form of proof is the gold standard of trust. And to a machine, it is completely invisible.
The machine reads the internet, not phone calls. Discretion - our virtue and our clients’ shield for a decade - picked up a price tag overnight, printed in a weekly report.
So the dilemma is not “publish or stay silent”. The question is what the machine actually needs. Look at those prompts again: nobody asks who we worked for. They ask who knows how.
You don’t answer “who knows how” with logos. You answer it with knowledge. How you approach a loyalty overhaul across three markets with different currencies and habits. Where integrations really break. Which decisions only get expensive two years in. All of that can be written without betraying a client: no name, no numbers, no strategy. The client’s name stays out of it, the machine gets material, the reader gets a candid account. Reference calls remain our closing argument - we just need buyers to get that far again.
None of this is a new discipline. It is inbound marketing - except the search engine is now a language model. A machine reads your content before a human does. And it starts with measuring, not writing. First find out which answers you are missing from - then you know what to write.
Alongside that, one habit that costs nothing: ask every prospect how they found you. Without that question, this column would not exist. In web analytics, such a buyer shows up as “direct” - or not at all. And read the prompts in your own report as an editorial plan. Buyers have already written your questions for you.
This column is the first dish from that recipe. Even here, I have not told you who the buyer is. And yet you now know what we know.
My fly-fishing mentor knew more than ChatGPT ever will. But he never wrote his knowledge down - and this year, a machine sold me the flies. Your knowledge faces the same fate. The question is not whether unwritten knowledge runs deeper than written. The question is which of the two your buyer will ever encounter.
P.S. ChatGPT had no idea about the Sava’s water level until I pasted it in myself. And it knows nothing about you that you haven’t put online for it to find. Is it recommending you today? Check - the way we checked ourselves.