Two ChatGPTs, One Question, and Two Different Dubais

On Saturday night I ran a small experiment. France plays Spain in the World Cup semifinal on Tuesday, and I wanted to watch it somewhere in Dubai with seven friends. So I asked ChatGPT for a recommendation. Twice. Once on my own account, the one that has known me for years, and once in a clean session, logged out, no history, no memory.

Same question, word for word. The two answers did not share a single venue.

The city I expected, and the two I got

Before reading either answer, I wrote down what I expected: Barasti, McGettigans, maybe Lock Stock & Barrel. The places everyone in Dubai names for a big match. The classics.

Neither answer contained any of them.

My logged-in session picked P7 Arena at Media One Hotel: AED 125 entry, two drink vouchers, doors at 8pm. Alternatives: a private simulator bay at Five Iron Golf for up to ten people, the DWTC fan zone from AED 63, a club in Business Bay promoting France v Spain on the big screen. The clean session recommended Bridgewater Tavern, The Yard at Topgolf, Irish Village, Huddle at Mall of the Emirates, and Kickers in Dubai Sports City.

Two complete answers. Two confident rankings. Zero overlap.

Small detail that stayed with me: my session quietly corrected my "7 friends" to plans for eight people, because it counted me. It also knew the venue was 20 to 25 minutes from my neighborhood without my saying where I live. The clean session, meanwhile, asked me which area of Dubai I was coming from. One of these machines knows me. The other was meeting a stranger.

Then I asked about a television

Second question, same two sessions. My TV is dead, the semifinal is Tuesday, I need a new 75-inch.

My account answered in a way I can only describe as decisive: "Buy this one: Xiaomi TV S Pro Mini LED 75." It referenced a comparison we had run together in June and confirmed the current UAE price, AED 3,099 on Noon. It even told me to order only if checkout confirmed delivery and installation by Tuesday afternoon, July 14. That is not a search result. That is a purchase decision, made, scheduled and closed.

The clean session shortlisted differently: the Hisense 75-inch Mini-LED as best overall value at AED 4,299, and it called it "my pick."

So I challenged my own session with three words: "Why not Hisense?"

It folded instantly. "You are right to challenge me: Hisense should have been in the shortlist." Then it revised the whole recommendation: "At AED 4,299, I would seriously choose the Hisense U7Q Pro over the Xiaomi." From "buy this one without overthinking" to a different brand, in one conversational turn.

The sponsorship that never came up

Here is the part I have not stopped thinking about.

Hisense is an official FIFA World Cup 26 sponsor, its fourth World Cup in a row. In both sessions, the product images in the answer carried the trophy and the official partner banner right on the box art. The machine was literally displaying the sponsorship while it spoke.

And in all the text, across both sessions, the World Cup sponsorship was mentioned exactly zero times. I asked why Hisense was or was not recommended, and I got dimming zones, nits, refresh rates, UAE pricing. Not one word about the brand having attached itself to the very match I was buying the television to watch.

The sponsorship bought the photo on the box. The argument that moved the purchase was made somewhere else entirely, from spec sheets and retailer listings, assembled in seconds by a machine that has never seen an ad.

The shelf has moved

I have spent my career around brands and retail, and what I saw on Saturday night is a shelf. A shelf where placement changes per person, where a three-word challenge reorders the products, and where the brand's most expensive marketing asset is present as a JPEG and absent from the reasoning.

My decision, somewhere between AED 3,000 and AED 4,300, was effectively made inside those two answers. Multiply that by every television, every restaurant table for eight, every hotel room, every B2B shortlist. Buyers increasingly decide with a machine. And what the machine says about a brand is volatile, personalized and, in most companies, entirely unread.

That last word is the one that matters. Unread. Ask a CMO what their agency said in last year's campaign and you will get the deck. Ask what the machine told a logged-in customer in Dubai at 11pm on a Saturday, and the honest answer is: nobody knows. There is a version of your brand you have never read, and it is the only one your customer met that night.

Why I think this is a discipline, not a task

The moment this stopped being an observation and became a conviction for me had nothing to do with televisions. It was my gardener, standing in my garden with his phone out, asking ChatGPT how to save my wife's cactus from the Dubai summer sun. A man who treats plants for a living, checking his own trade against the machine. That is when I understood the reach and the speed of it: the machine has become a default trust layer, even for experts of their own topic.

Then there is Lionel Henshaw, a professional photographer in Cape Town (lionelhenshaw.com). In February 2025 his inquiries suddenly jumped and he could not explain why. He asked new clients how they had found him, expecting the usual word of mouth. The answer, again and again, was ChatGPT. We share a friend, so we dug into it together and traced the machine's trust back to its source: a local newspaper article from four years earlier naming his work the best hotel photography of the year in Cape Town. A story he had almost forgotten was quietly earning him clients through a machine he had never thought about. Demand arrived before awareness. Lionel was lucky: the version of him the machine had assembled happened to be accurate. Most brands will not find out whether theirs is.

Some will file this under search optimization with new letters, and the market will likely settle on a term like AI visibility for part of it. I think that undersells what is happening. Reading what the machine says about your brand, in every market and every mood, finding where it drifts from your words, and teaching it your version is not a campaign tactic. It is closer to brand management itself, rebuilt for a buyer who asks a machine first.

This is the thesis we started AYAN on. We read what the machine says about your brand. We find where it drifts. We make sure it learns yours. The customer stays the hero of that story; the work is giving brands their voice back in the one conversation they currently cannot hear.

One number I would defend at any dinner table: Gartner predicts that 70% of customer interactions will begin and end with conversational assistants by 2028 (cited in Contentsquare's "What's Next in CX: 2026 Digital Customer Experience Trends"). My gardener and Lionel's clients are simply early.

Spain won, by the way. The forecast card in my session had France at 60%, a number the machine was quoting from a betting market. It does not predict; it repeats whoever sounds credible. That is the entire point.

On Tuesday night, somewhere in Dubai, eight friends ended up at a venue none of them chose, recommended by a machine none of them questioned. The brands in that answer earned their place through data they mostly do not know exists. The brands missing from it will never know they were absent.

I find that more interesting than any campaign I have seen this year.