Why AI Keeps Naming the Same Brands (And Your Open Window)

You are trying to get mentioned everywhere at once. Every model, every prompt, every listicle someone swore the AI companies scrape. That spread is the reason you are not the answer to anything.

Pick one category you can realistically own, become the brand named across every kind of buying question inside it, and do it before the current leader's thin lead hardens into a habit. The odds are knowable, and two studies published this summer put numbers on them. AI models searched for brands they already knew 3.2 times more often than unfamiliar brands, according to the geoSurge study "AI Searches What It Remembers," reported by Danny Goodwin in Search Engine Land on July 30, 2026. At the same time, 89.3% of estimated AI search demand still sits in categories with no clear owner, from Kevin Indig's analysis of six months of Semrush ChatGPT data, reported by Greg Jarboe in Search Engine Journal on July 29, 2026. Memory is compounding against you. The door is still open. Both are true this quarter. Owners held their position in 90.4% of month-over-month comparisons, so the second one is the one with a clock on it.

AI shortlists you from memory before it runs a single search

Ask ChatGPT for the best physio clinic in your city and it does not read the entire web, weigh the options, and hand you a fair winner. It runs a handful of background searches first, what the industry calls fan-out searches, meaning the several separate queries a model fires off to assemble one answer. What it chooses to search for is already biased. geoSurge measured how models behaved when they had a brand in memory versus when they did not: familiar brands got searched 55.7% of the time, brands outside a model's top ten got searched 17.4% of the time. If you are in the second group, more than eight in ten relevant moments never even become a lookup, and you never see the loss because it happens before any tracking tool you own starts counting.

Brand names are scarce in these searches, and concentrated when they appear. Only 31% of fan-out searches named a company at all, and when a model did search for a specific brand, 63% of those searches involved one of its five most familiar. Read that as a five-name shortlist the model reaches for first. Your budget is not competing against the whole market, it is competing to be one of the handful of names the model already has loaded, and the incumbents are being handed their slots for free.

By industry the familiar-brand search rate ran from 41% to 82%, against 9% to 23% for unfamiliar brands, across travel, automotive, finance, business software, education, food and restaurants, luxury, fitness and wellness, and fashion. The method behind all of it: 66 U.S. buyer prompts run 60 times each between May 29 and June 9, 2026, producing 3,960 responses, 13,281 fan-out searches and 1,416 brand-level observations. Some industries had as few as six prompts, so treat the per-industry spread as a signal about your sector, not a scoreboard. The geoSurge authors also say plainly that this shows a relationship between memory and search behavior, not proof that memory causes it.

Memory is not a locked gate either. In the same study, Gemini searched for Lemon Squeezy, a brand that was not among the ones it was measured as familiar with. Live search can still surface a name the model does not know well. That is the crack you work through, and it is narrower than a marketing deck will admit.

Nearly nine-tenths of AI search demand has nobody sitting in the chair

Indig's dataset is the other half of the picture, and it is the half that should change what you do this month. Six months of ChatGPT answers pulled from Semrush, January through June 2026, covering 1,094 U.S. categories, five prompts per category, more than 50,000 brands and over 600,000 citations. He defined a "clear owner" as one brand showing up in at least four of the five test prompts and beating the runner-up by five percentage points or more. In June 2026, only 15.2% of categories had one. Another 53.7% were open fields with no brand anywhere close to locking the door.

The part that matters for where you spend is the split by demand. The higher-volume half of categories carries 98% of AI search demand, and its owner rate is only 11.3%, against 19% in the lower-volume half. Net: 89.3% of estimated AI search demand sits in categories with nobody established. The busiest rooms are the emptiest ones. That is not a permanent condition, and it is the single best reason to move a quarter's budget now instead of next year.

Once someone does sit down, they stay. Clear owners held first place in 90.4% of month-over-month comparisons. The categories where the leader actually changed had a median lead of just 1.3 percentage points going in; the ones where the leader held had a median lead of 2.9 points. Jarboe's practical read is to treat anything under roughly three points as contested. So the operating question for your category is not "are we winning," it is "is the current leader under three points, and can we take that before the number doubles." Indig is careful to call these correlations, not a formula, and he is right to be.

I have watched this exact shape before, without any AI in it. A London ADHD clinic I worked with went booked solid for three straight months. Demand got heavy enough that they hired more specialists and outsourced the overflow. What I would credit for it was not new technology. It was refusing to be the seventh-best answer to nine questions when you can be the first answer to one.

Your citation report looks healthier than your pipeline for a measurable reason

Most owners I talk to are measuring the wrong object. They have a tool counting how often their domain gets linked inside an AI answer, the number goes up, and the phone does not. Indig's dataset explains the gap without any hand-waving: citations and brand mentions were weakly and slightly inversely related, at a correlation of -0.229. More citations did not trend with more mentions; if anything the line ran the other way. The most-cited domain in a category matched the most-mentioned brand just 20.8% of the time. Being the source the model reads and being the name the model recommends are two different jobs, and only one of them puts a customer in your calendar. If you want the mechanics of that split, the difference between being a citation and being a recommendation is worth ten minutes before your next agency call.

The traffic runs the other way too. The most-mentioned brand in a category was cited at least once 69.9% of the time, so citation is a useful ingredient, just not the outcome. Get cited and you are in the room's doorway. Get named and you are the recommendation. Wellows, analyzing 11.1 million AI citations, found only 6.8% of them carried any brand mention at all. Two-thirds of the time the answer names nobody, which is a different problem from losing to a rival: there is no incumbent in the way, only an answer that has never had a name attached to it.

None of this makes your Google ranking useless, and none of it makes your ranking sufficient. Kevin Lee, founder of Didit, put it in Search Engine Land on August 3, 2026: "Ranking first in Google no longer guarantees you'll appear in the AI answer for the same query." If your entire visibility plan is a rank tracker and a monthly PDF, you are measuring a channel that no longer reports to the one you are being judged on.

Pick one category you can defend, then answer all five buyer prompts inside it

Start with the category, not the content. Indig's dataset tested each category with five prompt types: a definition question, a comparison, an alternatives list, a use case, and a buying question. That is your target set, and it is the whole set. A brand that shows up for the definition and vanishes on the buying question is not an owner; it is a reference book.

Picture a physio clinic that does post-surgical knee rehab and also treats backs, shoulders, sports strains and office posture. Trying to be the AI answer for "best physio" in a major city is a fight against hospital groups and chains. Trying to be the answer for post-surgical knee rehabilitation in that city is a fight against a handful of practices, and the buying question ("who should I see after ACL surgery in Manchester") is exactly the one that fills a diary. Same clinic, same budget, wildly different odds.

Then you build presence inside that one lane until the model stops treating you as a stranger. Same pages, same reviews, same directories, all of it pointed at the one category name instead of spread across nine service lines. The step-by-step version of the execution lives in my step-by-step checklist for improving your AI visibility, which is the companion to this piece: this one tells you which lane to claim, that one tells you how to work it.

Count how often you are named, not how often you are linked

Baseline first, because you cannot argue with a number you never wrote down. Run your five prompts, note whether your brand is named in each, and record the runner-up. Your lead in points is the share of your runs that named you minus the share that named the runner-up. Ten runs of each prompt is enough to make that number stable, which is fifty runs a month total. That is your scoreboard, and it takes about an hour a month. If you want a structured version of the test, the walkthrough for checking what AI currently says about your business covers the setup.

Monthly is the right cadence, not weekly. Indig's dataset was built month over month, owners held their position in 90.4% of those comparisons, and the movement worth acting on shows up as a lead crossing or falling below the three-point line. Weekly checking produces noise and anxiety in equal measure, and it tempts teams into changing strategy before the last change has had time to register.

Four things will tempt you to declare victory early. A rising citation count is the loudest one, given the -0.229 correlation between citations and mentions and the 20.8% overlap between most-cited domain and most-mentioned brand. Screenshots of a single lucky answer are the second, because one run of one prompt on one model tells you nothing; geoSurge ran each prompt 60 times for a reason. Tracking every platform equally is the third, since you have limited hours and your buyers concentrate somewhere specific. And "brand awareness" as a defense of flat results is the fourth. Indig found owners had higher branded search volume than their runner-up in 55.7% of pairs, higher organic traffic in 48.4%, and a higher Semrush Authority Score in 52.5%, and he calls those correlations, not a formula. Coin-flip odds are not a strategy you get to bill for.

The studies measure U.S. prompts across a handful of categories over a few months, so the exact percentages will move. The shape will not. Models favor what they already know, and most categories still have nobody they know well.

Frequently Asked Questions

How do I actually get ChatGPT to recommend my business?

Stop trying to appear everywhere and pick one category you can realistically be known for. Test it with the five prompt types Semrush used, a definition question, a comparison, an alternatives list, a use case and a buying question, and see whether any brand already shows up in four of the five. If nobody does, that category is open, and 53.7% of the categories Kevin Indig analyzed in June 2026 were exactly that. Then build enough consistent, verifiable presence around that one category that the model starts treating your name as familiar, because geoSurge found models search for familiar brands 55.7% of the time versus 17.4% for brands outside a model's top ten.

If I rank number one in Google, will AI mention me?

Not automatically. Kevin Lee, founder of Didit, said it directly in Search Engine Land on August 3, 2026: "Ranking first in Google no longer guarantees you'll appear in the AI answer for the same query." Being the page a model reads and being the brand a model names are separate outcomes, and Indig's data puts the correlation between citations and brand mentions at -0.229. Keep your rankings, because being read is still how you get into the pool: the most-mentioned brand in a category had been cited at least once 69.9% of the time. Just do not treat rank as proof you are in the answer.

How long before I become the brand AI names in my category?

The public data does not give a timeline, and anyone quoting you one in weeks is guessing. What it does show is the mechanics: Indig's six-month dataset ran January to June 2026 and tracked position month over month, clear owners held first place in 90.4% of those comparisons, and the categories where the leader actually changed had a median lead of only 1.3 percentage points going in. So the practical answer is that unowned and narrowly-led categories move within months, and well-defended ones rarely move at all. Check monthly, measure your gap to the runner-up in points, and judge progress by whether that gap is closing rather than by how long it has taken.

The scattershot approach is not lazy. It is usually the most expensive thing on the marketing plan, funded by people working hard in nine directions because nobody made them choose one. Choosing is the work. If you want a second pair of eyes on which category you can actually take, and what your current gap to the leader looks like across all five buyer prompts, book a short AI visibility call and bring your five prompt results with you. In seventeen years of doing this across 300+ businesses, I have never seen a brand win a market it would not name out loud as the one it was going after.