In the Eyes of AI, Your Brand May Have Already Disappeared

A few days ago, an SEO friend of mine called me, and there was a hint of panic in his voice.

His site, he said, has a Domain Rating (Ahrefs' site-authority score) over 80, several million organic visits a month, and a keyword bank of hundreds of thousands of terms. By any standard of the past decade, that is textbook-level excellence.

But recently he asked an AI: for our category, recommend a few brands?

The AI rattled off five names.

He wasn't one of them.

Not a single one.

He asked me: man, have I wasted these ten years?

I didn't know how to comfort him, because just days earlier I had seen a study published by Fractl about exactly this phenomenon. The study used three models — GPT-4o, Gemini 2.5 Flash, and Claude Sonnet 4.6 — and ran 96 industry prompts, each prompt fifteen times per model, for a total of 4,320 answers and more than 8,500 brand mentions. Then the brands were matched against Ahrefs data: Domain Rating, traffic, keyword count.

After the comparison, the answer stings:

Being king on Google does not mean you still exist in AI's answers.

First, the conclusion: most of the time, the old rules still hold

Don't panic yet.

More than ninety percent of the brands in the data behaved the way you'd expect: those with high traditional search authority generally also had high AI visibility. What does that tell us? That AI isn't magic — it feeds on the content of the entire internet. The SEO you've done over the past decade was not wasted.

But.

The remaining ten percent or less is where things get genuinely interesting.

About 5% — 471 brands — have high authority, big traffic, and lots of keywords, yet AI almost never mentions them. On the infrastructure of SEO, they are market leaders; in AI's citations, they look like they were never written about at all.

Another 4% — 377 brands — are exactly the opposite. Unremarkable traditional numbers, yet AI recommends them frequently, at rates far beyond what their "SEO footprint" should warrant.

On one side, invisible giants; on the other, over-cited underdogs.

Inside this crack hide the new rules of brand competition for the next few years.

In AI's answers, there are "default brands"

What is a "default brand"?

It's the handful of names AI keeps recommending, no matter how no matter how many different ways you phrase the question.

Every industry has them. And these default answers are often not the biggest company in the industry, not the highest-traffic brand, and not number one on the traditional SEO leaderboard.

Let me share a few findings from the study so you can feel it for yourself.

Travel is the most concentrated. Booking.com was mentioned 285 times, Airbnb 227, Expedia 215. Together the three account for roughly one-fifth of all mentions in the industry. In other words, every five times AI recommends a travel product, one of those picks lands on these three.

Insurance is the most interesting. Lemonade ranked 213, ahead of State Farm's 172. Root Insurance ranked 165, ahead of Progressive's 114. You read that right: digital-native insurance companies ranked ahead of century-old incumbents. AI treats the new as default and the old as backup.

In health tech, Teladoc is in a league of its own with 275 mentions, a quarter ahead of second-place Amwell. The models have already "made their choice."

Only the wellness category has no absolute leader: Peloton, Headspace, Calm, Whoop, and Oura all sit above 168 mentions.

What does that mean? It means the "default answer" seat in this category is still open.

The competitive set in AI answers is far smaller than a search results page. For a given category, AI only remembers roughly five to ten names. If you can't get in, then no matter how high your Google ranking, you'll never make AI's generated shortlist.

Some brands aren't forgotten — they're filed in the wrong drawer

Those invisible giants — does AI really not know them?

No.

Take Microsoft and Spotify, both high on the FinTech "invisibility list." But ask about productivity software or cloud platforms, and Microsoft gets cited heavily. It's not unpopular; the models have simply filed it under another category. Ask about fintech, and Microsoft simply doesn't have a slot in the model's drawer.

Insurance again: Aetna, Cigna, Humana, and Liberty Mutual are all giants with Domain Ratings above 80 and millions of monthly visits, yet AI barely mentions them. In the same set of questions, Lemonade and Root rank ahead.

Retail is the same. Asked for lifestyle recommendations, the models favor brands with sustainability stories — Patagonia, Allbirds, Eileen Fisher — while Sephora, Samsung, and Whirlpool get pushed aside. Are Samsung's products bad? Not necessarily. It's that in the training data, third-party content more often ties that cohort of new consumer brands to "worth buying."

Your brand's low recall may not mean the model doesn't know you — it may mean the model has filed you under another category. Your buyers are searching drawer A, and you've been placed in drawer B.

This problem can't be solved by publishing more content. What you need is to repeatedly build the signal "you belong to this category" in the places models read: analyst reports, industry media, comparison pages, partner pages, customer stories, review sites.

First get AI clear about which drawer you belong to; then chase impressions.

The over-cited underdogs reveal the playbook

Now the other side: those 377 brands whose AI presence far exceeds their footprint. What did they do right?

monday.com is the most over-represented brand in the entire dataset, with an AI visibility score of 0.71. Its traditional numbers are actually not bad — nearly a million monthly organic visits and more than fifty thousand keywords — but the models cite it far more often than its SaaS peers.

Root Insurance, a fraction of the size of those century-old incumbents, receives more AI citations than any traditional insurer.

Six of the top fifteen over-represented brands come from education: Stanford, MIT, Khan Academy, LinkedIn Learning, IBM's data science programs, and Google Career Certificates. The models cite them not because their SEO is strong, but because they are trusted.

Then there are the smaller players like Doxy.me, Nike Training Club, and Google Flights, appearing again and again in reviews, roundups, and expert lists — and so getting remembered by the models again and again.

See the pattern?

The place these underdogs win is not their own websites — it's other people's content.

Reviews, rankings, comparison articles, expert evaluations, podcasts, YouTube transcripts, community discussions. The models learned "who belongs to which category" by reading the entire internet. What your own website says carries limited weight; what others repeatedly say you are — that is what you are in the eyes of the model.

So the next time you find a competitor with half your organic traffic showing up in AI's answers while you're absent, don't go study its SEO strategy.

Go study its media coverage. How did it earn those mentions?

What you're optimizing has never been just your own website. What you're shaping is the material AI reads when it learns about your category.

77% of brands live in only one model's world

There's another finding that may make many marketing departments squirm.

Across the entire dataset, brands cited by all three models (ChatGPT, Gemini, Claude) account for only 11%. Cited by two models, 12%.

The remaining 77% are cited by only one model.

Just one.

This means a brand can win in ChatGPT and lose in Gemini; be everywhere in Claude and nowhere to be found in ChatGPT. Notion is cited by Claude 17 times as often as by Gemini. Whoop appears in Claude at nearly four times its frequency in Gemini. Nearly half of State Farm's mentions come from ChatGPT, while Gemini contributes only about a quarter.

Three models, three palates. Claude favors SaaS and insurance — Notion, Linear, and Lemonade show up more with it; Gemini favors travel and healthcare — Booking.com, Teladoc, and Livongo get mentioned more by it; ChatGPT follows the crowd most — its rankings overlap the most with the other two.

So if your dashboard says "AI visibility is up," don't celebrate yet.

First ask: up where?

A blended average will hide the real problem. You may not have a general AI visibility problem at all — what you have may be just a Gemini problem, a Claude problem, or a category-prompt problem.

Test per model. Break prompts down by category and by intent. Watch which sources keep appearing. Then, and only then, decide which gap is worth closing.

Five takeaways for brand owners

Having read the study, I've compressed it into five points, ordered by how fast you can act on them.

First, measure Google rankings and AI recall separately. Two sets of books — don't mix them. Where they diverge is exactly where the strategy lives.

Second, stop creating only owned content. Go earn third-party validation: rankings, reviews, comparison pages, expert lists, podcasts, community discussions. For AI visibility, digital PR may be the most practical lever.

Third, track per model. ChatGPT, Gemini, and Claude each keep their own books. One model's gap is a hundred times more useful than one average score.

Fourth, fix your category association before chasing volume. The model has to know which drawer you belong to before it can recommend you.

Fifth, move while the "default answers" haven't hardened yet. Travel, mainstream SaaS, and parts of fintech have already tightened considerably; but in wellness, lifestyle, and parts of digital health, the default seats are still open. Once a model has repeatedly seen the same brands bound to the same categories, prying them loose will get harder year by year.

Finally

Back to my friend.

His decade of SEO wasn't wasted — those efforts are still feeding his site. It's just that from now on, being king on your own turf isn't enough.

The next time you do a brand visibility audit, don't stare only at your own website. Draw a map: which content did AI read to learn "who belongs to my category"? Reviews? Rankings? Review media? Communities?

None of those 377 underdogs won by sheer size. They won by being read, again and again, in the places where AI does its reading.

In the eyes of AI, what your brand is depends on what the entire internet says you are.

That line deserves a spot on every brand builder's monitor.