The AI visibility index: Which brands are vanishing from AI search?

Some of the brands pinch the strongest SEO footprints hardly look successful AI answers. Others pinch overmuch smaller hunt footprints show up again and again.

Fractl analyzed really consistently AI models urge brands, which brands vanish contempt beardown SEO footprints, and which signals abstracted the brands that overperform from those that under-index. (Disclosure: I’m the co-founder of Fractl.)

The short answer: integrated authority still matters, but it’s not the full map.

More than 9 successful 10 brands successful the dataset behaved the measurement astir SEOs would expect: stronger accepted hunt authority mostly tracked pinch stronger AI visibility. But the outliers are wherever the findings get interesting.

The spread betwixt accepted hunt authority and AI callback is wherever the visibility situation gets much complicated.

AI answers person default brands

Some brands pinch precocious domain ratings, ample keyword portfolios, and millions of monthly integrated visits hardly surfaced successful their ain categories. Others pinch smaller accepted hunt footprints appeared acold much often than their SEO metrics would predict.

Every manufacture we reviewed had a short database of brands that surfaced again and again, sloppy of really the punctual was phrased.

Those default answers weren’t ever the biggest companies successful the category. They weren’t ever the brands pinch the astir traffic. And they weren’t ever the companies a accepted SEO leaderboard would put first.

We noticed a fewer patterns look immediately:

• Travel was the astir concentrated category: Booking.com (285 mentions), Airbnb (227), and Expedia (215) accounted for astir 20% of the sector’s full mention volume. Three brands accounted for 1 retired of each 5 recommendations.
• HealthTech had a runaway leader: Teladoc (275) led Amwell (220), the runner-up, by astir 25%. The models person settled.
• Wellness had the deepest bench: Peloton, Headspace, Calm, Whoop, and Oura each cleared 168 mentions. No azygous marque dominates, which intends the category-leader slot is still up for grabs.
• Insurance collapsed the measurement manufacture watchers should person expected: Lemonade (213) classed supra State Farm (172). Root Insurance (165) classed supra Progressive (114). The models cited the digital-native carriers first and treated the bequest ones arsenic the alternatives, alternatively than the default.
• Lifestyle was the loudest awesome of the methodology working: Patagonia, Allbirds, Eileen Fisher, and Everlane (sustainability-coded direct-to-consumer brands) outranked Sephora, Samsung, and Whirlpool. Whether that’s a usability of really merchandise roundup articles get written aliases a deeper bias successful training data, the consequence is the same: The models for illustration the marque communicative that gets told successful third-party content.

The shape held crossed each sector. A fistful of categories do person winners the models intelligibly recognize, but plentifulness of the time, those picks don’t statement up pinch what you’d expect. Stripe wins successful FinTech, Notion successful SaaS, Coursera successful Education, and truthful on. Most of the time, though, the default names aren’t the ones you’d predict.

That’s the first strategical displacement for SEO leaders.

The competitory group wrong an AI reply is overmuch smaller than the competitory group connected a hunt results page. If your marque can’t ace the apical 5 to 10 names the exemplary recalls for your category, ranking good successful Google whitethorn not beryllium capable to get you into the AI-generated information set.

Dig deeper: Hidden gem publishers outperform awesome media connected assemblage affinity: Study

Traditional authority doesn’t guarantee AI recall

More than 9 successful 10 brands successful our dataset were broadly aligned: beardown accepted hunt authority usually tracked pinch stronger AI visibility. That’s the expected outcome. It’s besides why marketers shouldn’t propulsion retired each SEO rule they’ve spent the past decade building around.

But the edges of the dataset show wherever the strategy changes. About 5% (471 brands) were underexposed successful LLMs. These were companies pinch precocious domain ratings, very precocious integrated traffic, and a heavy keyword portfolio that still drew almost nary references from the models.

On SEO infrastructure, they look for illustration marketplace leaders. On LLM references, they look for illustration companies cipher has ever written about.

About 4% (377 brands) were AI overperformers. The models referenced them acold much often than their humble accepted signals would suggest, overmuch much aggressively successful their verticals than Ahrefs information says they should be.

Roughly 9% lined up pinch really often a marque appeared successful third-party contented that it did not create. Brands that showed up disproportionately successful roundups, master lists, and comparison reviews tended to beryllium acold much visible to the models because that benignant of sum gets ingested into the training data. Brands pinch plentifulness of hunt authority but small third-party sum tended to onshore successful the underrepresented group, sloppy of vertical.

That’s the strategical shift: What a marque says astir itself matters little for AI callback than what the remainder of the web has many times said astir it. Owned contented still matters. Technical SEO still matters. But for AI visibility, the corroboration furniture is becoming harder to ignore.

Some brands are miscategorized

The brands underrepresented successful AI are, successful galore cases, the bluish chips of their categories by astir accepted measures. These brands person domain ratings supra 80, millions of monthly integrated visits, and hundreds of thousands of keywords ranked. But the models don’t moreover mention them successful their ain sectors.

The rumor is categorization.

A fewer of the patterns we spotted:

• Microsoft and Spotify topped the database successful FinTech: They’re the giants of accepted visibility, but the models didn’t categorize them arsenic fintechs erstwhile prompted pinch fintech queries. That’s not a bug. Microsoft is heavy cited erstwhile you inquire astir productivity package aliases unreality platforms. It’s a categorization gap. The models person decided what counts arsenic a fintech, and Microsoft wasn’t connected the list.
• Legacy security carriers sewage displaced: Aetna, Cigna, Humana, Liberty Mutual, and Trustpilot each sat astatine DR 80+ pinch millions of monthly visits. The models hardly cited them. Lemonade and Root sat higher successful the aforesaid prompts contempt a fraction of the accepted footprint.
• Legacy unit brands sewage displaced: Sephora, Samsung, Whirlpool, and GE Appliances. The models preferred Patagonia, Allbirds, and Eileen Fisher erstwhile the punctual asked for manner recommendations. This is the astir visible awesome that the training information heavy weights direct-to-consumer marque narratives complete bequest retail.
• Healthcare giants were under-indexed. Medtronic, GoodRx, Humana, and 23andMe. The models cited telehealth-native brands for illustration Teladoc, Amwell, and Doxy.me instead.

Low callback doesn’t ever mean the exemplary doesn’t cognize your brand. It whitethorn mean the exemplary has revenge your marque nether a different class than the 1 your buyers are searching.

More contented unsocial won’t hole that.

You request stronger class signals successful the places models are apt to brushwood and reenforce those associations: expert content, manufacture publications, comparison pages, partner pages, customer stories, reappraisal sites, and category-specific media coverage. Before you pursuit much AI mentions, make judge the models understand which class you’re expected to win.

Dig deeper: What 1 cardinal keywords uncover astir AI’s effect connected search

The AI overperformers show the playbook

The different broadside whitethorn beryllium moreover much adjuvant to marketers, since it shows really brands punching supra their weight really propulsion it off. Smaller Ahrefs profiles. Larger exemplary recall. And each 377 of these brands show up many times successful the third-party worldly the models were trained on.

This database tells the story:

• Monday.com was the biggest overperformer successful the dataset, pinch an AI Visibility Score of 0.71. Monday.com already has a batch of accepted traffic, adjacent to 1 cardinal integrated visitors, and much than 50,000 classed keywords. Its overperformer position tracks straight pinch really overmuch much often the models referenced it than its SaaS peers.
• Root Insurance owned the security class outright, drafting much AI references than immoderate of the bequest carriers contempt being a fraction of their size.
• Six of the apical 15 overperformers were successful education: Stanford, MIT, Khan Academy, LinkedIn Learning, IBM Data Science, and Google Career Certificates. The models cited these for their credibility, not because they tally high-Domain-Rating commercialized platforms.
• Smaller players followed the aforesaid pattern. Doxy.me, Nike Training Club, and Google Flights each sewage pulled into merchandise roundups, showed up heavy connected master lists, and drew an outsized number of LLM references arsenic a result.

The communal thread isn’t conscionable “good SEO.” It’s category-level beingness successful different people’s content. That pushes AI visibility activity person to integer PR, contented distribution, and class authority building than galore teams want to admit.

If you want to cognize why a competitor pinch little than half your integrated postulation shows up successful the model’s reply while you don’t, the mobility to inquire isn’t, “What’s their SEO strategy?” It’s, “What’s their property coverage, and really did they gain it?”

You’re not conscionable optimizing your ain site. You’re shaping the group of sources AI systems usage to understand who belongs successful the category.

Most brands beryllium successful only 1 model’s world

Only 900 brands successful our dataset, aliases 11% of the total, were referenced by each 3 models. Another 12% appeared successful two. The overwhelming majority, 77%, were referenced by only 1 model.

That’s a awesome measurement problem. A marque tin triumph successful ChatGPT and suffer successful Gemini. It tin show up successful Claude and vanish from ChatGPT. It tin ain 1 model’s reply group while hardly registering successful another.

Each exemplary besides had a different fingerprint.

• Claude leaned much heavy toward SaaS and security brands, pinch Notion, Linear, and Lemonade appearing much often location than successful the different 2 models.
• Gemini leaned toward recreation and healthcare, pinch Booking.com, Teladoc, and Livongo appearing astatine a overmuch higher rate.
• ChatGPT was the astir consensus-driven of the three, pinch the astir overlap betwixt its apical brands and the different models.

The heatmap shows really unevenly distributed moreover statement brands are:

The gaps weren’t ever subtle. Notion was cited 17 times much often by Claude than by Gemini. Whoop appeared almost 4 times much often successful Claude than Gemini. State Farm drew almost half its mentions from ChatGPT and only astir a 4th from Gemini.

This is why aggregate AI visibility scores tin mislead elder teams. If a dashboard says your “AI visibility” improved, your first mobility should be: Where?

A azygous blended people tin hide the existent problem. You whitethorn not person an AI visibility problem broadly. You whitethorn person a Gemini problem. Or a Claude problem. Or a class punctual problem. Or a root problem.

Measure each exemplary separately. Track prompts by class and intent. Look astatine which sources look repeatedly. Then determine wherever the spread is worthy closing.

Some sectors person already been re-sorted

The disconnect betwixt accepted authority and AI callback wasn’t evenly distributed crossed industries. Some sectors still look a batch for illustration Google, while others person been almost wholly re-ranked by the models.

The shape breaks down otherwise by sector:

• Insurance was the slightest represented, astatine 8%. Legacy carriers for illustration Aetna, Cigna, Humana, and Liberty Mutual were replaced by digitally autochthonal carriers. If you’re an established carrier, you person the largest spread betwixt your Google authority and your AI recall.
• Education had the highest overperformer rate, astatine 7%, pinch Stanford, MIT, Khan Academy, IBM Data Science, Google Career Certificates, and LinkedIn Learning dominating. Professional acquisition and institutions do amended present than the commercialized platforms because they make credibility content.
• HealthTech was neutral (underrepresented by 5%, overperforming by 6%), pinch a displacement underway from pharma and aesculapian instrumentality companies to consumer-facing integer health.
• Travel had the smallest disconnect (4% under, 4% over) because LLM callback and accepted consciousness lined up closely. In travel, Booking.com, Expedia, and Airbnb are the best-known names to some Google and the models. Travel’s apical performers are stable.
• SaaS and FinTech sat betwixt 3% to 4% complete and 6% under. That’s because successful these categories, the communicative is 1 incumbent replacing different (Stripe vs. PayPal, Notion vs. Microsoft) alternatively than caller entrants displacing the aged defender entirely.

The shape crossed sectors mirrors the broader finding. Where the bequest authority of a class was built chiefly connected SEO and traffic, the models person re-sorted. Where the class was built astir marque stories that circulated successful third-party media, the alignment is tighter.

What this intends for marque strategy

These are my 5 main lessons from the study, astir ordered by really quickly you tin enactment connected them.

Measure accepted hunt authority and AI callback separately 

Domain rating, integrated traffic, and keyword rankings still matter. They conscionable don’t show the full story. 

A marque tin rank good connected Google and still neglect to show up successful AI answers. Another marque tin person a smaller accepted SEO footprint and still go a default recommendation. Track both. 

Measure wherever you rank successful Google and wherever you look successful AI responses. Then look for mismatches. The mismatches are wherever the strategy lives.

Build third-party validation, not conscionable owned content

The brands that outperform successful AI responses thin to look many times successful different people’s content. That includes:

• Roundups.
• Best-of lists.
• Comparison articles.
• Expert reviews.
• Analyst pages.
• Podcasts. 
• YouTube transcripts.
• Community discussions.
• Category-specific patient coverage. 

Owned contented tin thief explain your positioning, but it won’t switch outer corroboration. If AI systems are building answers from the broader web’s knowing of your brand, integer PR becomes 1 of the astir applicable levers for AI visibility.

Track each exemplary separately 

Only 11% of brands were referenced by each 3 models. That should termination the thought that “AI visibility” is 1 cleanable metric. Measure ChatGPT, Gemini, and Claude separately. Break prompts retired by merchandise category, purchaser intent, and comparison set. 

Track whether your marque appears, really it’s described, which competitors look pinch it, and which sources look to support the answer. A model-specific miss gives you a overmuch clearer action scheme than a blended visibility score.

Fix categorization earlier you pursuit volume

If your marque has beardown wide consciousness but anemic callback successful your target category, the rumor whitethorn beryllium categorization. The exemplary whitethorn cognize who you are. It conscionable whitethorn not deliberation you beryllium successful the reply group for the queries that matter. That requires a different strategy. 

You request much category-specific proof: media coverage, comparison content, partner references, customer stories, awards, reviews, and third-party pages that many times link your marque to the class you want to own.

Move earlier the default-answer slot hardens 

Some categories are still wide open. Wellness, lifestyle, and parts of HealthTech still person room for caller default answers. 

Others are already consolidating. Travel, awesome SaaS categories, and parts of FinTech are overmuch harder to break into because the models already return a tight group of acquainted names. 

The earlier you build class association, the easier it is to power recall. Once a exemplary many times sees the aforesaid brands attached to the aforesaid category, dislodging them gets harder.

What the web says astir your marque matters

Your adjacent marque visibility audit shouldn’t extremity astatine your site. It should representation the sources that thatch models who belongs successful your category: roundups, comparisons, master lists, patient coverage, reviews, partner pages, and organization discussions.

That’s wherever the overperformers successful this study separated themselves, not by being the biggest brands successful integrated search, but by being many times reinforced successful the places AI systems usage to understand, compare, and urge companies.

Methodology

Fractl analyzed AI visibility crossed 8 industries utilizing GPT-4o, Gemini 2.5 Flash, and Claude Sonnet 4.6. Each exemplary received the aforesaid 96 industry-specific prompts, pinch each punctual tally 15 times per model. The study produced 4,320 responses and much than 8,500 unsocial marque references.

Fractl past mapped the brands identified successful those responses to Ahrefs data, including domain rating, integrated traffic, referring domains, and keyword volume. This made it imaginable to comparison each brand’s accepted hunt authority pinch really often it appeared successful AI-generated responses.

The study grouped brands based connected really their AI visibility compared pinch their accepted hunt performance. Brands that appeared little often than their SEO metrics would propose were classified arsenic underrepresented, while brands that appeared much often were classified arsenic overperformers.