Your brand is recognized by AI but still not recommended

Nike, New Balance, and Reebok share an identical Google Knowledge Graph description: “Footwear company.” Across five AI systems, all three are correctly identified as athletic footwear brands. Nike appears in 71% of athleisure recommendation responses. New Balance and Reebok appear in 0%.

That single finding, drawn from research published in Search Engine Land by Maryanna Franco on June 11, 2026, makes the case that GEO has been solving the wrong problem. The field has focused on entity clarity: structured data, SameAs markup, well-formed About pages. Those signals help AI systems recognize a brand. They do not determine whether that brand enters the recommendation set when a user asks a category question with no brand name attached.

Recognition and recommendation are separate signals with different inputs.

Franco and co-author Joao da Silva ran 14,140 API calls over seven days, testing 12 athleisure brands across ChatGPT, Gemini, Perplexity, Claude, and Google AI Overviews. They split prompts into two types: recognition (“What is this brand?”) and recommendation (“Best athleisure brands”). The citation sources split sharply by prompt type. For recognition prompts, own-brand content drove between 23% and 49% of citations depending on the platform. For recommendation prompts, own-brand citations dropped to 18% on ChatGPT and to effectively zero on Gemini, Claude, Perplexity, and Google AI Overviews. Third-party sources accounted for 82% to 100% of what each system cited when no brand name was in the query.

The authors call the gap between recognition and recommendation “the framing gap.” The mechanism behind it is co-mention density: how frequently a brand appears alongside the brands that already define a category, across articles, roundups, editorial comparisons, and review pieces. lululemon and Alo Yoga co-occur 534 times in UK-indexed athleisure content. Nike and lululemon co-occur 482 times. New Balance and lululemon co-occur so rarely that the pair does not appear in the top co-mention pairs at all.

LLMs do not infer category adjacency from a brand’s attributes. They pattern-match against co-occurrence in the corpus they draw on. If a brand has never appeared consistently alongside the cluster leaders in category-aligned external content, the model has no signal to associate it with that category. Knowledge Graph says “Footwear company” for New Balance; the third-party corpus confirms footwear; athleisure queries retrieve the athleisure corpus. New Balance is not in it.

The authors frame three distinct problems that most GEO audits collapse into one. Entity clarity, which gets a brand recognized, is a problem solved on the brand’s own site. External credibility, which gets a brand considered, is a PR and corroboration problem. Co-mention density in the right category cluster, which places a brand in the concept graph for a specific recommendation query, is a category-positioning problem. Each requires a different solution.

The study has real constraints worth noting. It covers one category in one geography, with UK-indexed sources from May 2026. Cross-category validation is ongoing. The co-mention figures are raw co-occurrence counts, not a weighted influence model. Still, the 71% versus 0% result across three brands with identical entity descriptions is a clean natural experiment. The methodology and extraction code are published on Zenodo for independent replication.

For GEO and SEO teams, the 90-day priority is auditing co-mention exposure, not only entity markup.

The practical diagnostic is direct: pull your brand’s press coverage from the past six months and check whether the articles that mention you also name the two or three brands that already define your target category. If your coverage places you in isolation (“a leading provider of X”) rather than alongside your category cluster (“alongside Y and Z in the premium X space”), you are building recognition signals while the recommendation gap stays open. The fix is off-site: earn coverage in roundups that name your category peers, pitch podcast introductions that place you in explicit relation to cluster leaders, and seek inclusion in analyst and retailer taxonomy pages where category leaders already appear. Schema and SameAs markup will not close a gap that lives in the co-occurrence structure of third-party content.

This analysis is based on reporting by Maryanna Franco published in Search Engine Land on June 11, 2026.