New research from Suganthan Mohanadasan reveals a fundamental flaw in how most GEO practitioners think about AI brand visibility: ChatGPT injects brand names into its search queries before it retrieves any web results. In 21 of 27 open-ended recommendation prompts tested, the model wrote competitor names into its initial search query without any user input — deciding which brands to mention before a single page was fetched.

The citation impact is dramatic: brands that appear in ChatGPT's pre-search query get cited at a 68.9% rate, versus just 2.1% for brands not included in the query — a 33x gap. And only 3.1% of pages retrieved by ChatGPT's search are ever cited in the final response. The implication: most of what GEO practitioners optimize for (technical signals, schema, freshness) happens on the wrong side of the decision. The model names you before it even starts looking. Getting into ChatGPT's vocabulary — through years of training data exposure — matters 33x more than winning the retrieval round.

A survey of 163 SEO practitioners found a stark disconnect: practitioners rate AI visibility data highly valuable (4.20/5) but show far less willingness to fund the platforms providing it (3.19/5). The top barrier is not price — only 7% cited cost. The real objection is methodology opacity: 24% don't trust the numbers, and 20% can't connect the data to business outcomes. Venture-backed GEO platforms face an impossible position: disclosing methodology destroys their competitive advantage, but without it practitioners won't trust the data.

Healthcare is the vertical where AI search dominance is most complete: AI Overviews appear on 89% of health-related queries, and 3 in 5 U.S. adults used AI for health questions in the past 3 months. The content signals that determine which sources get cited in high-stakes YMYL queries are stricter than general GEO — named authors with verifiable credentials, answer-first formatting (direct 2-4 sentence response before elaboration), and cross-platform entity consistency matter disproportionately. The same signals that make health content citation-worthy also apply to any B2B content targeting expert-level queries.

"The 33x pre-query brand injection finding is the most important GEO research published this year. ChatGPT doesn't search, then decide - it decides, then searches. That sequence means the entire technical optimization layer most GEO practitioners obsess over (schema, robots.txt, llms.txt, freshness signals) is downstream of the real decision. I'm not saying those things don't matter - they matter for the retrieval round. But getting into the model's vocabulary through consistent training data exposure - reviews, editorial mentions, category comparisons over years - is worth 33x more than winning the retrieval round. The brands that understand this will build brand presence the way it actually works, not the way it's easy to measure."