Most brands have no idea whether AI models recommend them. I decided to find out systematically.
Last year I started noticing a pattern in client conversations: companies with strong SEO metrics, solid backlink profiles, good content programs — but no idea what AI models said about them when buyers asked category questions.
So I built a dataset to answer it.
The GEO Brand Citation Index tracks 41 brands across ChatGPT-4o, Perplexity AI, and Gemini 1.5 Pro. For each brand, I recorded citation frequency, the context in which they were mentioned, and cross-platform consistency. The full dataset is open at thegeolab.net.
Here's what I found — and why it matters for anyone building a product or tool that buyers evaluate using AI assistants.
The Methodology
The process was straightforward but time-intensive.
For each brand in the index, I ran category-level prompts across all three platforms — the kind of queries a real buyer would ask when evaluating options: "best tool for X," "top platforms for Y," "which companies do Z well."
I recorded:
- Whether the brand appeared in the response
- The position and framing of the mention
- Whether the response was recommendatory, neutral, or negative
- How consistent the answer was across platforms
41 brands, three platforms, hundreds of queries. The output is a citation frequency score for each brand on each platform, plus a cross-platform delta metric showing divergence between ChatGPT and Perplexity.
The Surprising Finding
The thing that surprised me most: SEO metrics don't predict AI citation.
Brands with high domain authority, lots of backlinks, and consistent content output can still be near-invisible in AI-generated responses. I've labeled these "Ghost brands" in the dataset — real market presence, minimal AI presence.
The inverse also holds. Some brands with modest SEO metrics show up consistently across all three platforms, in recommending contexts, with high frequency.
The difference comes down to something simpler: what independent sources say about you.
AI models don't index your content directly. They synthesize representations of categories from everything they've been trained on — analyst reports, trade coverage, independent reviews, research citations. A brand cited consistently by credible third parties, in consistent language, describing consistent capabilities, builds a strong model representation. One that only controls its own narrative — no matter how much content it produces — doesn't.
Four Archetypes in the Data
The 41 brands in the index cluster into four patterns:
Dominant — consistently cited across all three platforms, in recommendatory contexts. These brands have achieved what amounts to default status in the model's category representation.
AI Memory — strong on ChatGPT, weak on Perplexity. ChatGPT draws on static training data. Perplexity retrieves live web content. AI Memory brands built authority at some point but haven't sustained the independent citation signal. Their ChatGPT presence is essentially a historical artifact.
Ghost — strong SEO, weak AI citation across all platforms. The most strategically important segment for most teams to think about. These are brands losing discovery share in a channel they're not measuring.
Unclassified — inconsistent patterns, often tied to fragmented or inconsistent category positioning across third-party sources.
Why ChatGPT and Perplexity Don't Agree
One operationally useful finding: the platforms behave differently enough that a single "AI visibility" score is misleading.
Perplexity retrieves live web content. A credible article mentioning your brand today can influence Perplexity's responses within days. ChatGPT is trained on historical data — your current earned media may not be reflected for 12–18 months, if at all until the next training cycle.
The practical implication: if your buyers research on Perplexity, recent earned media in high-signal outlets is high-leverage right now. If they're on ChatGPT, you're playing a slower compound game of building durable third-party authority that survives training cycles.
Knowing which platform your audience actually uses is now a meaningful strategic input.
Running Your Own Audit
The methodology behind the dataset translates directly into a repeatable audit any team can run:
- Pick your five most important category queries — what a buyer asks when evaluating tools in your space
- Run each on ChatGPT, Perplexity, and Gemini — note where you appear, what context, how competitors are framed
- Assess your third-party citation profile — analyst coverage, independent reviews, trade mentions, research citations
- Map the gap between how you describe yourself and how independent sources describe you — divergence here directly suppresses AI citation
- Track quarterly — model behavior shifts with training updates and retrieval changes
The brands establishing this baseline now are building a measurable lead. The window is open in a way it won't be in two or three years.
The Dataset Is Open
The full GEO Brand Citation Index — citation scores, archetype classifications, delta metrics, methodology — is available at thegeolab.net.
If you're working on anything related to AI visibility measurement, brand monitoring in LLMs, or GEO strategy, the dataset is there to use.
Note: This article was written with AI assistance.
Top comments (0)