AI Visibility Is Not One Score: How Brands Should Audit ChatGPT, Gemini and Claude

Posted on Oct 16 • Originally published at scalevise.com

A single AI visibility score can hide the answer marketers most need: whether a brand appears when a prospective buyer asks a specific AI assistant for help. Fractl's research into generative engine optimization finds that brand recall differs substantially by model, category and prompt intent. For businesses trying to earn mentions in ChatGPT, Gemini or Claude, the practical implication is clear: visibility should be measured separately, not blended into one headline number.

Fractl analyzed 8,090 keywords across 25 verticals and 22,410 leading domains. Its dataset covered 4,320 responses from GPT-4o in ChatGPT, Gemini 2.5 Flash and Claude Sonnet 4.6, using prompts intended to reflect buyer behavior in each category. The results, detailed in Fractl's Generative Engine Optimization study, show why a conventional ranking-style dashboard is an incomplete way to evaluate AI discovery.

Only 11% of brands in the dataset were referenced by all three models. About 77% were mentioned by just one. That means a brand can look visible in one assistant while being absent from another, even when users ask closely related questions. A blended score may smooth over that gap, but it cannot tell a marketing team where visibility is actually being won or lost.

The central finding is not that one model is universally better at naming brands. It is that each model can produce a different set of recommendations, and those differences change with the category and question. A travel prompt, an insurance question and a lifestyle recommendation do not create the same competitive field. Fractl observed concentrated travel results, a greater presence for digital-native insurance players, and direct-to-consumer brands appearing prominently in lifestyle-related results.

That makes category context as important as the model itself. A business that sells into more than one use case should not test one broad question and treat the result as representative. It should assess the buyer questions that matter to its own products, services and market position.

Fractl also identified a default brands effect. Some brands surfaced repeatedly across prompts and categories, even when they were not the leaders in traditional SEO metrics. At the same time, conventional search authority generally correlated with AI recall for most brands. The relationship was not absolute: roughly 5% of brands were underexposed in large language models relative to their search standing, while about 4% were AI overperformers.

For marketers, this is a useful distinction. Strong organic search performance remains relevant, but it is not a guarantee that an AI assistant will mention a brand. Nor does weak conventional visibility necessarily rule out strong AI recall in a particular category.

A useful audit starts with the questions buyers are likely to ask, rather than a generic prompt such as "best companies in [category]." Group prompts by category and intent. For example, separate research-oriented questions, comparison questions and recommendation requests where those reflect real customer journeys.

Then document which brands each model names, how often they recur across the prompt set, and which sources the models cite where citations are provided. Fractl's GEO framework recommends looking at these cited sources because they can indicate the types of coverage associated with credibility in an AI-generated answer.

This approach does not turn AI recall into a fully controllable metric. It does make the gap between a brand's desired category position and its current AI presence visible enough to act on.

Fractl argues that improving recall depends less on simply publishing more generic content and more on reinforcing clear category associations across the wider web. Owned pages remain part of that work, especially when they explain what a company offers in precise, useful language. But the study's framework places substantial weight on third-party coverage and digital PR that support the same association through high-authority publishers.

In practice, that means teams should connect content work with earned-media efforts. If a business wants to be associated with a defined product category, its website, expert commentary, research, partnerships and relevant external coverage should consistently communicate that positioning. The goal is not to chase mentions indiscriminately. It is to build credible, category-relevant signals that make the association easier to recognize across the web ecosystem.

The study's scope also sets important limits. It tested three AI models and a fixed prompt set, so its percentages should not be treated as a permanent market-wide benchmark. Results may change as models, their underlying data and the ways people phrase queries evolve. The durable lesson is methodological: measure the models and questions that matter to the business, then avoid treating one aggregated score as the whole story.

For businesses that rely on search discovery, this shift affects everyday marketing workflows. SEO reporting can remain a useful baseline, but it should sit alongside a recurring AI visibility review that tracks category prompts, competitor recall and external sources. That provides a more specific basis for deciding whether to strengthen a product page, develop a research asset, pursue relevant coverage or refine the way a category is described.

Businesses that cannot see how AI assistants describe their category risk spending on content without knowing whether it changes discovery. GEO Search Leads helps teams turn model-by-model findings into a practical plan for stronger AI search visibility and qualified lead opportunities, connecting category positioning, cited-source analysis and measurable follow-up work. Start an AI visibility audit.

What does AI visibility mean for a brand?

AI visibility is the extent to which an AI assistant mentions or recommends a brand in response to relevant user prompts. Fractl's research indicates that this visibility varies by model, category and prompt intent.

Why is one AI visibility score misleading?

Fractl found that only 11% of brands in its dataset were referenced by all three tested models, while about 77% were mentioned by only one. A combined score can conceal major differences between assistants.

The study analyzed responses from GPT-4o in ChatGPT, Gemini 2.5 Flash and Claude Sonnet 4.6.

How can a business improve its AI brand recall?

Fractl recommends auditing relevant prompts by model and intent, reviewing cited sources, and strengthening category associations through owned content, third-party coverage and digital PR.

Fractl's findings make the case for treating AI visibility as a set of model-specific, category-specific signals rather than a single marketing metric. Brands that audit the buyer questions relevant to their market, identify gaps across assistants and build credible category signals across the web will have a more useful foundation for AI search strategy than those relying on one blended score.