What Is AI Visibility? Why One Score Doesn’t Work Across ChatGPT, Claude and Gemini

One thing I keep seeing working in the AI Search space is that people talk about ‘AI visibility’ like it’s one thing. It isn’t.

To show why, I ran 200 identical prompts, a mix of ecommerce, travel, services and pure informational questions, through Ebb, Propellernet Labs‘ AI visibility tracking tool. Ebb tracks how brands appear across ChatGPT, Claude and Gemini, and surfaces the fan-out queries, the follow-up searches those platforms run before answering, that sit behind each response. That’s what let me see not just whether each engine mentioned a brand, but how it went about deciding what to say.

Three engines, three different research strategies

ChatGPT ran no searches on 96% of factual questions, things like ‘what causes hay fever?’ or ‘how does compound interest work?’, answering straight from its training data. But it searched every single time a prompt had a commercial angle, like ‘best company for X’ or ‘where to buy Y’.

Claude searched on both factual and commercial prompts, but ran a consistently modest number of searches regardless of whether the question was broad or specific.

Gemini searched on almost everything, and the number spiked hard when a prompt combined multiple details. One example, ‘noise-cancelling wireless earbuds under £100’, triggered a notably high number of searches, while a less specific version of a similar prompt triggered far fewer. This is a single example rather than a tested pattern, but it points to something worth watching.

This was a single snapshot of all 200 prompts run at the same time, not repeated over multiple runs. I’ll be rechecking weekly to see how consistent these patterns actually are, and will update this piece as that picture develops.

Why this matters for strategy

None of these approaches match, and that’s the point. If you’re collapsing all of that into one ‘AI visibility’ score, you’re hiding more than you’re showing.

The tactics aren’t the same either. Chasing visibility on Gemini means chasing a model that searches constantly and reacts to specific, stacked detail, so structure and specificity in your content matters. Chasing visibility on ChatGPT for a factual, brand-related question is often a different job entirely, since it’s not searching live, it’s answering from what it already learned. This suggests a training data and knowledge base problem, rather than a live ranking one.

So which LLM should you actually be reporting on and optimising for? The answer depends on your audience and your category, not a single blended metric.

How we’re approaching it

This is exactly why we built Ebb inside Propellernet Labs. Fan-out queries, and the different ways each LLM decides whether and how to search, are the detail that gets lost when brands look at a single visibility score. Ebb tracks AI responses across ChatGPT, Gemini and Claude, surfaces the fan-out queries behind them, and benchmarks against named competitors, so you can see which platform is actually worth your focus and what’s driving your presence, or absence, in it. 

If you want to understand which engines matter for your brand and how to show up in them, get in touch. You can also sign up for the Ebb waitlist here, ahead of wider roll out later this year.