Methodology
How CiteTide measures your brand's visibility in AI search — openly, no black boxes.
How we create tracked queries
We don't measure whether AI knows you by name. We measure whether it recommends you when someone searches for a solution in your category without already knowing your brand. That's why the queries we track deliberately don't include your brand name — they're natural questions the way a real person would ask them while looking for a product like yours ("what's the best tool for X", not "is X any good"). We generate queries from your website's actual content, not just its name — a name can be misleading, and we can identify a product's real focus more reliably from what's actually on the site. You can always edit the suggested queries or add your own.
How we detect a mention
For every AI answer, we check whether your brand or its variants (which you configure) appear in it. The comparison is case- and diacritics-insensitive, so it catches a mention no matter exactly how the model writes it. The same way, we track which competitors AI names instead of you and which sources it cites — so you don't just see "mentioned / not mentioned", but also who else shows up in your category and where the model draws from.
Why we measure repeatedly over time
AI model answers aren't deterministic — the same query won't necessarily return an identical answer twice. That's why we don't measure just once. We measure each tracked query repeatedly (daily on paid plans, weekly on the Free plan), so instead of a single snapshot you see how things develop over time. A single result can be random; only repeated measurement shows the real state and how it changes. We don't repeat the same query multiple times within a single measurement run — instead, we build on the trend across days, which evens out random noise.
Which models we measure on
We measure across four AI systems: ChatGPT (OpenAI), Perplexity, Gemini (Google), and Claude (Anthropic). For every measurement we record the exact model version it ran on — models get updated over time, and a result from an older version isn't necessarily comparable to a newer one. This way you know you're measuring against what's actually running, and you can tell changes in visibility apart from a mere model change on the provider's side.
How the visibility score works
We calculate visibility as the share of tracked queries in which a given model mentioned you. If you track ten queries and Claude mentions you in four of them, your visibility on Claude is 40%. We calculate the score separately for each model, not as one blended number — because models differ, and the difference between them is valuable information in itself. When one model mentions you regularly and another almost never does, it shows you exactly where the gap is and where there's something to work on. We build our recommendations on this breakdown: instead of a generic "improve your visibility", we target concrete findings like "Perplexity doesn't cite you for this query, but ChatGPT does".
What we don't measure yet
We're upfront about the limits of the method. We don't measure Google AI Overviews (AI-generated summaries directly in Google Search) — Google doesn't provide an interface that would make this reliable and accessible. We don't do geographic simulation (answers can vary depending on where the user is asking from) — we measure from a single environment. And as mentioned, we don't repeat the same query multiple times within a single measurement run; we rely on the trend over time instead. We're gradually pushing these limits as what's technically feasible expands and as it actually helps our users.