Perplexity read this brand's pages 12 times. Nine times it still recommended someone else.

Most brands assume that if an AI engine cites their website, they are winning. The citation means the engine read the page. The finding from a real 46-prompt panel: being cited and being recommended are two separate things, and the gap between them is larger than most people expect.

What the panel showed

We ran a full AI visibility audit for a regional internet provider. Forty-six buyer-intent prompts, each asked of four engines that use live web search: ChatGPT, Gemini, Perplexity, and Claude. That is 184 AI responses in total.

After the panel, we counted two things for each engine: how many times the brand's own domain appeared in the cited sources, and how many times the brand was named in the actual response text.

Engine	Times cited (URL in sources)	Times mentioned (named in answer)	Cited but not mentioned
Perplexity	12	8	9
Claude	10	15	1
ChatGPT	8	15	0
Gemini	0	14	0

Perplexity cited the brand's domain 12 times. In 9 of those 12 cases, the engine retrieved the brand's pages and still wrote an answer that named a competitor, not the brand. The brand's pages were read; the brand was not recommended.

What it means when an engine cites your URL

When an AI engine with web search runs a query, it fetches several pages before writing its response. That fetch step determines which URLs end up in the citation list. A URL in the citation list means the engine retrieved the page and used it as source material when generating the answer.

What it does not mean is that the answer favored the brand on that page. The engine may read your page, extract context about your service area and offerings, and then write a response that recommends a competitor based on other pages it retrieved in the same pass. The citation is an input. The recommendation is the output. They are not the same thing.

The Gemini inversion

Gemini's column in the table runs the opposite direction. It named the brand in 14 responses but cited the brand's domain zero times. This is because Gemini routes all its citations through vertexaisearch.cloud.google.com rather than returning the specific source URLs it retrieved. Every Gemini citation in a response points to Google's own intermediary, not to the original page.

So Gemini mentioned the brand 14 times without ever showing the brand's URL in its citation list. For Gemini, the citation count is always zero for every brand, regardless of how often Gemini surfaces the brand in answers. The metric that matters for Gemini is mention rate, not citation presence.

This is also why a per-engine breakdown is necessary. If you average the four engines together, the Gemini zeros pull down what looks like a citation rate, while masking that Gemini is actually naming the brand as often as ChatGPT and Claude are.

Why Perplexity reads and moves on

The 9 Perplexity cases where the brand was retrieved but not recommended follow a consistent pattern. Perplexity cited the brand's domain alongside several other sources in each query. The final response named one of the other sources as the primary recommendation. Perplexity appears to weight presence on the aggregator and comparison pages it retrieved more heavily than the brand's own pages, at least for geographic availability queries.

This is consistent with what the citation-source data showed for this panel overall: the most-cited domains were ISP directory and comparison sites, not individual provider pages. Perplexity retrieved the brand's pages as supplementary context, then built its answer from the directory infrastructure it had assembled.

Being on those directory pages would have put the brand into Perplexity's answer directly, rather than just into its source list. Getting retrieved is a prerequisite, not a guarantee.

What to track instead

Citation presence is not a useful optimization target on its own. A brand can increase its citation count by being on pages engines retrieve frequently, but if those citations are not converting into answers, the gain does not reach buyers.

The more useful metrics are mention rate (how often the brand appears by name in the response text) and recommendation rate (how often the brand is the specific option suggested). Both of these measure outcomes, not inputs.

Citation data is still valuable, but as a diagnostic: if an engine is citing your domain and not mentioning you, the content on your pages is being retrieved but not winning the answer. The fix is different from the case where you are not being retrieved at all.

• Not cited, not mentioned: you are absent from the pages engines retrieve for this query type
• Cited but not mentioned: engines find your pages but the answer goes elsewhere, often to better-represented aggregator sources
• Mentioned but not cited: the engine has absorbed information about your brand from other pages it indexed
• Cited and mentioned: engines are retrieving your pages and naming you in responses

Each case has a different fix, and conflating citation with recommendation leads to optimizing for the wrong one.