A brand appeared in 28 percent of AI responses. The recommendation rate was under 1 percent.

We ran 184 AI responses for one brand across four engines. The blended mention rate was 28 percent, or 52 responses. When we looked at what triggered those 52 mentions, 51 came from prompts that named the brand in the question. Against organic buyer-intent prompts, the kind someone types when they have not heard of you yet, the brand appeared zero times.

The audit

We ran a full prompt panel for a brand in a locally competitive services category: 46 buyer-intent prompts, each run against ChatGPT, Claude, Gemini, and Perplexity using live web search. That is 184 responses total.

The panel was divided into two types. Thirty prompts were organic: category-level buyer questions where no brand name appeared ("best internet provider near [city]", "rural internet options in [county]"). Sixteen prompts were branded: questions that named the company directly, covering comparisons, reviews, and direct questions about the brand. Each prompt ran on all four engines, so the organic prompts produced 120 responses and the branded prompts produced 64.

Where the 52 mentions came from

The 28 percent blended mention rate is 52 of 184 responses. The source of those mentions is not evenly distributed across both prompt types:

When a prompt names the brand, the engine almost always mentions it back: 81 percent of the time. That is not surprising. If the question is "Is [brand] worth it?", the engine is going to discuss the brand. When the question is about finding a service provider without naming anyone, the brand appeared in none of the 120 responses across all four engines.

The one recommendation

Of the 52 mentions, one was an explicit recommendation. It came from Perplexity, on the prompt "HillFi vs Starlink: which is better?" The engine concluded that the local ISP was the better option for fixed-home internet within its coverage area.

That prompt is a branded comparison. The person who typed it already knew both names. The engine's job in that context is to compare two options the user identified. A favorable answer there does not represent a new buyer discovering the brand; it represents an already-aware prospect getting confirmation. All four branded-comparison prompts that named the brand against a competitor ran across all four engines, and this was the only one that returned an explicit recommendation.

Why mention rate and recommendation rate tell different stories

Mention rate is a low bar. An engine can cite a brand, list it among others, note its existence, or acknowledge it in passing without recommending it as the answer to a buyer's question. Recommendation requires the engine to actually propose the brand as the right choice for the buyer's need.

For this brand, the gap between the two numbers is wide. Mentioned in 28 percent of responses, recommended in 0.5 percent. The higher number reflects a brand that is known and gets acknowledged when named. The lower number reflects how often an engine actually steers a buyer toward it.

Neither number on its own tells you what you need to know. Mention rate without the branded/organic split tells you how often you show up; it does not tell you how often you show up for a buyer who has never heard of you. Recommendation rate without that same split can look better than it is, because branded comparison queries can produce recommendations even when organic discovery is zero.

The number that predicts new-customer acquisition

The signal that predicts whether AI search is sending you new customers is recommendation rate on organic prompts: how often an engine steers an undecided buyer your way when your name was never in the question. In this audit, that number was zero across all four engines and all 30 organic prompts.

A blended mention rate of 28 percent can read as modest but acceptable. A recommendation rate of 0.5 percent, sourced entirely from branded queries, says something different: AI engines know this brand exists, but when a buyer who does not know the name asks where to go, the engines send them somewhere else.

How we measure it

Every CitedWell audit separates organic and branded panels and reports both mention rate and recommendation rate for each prompt type, broken out by engine. The organic columns are where the new-customer signal lives. The branded columns show awareness and reputation. Both matter, and they need to be read as separate numbers rather than averaged together into a single visibility score.