Definition
Recommendation rate is the percentage of tracked AI answers in which the engine actively recommends your brand, for example by naming it as a good choice, ranking it near the top or suggesting it for the user's situation. It is stricter than mention rate, which counts any appearance of the brand name, and different from citation rate, which counts answers that link to your pages.
The distinction matters because the three can diverge. A brand can be mentioned in a list of options to avoid, or as an example of an expensive tool. It can be cited as a source for a statistic without being recommended. And it can be recommended without a link. Reading the answer for intent is what separates a recommendation from a mention.
To calculate it, divide the number of answers that recommend your brand by the total answers sampled for a set of prompts. Classification is the hard part. Some teams label answers manually, others use a model with a fixed rubric, and either approach needs spot checks. Pair it with sentiment monitoring, since a recommendation can come with caveats.
Sample size limits how far to trust it. As a rough rule, the margin of error on a rate from n answers is about 1 divided by the square root of n, so 100 answers gives roughly plus or minus 10 percentage points. Answers also change between identical runs, which is why prompt sampling matters. Small week-to-week moves are often noise.
Recommendation rate sits alongside other KPIs such as AI visibility score and share of model. See our overview of AI search visibility KPIs for how they fit together, and use prompt tracking to collect the underlying answers.
Examples of Recommendation Rate
- A CRM brand is mentioned in 60% of answers to category prompts but recommended in only 25%, because many answers list it among tools with caveats.
- An answer says 'Tool X is often used for this, though it is expensive and many teams prefer Y'. A mention-only metric counts it, while a recommendation metric would not.
- A team labels 200 answers with a three-level rubric (recommended, mentioned neutrally, discouraged) and reports the first level as its recommendation rate.
- A brand sees its recommendation rate rise from 18% to 22% on 80 answers and holds off on reporting a win, since the change is within the margin of error.
