Popular brands earned more citations in an AI search audit
A November preprint tested Italian commercial prompts across three AI APIs and examined how brand popularity related to mentions and citations.
A November 5 preprint on generative search reports that brands in the highest popularity quintile were cited 1.4 to 2.8 times as frequently as the average brand, depending on the engine. The study used Italian commercial queries and measured popularity using Italian Google search volume for brand names.
The audit separated mentions from citations
The authors collected 19,200 answers from GPT-5.5, Gemini 3.6 Flash, and Perplexity Sonar. They used 80 unbranded commercial prompts across insurance, travel, fashion and beauty, and sport. They varied the buyer context, ran prompts repeatedly, and tested configurations with search enabled or disabled.
A mention meant that the brand appeared in the answer text. A citation required its official website to appear among the sources. The audit included 1,176 brands, but the citation analysis covered the 683 with known official domains. A comparison article that names a brand can inform an answer without the brand’s own site appearing among the citations.
The page experiment tested a different outcome
In a separate experiment, the authors used 24 prompts and supplied five pages per task. They changed the brand or product names and the wording on one page, then asked the models to rank the supplied pages with web search disabled. That setup helps examine how names and wording affect the ranking of a fixed set. It cannot show whether a rewritten page would enter the candidate set in a live search.
What to record in a publisher audit
Use a fixed set of unbranded questions for one market and product category. Record whether the engine actually searched, which brands it mentioned, and which official sites it cited. When the API exposes the first search query, note whether the query already contains a brand name. Repeat the prompts before deciding whether a pattern holds.
The findings apply to the study’s prompts, model configurations, and popularity measure. The AI visibility measurement crosswalk explains why mentions and citations belong in separate columns. The citation stability analysis provides context for repeated observations.
GrowthSpree’s B2B software study measures domain-level referring domains, a different variable from brand popularity. Its paired queries also change buyer context. Our recount of the AI Overview source data shows how the result changes when each topic gets equal weight.
Keep learning
Continue this topic
Audit citations
Locate the failing answer layer
Compare evidence
Separate mentions, citations and visits
Measure referrals
Check AI search visits in GA4
Research
Ask a question or join the discussion