US market research

If AI Cites Our Website, Will It Name Our Brand? A Reanalysis of 4,682 Responses

Published

This article was first published in Korean by Intrix Lab on October 5, 2026.

Usually, but not always. As of 2026-09-28, we reanalyzed 4,682 responses in which AI searched before answering questions that did not name a brand. Among responses that cited a brand's own website, model A named that brand in 75% and model B in 56%. Without a citation to the brand's website, the rates were just 0.4% and 6.3%. Citations are a gateway to recommendations, but even with a citation, the name was missing from 1 in 4 responses for model A and nearly half for model B.

What did we reanalyze?

A study released on arXiv in September 2026 divided the process of brand mentions in AI search into stages. It claimed that when two conditions coincide, a citation to the brand's own domain and an AI search query containing the brand name, the mention rate can rise to as high as 100%. We used responses already held by Intrix to check whether the same pattern appeared in Korean-language questions.

We made no new AI calls. We reread research responses collected from 2026-08-19 through 2026-09-22. The subjects were 6 businesses: a design platform, two law offices, a branding company, an online store, and Intrix. We excluded responses to questions containing a brand name and examined only responses that involved a search. We documented and fixed the analysis rules before opening the data.

Does a citation lead to a brand name mention?

Brand's own websiteModel AModel B
Cited39/52 (75%)30/54 (56%)
Not cited15/3,970 (0.4%)38/606 (6.3%)

The difference is large. In responses without a citation to the brand's website, the likelihood of the brand being named was in the single-digit percentages or lower for both models. This means that to get your name into an AI response, you first need a page that gets cited for that question.

Will AI search for our name?

It rarely did. For questions that did not name a brand, model A independently added a brand name to its search query in 26 of 4,022 responses (0.6%). Model B did so in 0 of 660 responses. The study's proposed "path combining a citation to the brand's own website and a branded search query" appeared in only 11 cases for model A. The brand was named in 10 of those 11 cases, but the sample fell below our predefined minimum of 20, so we recorded the result as "not replicated."

That makes it difficult to wait for AI to seek out our name. For questions from customers who do not know the name, a page answering their question must first be cited to create an opportunity for the name to be mentioned.

When does AI cite a site but leave out its name?

One law office's pages were cited 40 times across the two models, but its name appeared only 9 times. The remaining responses used the office's published fee information as evidence while referring to it only as something like "an example published by a local office." AI took the information from the page without conveying whose information it was.

The next checks are whether the business name is clearly stated in the page body and title, and whether company information is readable through structured data. Earning citations and earning name mentions need separate attention.

What happens if citations and mentions are counted together?

Performance is overstated. Model A inserts citation links into the response body. If the domain text inside those links is also counted as a brand mention, the mention rate among cited responses becomes 92% (48/52). Counting only cases where the name was actually stated gives 75%. When reviewing an AI visibility report, it is safer to first check whether citations and name mentions are counted separately.

To separately check which pages AI cites and whose names it mentions in your industry, you can start with a First Measurement.

How far can we trust these results?

The subjects were 6 businesses, and the responses came from Korean-language questions submitted through APIs. They cannot be assumed to match the consumer app experience, and we do not generalize that the same effect size will appear in other industries. Citations and mentions moving together shows an association, not causation. AI may have already known the brand and therefore both cited it and named it.

We did not test another claim from the study: the effect of prior exposure history. Model B mostly answered without searching, leaving a smaller set of 660 responses that involved a search. We recorded two definitions corrected after the analysis in the preregistration document.

Frequently asked questions (FAQ)

Q. If our website is cited, can we count that as a recommendation?
No. Among responses with citations, the name was missing from 1 in 4 for model A and nearly half for model B. Citations and name mentions need to be counted separately.

Q. Can we make AI search for our name?
That was rare in this dataset. For questions that did not name a brand, model A added the brand name to its search query in 0.6% of responses, while model B did so in 0 cases. Getting a page cited because it answers the question is the more realistic path.

Q. What happened in responses without a search?
Model A never named a target brand in 988 responses. Model B named one in 281 of 2,634 responses (10.7%). This means model B sometimes recalled businesses from memory without searching.

← Back to insights