You Can Win the Mention and Still Lose the Buyer

Imagine someone asks an AI assistant, “What running watch should I buy?”

The answer mentions Garmin, Coros, and Polar. All three brands appeared, so all three could technically count that response as a visibility win.

But then the assistant explains that Coros is the best choice for a trail runner who cares about multi-day battery life.

Now the answer looks different. Garmin and Polar were mentioned. Coros won the recommendation.

That distinction matters because people are increasingly using AI to narrow their options before visiting a company’s website. McKinsey found that half of surveyed consumers use AI-powered search, while 44% of those users consider it their primary and preferred source of information. McKinsey also projects that $750 billion in U.S. consumer spending could flow through AI-powered search by 2028.

Adobe is seeing the shift in actual website traffic. Its analysis of more than one trillion visits to U.S. retail sites found that traffic from generative AI sources increased 1,200% between July 2024 and February 2025. In Adobe’s accompanying consumer survey, 47% of people who used generative AI for shopping said they used it for product recommendations.

The numbers are still developing, but the direction is becoming difficult to ignore: AI is starting to influence which companies enter a buyer’s consideration set.

A mention can mean several different things

When marketers talk about AI visibility, they often reduce it to one question: Did our company appear?

That is useful, but it leaves out what happened after the company appeared.

There are at least four different outcomes to measure:

• Recognition: Does the AI know what the company does?

• Inclusion: Does the company appear when someone asks about its category?

• Positioning: How does the AI describe the company’s strengths, weaknesses, and ideal use cases?

• Recommendation: Does the AI actually suggest the company for the buyer’s specific situation?

A company can perform well at the first two stages and still lose at the last two. It might appear regularly but be framed as too expensive, too limited, or better suited to a different type of customer. A competitor may be mentioned less often but recommended more decisively when it does appear.

A simple mention count would miss that.

Recommendations appear to carry more weight

A 2026 preprint from Scrunch AI examined the difference between neutral brand mentions and genuine recommendations across ChatGPT, Claude, and Gemini conversations.

The researchers connected opt-in conversation data with subsequent online behavior. Among users with no recent observed engagement with a brand, an AI recommendation was followed by a 4.3-percentage-point increase in same-name Google searches, a 2.4-point increase in visits to the brand’s website, and a 1-point increase in visits to brand-specific retailer pages.

Neutral name-drops were followed by much smaller changes: 1.8 points for search, 1.1 points for brand-site visits, and 0.3 points for retailer visits.

In other words, a genuine recommendation was associated with two to three times more downstream activity than an incidental mention.

There are important limitations. The paper is a preprint and has not yet been peer-reviewed. The study was observational, so it does not prove that the AI response caused the behavior. It also measured searches and website visits, not completed purchases.

Still, the distinction is important. Being named and being recommended are not interchangeable outcomes.

Context determines who wins

AI recommendations are not universal rankings. They depend on what the buyer asks, the needs included in the question, the category, the location, and the platform producing the answer.

Research published at EMNLP also found evidence of brand bias in large language models. Global brands were more frequently associated with positive attributes, while country and economic context affected the types of brands models recommended.

That means a company cannot understand its position by running one broad prompt or collecting a handful of favorable screenshots. It needs to test the situations in which real buyers would actually evaluate and compare their options.

This is a central part of how we think about measurement at Signal Harbor. We independently examine whether companies are included, how they are described, when they are recommended, how competitors are positioned, and whether those patterns remain consistent across repeated tests.

The goal is not simply to prove that a company appeared.

It is to understand whether the company is present when the decision is being made and whether the AI gives the buyer a reason to choose it.

You can win the mention and still lose the buyer.

Sebastian Miller
Co-Founder, Signal Harbor
signalharborconsulting.com
sebastian.miller@signalharborconsulting.com
402-306-2213

Works Cited

Iannelli, Michael, and Alan Ai. “From Prompt to Purchase: How AI Brand Recommendations Move Consumers on the Open Web.” arXiv, 9 June 2026, arxiv.org/abs/2606.10907.

Kamruzzaman, Mahammed, Hieu Minh Nguyen, and Gene Louis Kim. “‘Global Is Good, Local Is Bad?’: Understanding Brand Bias in LLMs.” Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, Association for Computational Linguistics, 2024, pp. 12695 to 12702. doi.org/10.18653/v1/2024.emnlp-main.707.

Pandya, Vivek. “Adobe Analytics: Traffic to U.S. Retail Websites from Generative AI Sources Jumps 1,200 Percent.” Adobe Blog, 17 Mar. 2025, blog.adobe.com/en/publish/2025/03/17/adobe-analytics-traffic-to-us-retail-websites-from-generative-ai-sources-jumps-1200-percent.

Silliman, Elizabeth, Julien Boudet, and Kelsey Robinson. “New Front Door to the Internet: Winning in the Age of AI Search.” McKinsey & Company, 16 Oct. 2025, mckinsey.com/capabilities/growth-marketing-and-sales/our-insights/new-front-door-to-the-internet-winning-in-the-age-of-ai-search.