Since September 15, the visibility of food brands in generative AI responses has been tracked. LSA, in partnership with the startup Botrank.ai, is publishing a barometer dedicated to this topic, which will be updated quarterly.
What exactly does this barometer measure?
One hundred prompts, organized into four themes—drive, price, products, and promotions—selected from popular queries in Google Trends and the Keyword Planner. Each prompt was run daily for sixty days on ChatGPT, Gemini, and Google’s AI Overviews between June 26 and August 26, 2026, without any prior history or context, in order to replicate the conditions of a new search. A total of 18,000 responses were analyzed. Two metrics: the visibility rate—that is, the share of responses in which a brand is mentioned—and the share of voice among all brand mentions.
On the topic of price, two retailers—Lidl and E.Leclerc—are mentioned in more than 95% of the responses, with only one-tenth of a point separating them. Their ranking relative to each other matters less than their combined dominance over the rest: when it comes to price, nearly all responses mention these two names, and all other retailers share the remaining share of the market.
The value of this tool lies less in the scores themselves than in their consistency over time. There are already several AI-based visibility metrics in circulation, but each has its own prompts, models, and set of retailers: they aren’t comparable to one another, so no one can tell whether they’re improving or declining. By repeating the same method every quarter, LSA and Botrank.ai are creating the first point of comparison over time on this topic. A retailer that revamps its pages, content, or product data will be able to check the following quarter whether these changes were reflected in the AI’s responses.
Visibility is something you build; you can't buy it
The most insightful part of the study focuses on the sources the models draw from. For price-related queries, the most frequently used source is a retail chain comparison published by Que Choisir. More broadly, the domains most frequently referenced by the three search engines are Quechoisir.org, Facebook, Carrefour.fr, Reddit, and YouTube. Carrefour.fr appears on the list primarily due to its pages dedicated to delivery, which explains its presence in e-commerce-related topics. And Lidl.fr is the only retailer’s website to rank so highly on the topic of pricing, driven by a page where the brand documents its comparison method and regularly updates its product sample.
These examples all convey the same message: the content being indexed is content that clearly presents a method, data, and structure on a specific topic. Botrank.ai identifies three areas for improvement: the technical accessibility of pages, the accuracy and structure of the content, and the authority established through third-party sources. None of these three involves purchasing ad space.
This is the distinction we’ve been careful to maintain ever since the topic first arose: being mentioned by a model and purchasing ad space in a virtual assistant are two separate mechanisms. The appropriate framework for comparison is that of organic search and paid search. The introduction of ad inventory in virtual assistants adds another avenue; it does not replace any of the existing ones.
Finally, the barometer reminds us that authority is not limited to owned content. On the topic of freshness, a Reddit thread in which users discuss where they buy fruits and vegetables contributes to the responses generated. A model does not automatically prioritize an investigative report over a conversation among customers; it draws on what is available, readable, and referenced elsewhere.
What is the scope for which these scores are calculated?
One final point to note: The 100 prompts were drawn from popular search queries on Google Trends and the Keyword Planner, then categorized into "drive," "price," "products," and "promotions." The initial dataset therefore comes from the search engine, and the four selected themes are those used in retail comparisons: which store is the cheapest, and which has the best products.
This is the starting point available today, and it aligns with the areas in which retailers have always compared themselves. It’s simply worth keeping in mind when reading the scores: they measure visibility in these four areas, not across the full range of what a customer might ask a shopping assistant regarding groceries.
The next edition will reveal whether the rankings have shifted. What this initial analysis already shows is that a retailer’s visibility in a search assistant depends on a corpus of sources that is largely external, and that this corpus can be refined. If you’re making progress on these topics—whether from a retailer’s or a brand’s perspective—we’d be happy to discuss what the data reveals and what a retailer’s content makes intelligible to a model.
