Your restaurant has real awards and accolades, yet in AI answers a competitor with no title at all shows up first. It's not a mistake: awards sitting still on a website don't speak as loudly as five hundred monthly mentions across Instagram, forums and reviews. The AI doesn't read your trophy cabinet — it measures the volume of conversation around your name, and that competitor has built a steady buzz you don't have. Stimulating organic mentions systematically is possible without an advertising budget, and the results in AI answers arrive in measurable timeframes.
Single mentions don’t count. Aggregate volume does. A brand that gets 500 mentions/month across Twitter/LinkedIn/Instagram is perceived by AI as a salient entity, even without “quality backlinks”.
It’s something I often repeat to my fine-dining restaurateur clients, because it’s the sector where the asymmetry shows up most clearly: there’s the restaurant with two historic accolades and zero social conversation, and there’s the one that every week ends up in hundreds of stories, Instagram reviews, LinkedIn posts from chefs and sommeliers. When you ask ChatGPT or Perplexity “best seafood restaurants in Liguria”, it’s the second one that comes up.
In this article I explain why, what the research says about entity salience in discourse, and how I reconstructed the pattern by reverse engineering five Michelin-starred Ligurian restaurants.
What aggregating mentions means for an AI model
When a model like Claude or Gemini is trained, it doesn’t read your social posts one at a time to decide “okay, this brand deserves it”. It reasons in terms of mention clusters — groups of references to the same entity scattered across the corpus. The larger the cluster, the more salient the brand becomes, meaning “notable”, memorable, useful to cite when the user asks a relevant question.
The link between mention cluster size and entity salience has been measured explicitly.
“This is easy to confirm: a significant and strong Pearson correlation exists between cluster size and the entity salience score, with 𝑟2 = 0.4505, 𝑝 < 0.00001.”
Translated: the size of an entity’s mention cluster correlates strongly and in a statistically robust way with the salience score that entity receives in discourse. It’s not a marginal effect, it’s one of the heaviest variables.
The operational consequence for your restaurant, your winery, your practice is simple: every time someone names you — in a post, a comment, a story, a written review — you add a brick to the cluster. The single brick is invisible. Two hundred bricks a month become a wall the AI recognizes.
Why the mention beats the backlink in the AI economy
In the previous articles in this series I told you about how the backlink works as a citation proxy and how the weight of an implicit reference factors into the authority the AI perceives. Here we’re one step beyond: a social mention almost never has a link, yet it counts all the same because the model works on text, not on a link graph.
A tweet that says “dinner last night at [restaurant name] in Imperia, the red prawn crudo was to die for” is a textual occurrence of your entity next to strong context words: city, dish, judgment. Multiply it by five hundred a month and you enter the conversation corpus of your sector.
A post with 50,000 likes is memorable to you, but the AI model prefers 500 posts with 100 likes spread over six months.
The reverse engineering I did on Michelin-starred Ligurian restaurants
I told you the fine-dining sector is a good test bed. I took five Michelin-starred seafood restaurants in Liguria — I’m leaving out the names out of respect, but they’re establishments between Imperia, Sanremo, Genoa and the Cinque Terre — and for each I measured two things:
- the monthly volume of textual social mentions (public Instagram, LinkedIn, X posts, plus written reviews on Google and TripAdvisor) over a 60-day window
- the number of spontaneous citations when I queried Perplexity, ChatGPT and Gemini with 12 queries like “best seafood restaurant in Imperia”, “Michelin-starred restaurants in Liguria seafood crudo”, “where to eat Ligurian fine-dining seafood”
I compared the result by hand, it’s an indicative test, not a study. A tiny sample — five restaurants, 36 total queries — but the pattern was so clean it’s worth telling: the two restaurants with a monthly mention volume above 400 showed up in 9-10 answers out of 12, while the two with fewer than 120 mentions showed up in 2 answers out of 12 or in none. The intermediate one, around 250 mentions/month, came up in 5 out of 12.
The Michelin star was not the discriminating variable: all five had it. The volume of social conversation was.
A sharing template for customers: a small card on the table saying “if you feel like it, tag us — @restaurantname #restaurantnameimperia” works better than you’d think.
What the research says about the signal you’re sending
Salience research doesn’t stop the reasoning at the single sentence. It goes further.
“Taken together with the other findings above, this suggests that much more attention should be devoted to discourse structure and phenomena above the sentence level in the study of formal linguistic markers of entity salience.”
In plain terms: the signals that make an entity salient sit above the single sentence — in the structure of the discourse, in recurrence, in the distribution of mentions across the text and over time.
It follows that your goal isn’t the perfect tweet, it’s sustained density. Three isolated viral posts are worth less than two hundred steady average mentions a month. That’s why the Ligurian restaurants that invest in the chef’s daily storytelling, in customer tags, in behind-the-scenes content, win on AI visibility over those that post the photo of the dish once a week.
If you’re interested in the foundation all of this rests on, in my articles I explained how an entity gets recognized through NER and how an entry in Google’s Knowledge Graph raises the starting baseline. Social mention aggregation is the layer that sits on top of entity recognition and feeds it with frequency.
The test you can run in 20 minutes on your restaurant
No paid tools, no dashboard. An honest entry-level audit — the real analysis requires professional tools, but to figure out where you stand these steps are enough.
- Open Perplexity and run 6 queries on your sector + city: “best seafood restaurant [your city]”, “fine-dining restaurants [your province]”, “where to eat raw seafood [coast/region]”. Count how many times you come up.
- Open Google Trends and compare your restaurant’s name with two direct competitors over the last 12 months. If you’re flat at zero and they aren’t, you have a conversation problem, not a kitchen problem.
- Run a manual search on Instagram with your location’s tag + “restaurant” + “seafood”. Count how many of the first 30 posts mention you. Binary threshold: fewer than 3 out of 30 = you’re outside the local conversation.
These are first-level checks. They tell you whether you’re inside or outside the cluster, not how deep inside.
The mistakes I see most often
Thinking it’s enough to post yourself. Your post from your official account counts, but in the AI corpus it weighs far less than a hundred different mentions made by customers, food bloggers, local journalists, sommeliers. Your voice is one. The chorus is a hundred.
Never asking for the tag. In Liguria there are restaurants that hand the customer the dish without a single prompt to tag, and others that print an elegant card with the Instagram handle on the menu. The second group gets mentioned 4-5 times more.
Betting on the viral post. A post with 50,000 likes is memorable to you, but the AI model prefers 500 posts with 100 likes spread over six months. The research says it: it’s the structure of discourse above the single sentence that generates salience.
Ignoring long written reviews. A 300-word Google review with the name of the dish, the sommelier, the experience is gold in the corpus. A star with no text is noise.
What to actually do for your restaurant
- A sharing template for customers: a small card on the table saying “if you feel like it, tag us — @restaurantname #restaurantnameimperia” works better than you’d think.
- A unique brand hashtag, short, memorable. Not “seafoodrestaurant” — too generic — but something that ties your name to a dish or a concept.
- The chef narrated in the first person on Instagram and LinkedIn: the chef telling the story of the fish market in the morning is a continuous generator of mentions from those who follow him in the sector.
- Collaborations with local Ligurian wineries and producers: every time the oil or the wine on your menu gets named alongside you, you build a cluster of related entities.
- Monthly events: a four-hands dinner, a book launch, a tasting. The event entity generates mentions with date, place and name, which are very strong anchors in the corpus.
None of this is a magic factor. It’s not enough on its own. It adds up to the technical structure of the site, to your work on E-E-A-T for AI, to editorial consistency. But it’s the lever that most directly increases the volume of the cluster the research talks about.
The thread: from the mention to the AI answer
The point of all this stays the same: to be cited when a potential customer asks ChatGPT or Perplexity “where can I eat fine-dining seafood in Liguria tonight”. You don’t get there with an authoritative backlink, you get there by building — over time — a volume of conversation the model can’t ignore when it generates the answer.
In the following articles in this series I’ll talk to you about how to measure the editorial quality of mentions, how to work with food bloggers and local outlets without paying them, and how to bring the same logic to the trade press.
