The Great Restaurant That AI Could Not See

More and more, people no longer open a search engine and scroll. They ask an assistant a question and take the answer it gives.

Imagine someone new to Edinburgh, hungry, asking an AI assistant for a good Indian restaurant nearby. They do not receive ten blue links to sift through. They receive a short, spoken-style reply naming perhaps three or four places, with a sentence on each. That short list is now the front door. For most people, it is the whole doorway.

This is a different thing from ranking well on Google. A search engine shows a long, ordered list of the open web, and a business that sits on the second page is at least still on the list. An AI assistant does not list the web. It names a handful of options and leaves everything else unsaid. There is no page two to be found on. A restaurant that is not named is not ranked low. It is simply absent from the conversation, as though it did not exist.

We call how often, and how prominently, a brand is surfaced when people ask assistants for suggestions its AI visibility. It is measured across the surfaces people actually use to ask: ChatGPT, Google AI Overviews, Google AI Mode, and Gemini. A business can have a strong real-world reputation and still be invisible here, because the assistant draws on what it can read, not on what regulars know.

This is a case study of one such business. The Prahna Indian Grill is a genuinely well-regarded independent restaurant in Corstorphine, Edinburgh, with awards to show for it. Yet to the assistants, it was nearly invisible. What follows is a measured account of why that happened, what was changed, and how far it moved.

The Prahna Indian Grill sits in Corstorphine, on the western edge of Edinburgh. It is an independent restaurant, not a chain, and by every ordinary measure it had already earned its standing.

Its reputation was not in doubt. The Asian Catering Federation named it Scotland's Best Asian Restaurant, and it holds a Tripadvisor Travellers' Choice award for 2025. Diners who found it tended to return. By the only metrics that mattered for most of the restaurant's life, Prahna was a success.

None of that, it turned out, was the same thing as being visible to artificial intelligence. When people asked an AI assistant where to eat Indian food in Edinburgh, Prahna was rarely part of the answer. Quality and AI visibility are separate properties. A restaurant can be widely loved by the people who know it and almost entirely absent from the systems that increasingly decide who gets discovered.

The April baseline

When the field was first audited in April, Prahna ranked 22nd in the field of Edinburgh Indian restaurants for AI visibility. Its profile was distinctive: a "Cult Favourite", rarely surfaced by the assistants, but recommended with real warmth on the rare occasions it did appear. The problem was not that AI disliked the place. It was that AI seldom saw it at all.

The gaps were specific. Prahna was nearly absent from Google AI Overviews. In roughly 15% of the mentions it did receive, the assistants misspelled the name as "Prana", splitting whatever recognition it had across two spellings. And its menus existed only as PDF files, a format AI reads poorly, so the very information a diner most wanted was effectively closed to the machines doing the recommending.

To see a restaurant the way an AI assistant does, we asked the assistants the questions a real diner asks, and watched which names came back.

We put the everyday requests people make — for a good curry house, somewhere to take the family, a particular dish or a particular part of the city — to the AI tools people now use to find places: ChatGPT, Google's AI Overviews and AI Mode, and Gemini. For each answer we noted which Edinburgh Indian restaurants were named, and how near the top they appeared.

Taken together, that gives a single picture of the field: how visible each restaurant is to AI, and its rank within the field, from most visible to least. We measured Prahna's rank against the whole field before the work began, and again three months later.

An AI assistant can only recommend what it can read and trust. The work was not a marketing trick. It was hygiene: making a good restaurant legible to the systems that now sit between it and its next customer.

The diagnosis pointed to plumbing, not reputation. Prahna was already well regarded by people; it was illegible to machines. So theprahna.co.uk was rebuilt around a single principle: every fact a diner might ask about should exist somewhere a machine can read it, in a form it can trust.

In practice that meant a small number of changes, each ordinary on its own.

The menus and the answers to common questions were published as structured web pages rather than as PDF files. PDFs are convenient for printing and poor for reading; an assistant scanning the open web struggles with them. The same information, set out as plain pages, became something the models could actually parse.

A short set of plain-language pages was added: an FAQ, and a handful of pages answering the specific intents a real diner brings, such as what the restaurant is known for and who it suits. These give an assistant clear, quotable material rather than leaving it to guess.

The brand-name signal was corrected. The site, its redirects and its structured data were aligned to say the name consistently, so the misreading as "Prana" had no foothold. This addressed the roughly 15% of baseline mentions that had used the wrong spelling.

AI crawlers were explicitly permitted to read the site. And every old link was preserved, so the standing the restaurant had built up over years — the trust signals that accumulate as other sites link to it — was carried forward rather than discarded.

None of these steps is novel. They are the things a well-built site should do anyway. What was new was treating AI assistants as a serious reader of the site, and building for that reader deliberately.

Three months on, Prahna had moved from #22 to #9 — across into the top 10, the goal we had set.

This was not a place or two within the same band. The restaurant climbed from the lower third of the field to its upper reaches. The table below shows where it now sits among Edinburgh's Indian restaurants.

The change was visible in two further ways. First, presence. At baseline Prahna appeared on only some of the four AI surfaces and was nearly absent from Google AI Overviews. It now appears across all four — ChatGPT, Google AI Overviews, Google AI Mode and Gemini. Second, prominence. When an assistant recommends the restaurant, it now tends to place it around #3 to #4 in the list rather than at the margins.

The most telling evidence

The clearest evidence of cause is also the simplest. The pages built during the intervention — the structured menu pages, the intent-specific pages and, notably, the new FAQ — now appear among the sources the assistants cite when they describe the restaurant. The machines are reading back the answers that were written for them.

The clearest sign the work mattered is that the assistants now cite the pages built for them — drawing on the answers that were written for them.

Citation is the most direct sign of all: the assistants are pointing at specific pages that did not exist three months earlier. The map below places Prahna among its peers — on how often it is surfaced, and how prominently it is recommended.

AI visibility is a new discipline, and a distinct one. It is not the same as a brand's reputation, its marketing spend, or even the quality of what it offers. It is a separate question: when someone asks an AI assistant for a suggestion, whether a brand's name comes up at all, and where in the list it sits.

The Prahna's case makes the point plainly. This was already a great, well-liked restaurant, recognised by the Asian Catering Federation as Scotland's Best Asian Restaurant and holding a Tripadvisor Travellers' Choice 2025. None of that was enough for AI to see it. At baseline it ranked #22, nearly absent from Google AI Overviews, its name often misspelled, its menus locked inside PDF files that machines read poorly.

Prahna's case points to machine-legibility rather than money or merit — though a single before-and-after cannot prove that on its own. What can be said plainly is this: the website was rebuilt so the assistants could read and trust it, the restaurant rose to #9, into the top 10, and it now appears across all four AI surfaces. The clearest evidence that the work mattered is direct — AI now cites the exact pages that were created.

Two observations follow from this case. The first is that AI visibility can be shaped deliberately, by making a brand easier for machines to understand. The second is that it can be measured and tracked over time, so a brand can see where it stands rather than guess.

About Renown

Renown is an AI-visibility intelligence company. We measure how brands appear when people ask AI assistants for recommendations, and we track that visibility continuously over time. Contact us at hello@tryrenown.com or visit tryrenown.com.