Where should we eat tonight is now an AI question, asked with context no search box ever saw: the occasion, the diet, the group, the mood. The assistant answers with named restaurants and reasons. Here is how it builds that list.
The moments are occasions: the anniversary that needs candlelight, the client dinner that needs quiet, the birthday group of twelve, the visitor who wants what this city does best, the celiac guest who needs certainty. Each occasion phrase pulls different evidence, and a restaurant can own date night answers while never appearing for group dinners two blocks away.
For dining recommendations, assistants lean on review platforms with volume and recency, local food press and best-of lists, community threads where locals argue honestly, and restaurant sites with parseable facts: menus as text, dietary notes, private-dining details, hours that match everywhere. PDF menus are invisible to most of this machinery; a text menu with marked gluten-free items is quotable evidence.
The restaurants that keep getting named are legible to machines: text menus, consistent hours, dietary clarity, plus a review base whose language matches the occasions they win and appearances in the local lists assistants keep citing. Neighborhood spots outrank famous names for specific occasions constantly, because the evidence for the occasion is theirs.
The test costs nothing: ask ChatGPT, Gemini, Claude and Perplexity the questions above, phrased the way a real person would say them, and write down who gets named and what sources appear. Run each question twice on different days, since answers vary and patterns matter more than single runs.
For operators, the fixes are cheap: menus as text with dietary marks, identical facts across Google, Yelp and the site, and cultivation of the local press and lists your city's answers cite. For restaurants that want the systematic version, the free Aethon audit runs your real buyer questions across all four assistants with screenshots, and the restaurant playbook turns the gaps into a work plan.
By matching the occasion described to verifiable evidence: reviews with recency and volume, local food press and lists, community threads, and restaurant sites with text menus and clear facts. Occasion-matching language in reviews weighs heavily.
No. Answer placement is earned through the sources assistants read. Ads are separate and labeled where they exist at all.
Because for the specific occasion asked, quiet, gluten-free, group of twelve, its reviews and local coverage carry exactly that language, and the assistant matches evidence to request rather than fame to request.
Text menus with dietary marks, identical hours and facts everywhere, and steady review flow. Then check the four assistants monthly on your occasions. The free Aethon audit shows which occasions you currently win.