AI visibility
How easily AI assistants can find, understand, and recommend a business when someone asks them a relevant question. It is built from readable structure, consistent facts, and trustworthy signals across the web, and it is measurable: ask the assistants the questions your customers ask and record who gets named. Visibility is the entry requirement for recommendation, and most businesses have never checked theirs.
A diner asks ChatGPT for a quiet dinner spot nearby; AI visibility decides whether your restaurant is one of the five names that come back.
GEO (generative engine optimisation)
The practice of making a website and its wider footprint easy for generative AI systems to read, trust, and cite, so the business appears inside AI-generated answers. Where classic SEO chases positions on a results page, GEO shapes the sources an assistant draws from when it writes its answer. The tactics overlap with SEO but the target is the answer, and there is no page two.
GEO work for a restaurant means the assistant can quote your menu, your hours, and your story instead of guessing from a directory listing.
AEO (answer engine optimisation)
A close sibling of GEO, focused on question-and-answer surfaces: making sure that when an engine answers a question directly, your content is the answer it uses. In practice AEO means publishing content structured around real questions, marked up so machines can lift it cleanly. The terms GEO and AEO are often used interchangeably, and for a restaurant the work is the same.
A page that answers "do you take walk-ins" in plain text with FAQ markup is AEO in its smallest useful form.
Answer engine
Any system that responds to a question with an answer rather than a list of links: ChatGPT, Perplexity, Claude, Google AI Overviews, and voice assistants all qualify. Answer engines compress the choosing step, so being absent from the answer means being absent from consideration entirely. They reward sources they can read with confidence and quietly skip the rest.
When a diner asks where to take a client for lunch, the answer engine hands back three names; there is no page of blue links to fall back on.
llms.txt
A plain text file at the root of a website (yoursite.com/llms.txt) that gives AI systems a short, structured summary: what the site is, who it serves, and which pages matter most. It is a young standard with thin adoption, which is exactly the opportunity: the businesses that publish one hand assistants a cleaner brief than their neighbours do.
A restaurant llms.txt names the cuisine, the city, the menu page, and the booking link in under a page of text.
Generate yours with the llms.txt Generator
Structured data
Machine-readable facts embedded in a web page, stating in a fixed vocabulary what prose only implies: this is a restaurant, here is the address, these are the hours, this is the price range. Assistants and search engines lean on it because it removes guesswork. A page can read beautifully to a human and say nothing to a machine; structured data closes that gap.
With structured data, "open until midnight on Fridays" is a fact an assistant can quote, not a sentence buried in a footer image.
JSON-LD
The format most structured data is written in: a small block of JSON placed in a script tag in the page head, invisible to visitors and legible to machines. Google recommends it and AI systems parse it readily. It lives apart from the visible HTML, so it can be added to almost any website without redesigning anything.
A restaurant's JSON-LD block can carry the full menu, section by section, dish by dish, prices included.
Restaurant schema
The schema.org type built for eating places, with hospitality-specific fields: servesCuisine, priceRange, acceptsReservations, opening hours, menu. Declaring the specific type (Restaurant, CafeOrCoffeeShop, BarOrPub, Bakery) tells machines exactly what kind of place this is. Many restaurant sites either have no schema or a generic type copied from a template, which reads as an unlabelled business.
The difference between LocalBusiness and Restaurant in your markup is the difference between "a business exists here" and "book a table here".
Build yours with the Restaurant Schema Generator
FAQPage schema
Markup that turns a visible list of questions and answers into structured data machines can lift verbatim. It suits the practical questions every hospitality business answers by phone all day: parking, dogs, allergies, groups, walk-ins. Marked up, your own wording becomes the answer an assistant gives, rather than a guess assembled from reviews.
A ten-question FAQ with FAQPage markup lets the assistant answer "can they seat twelve on a Saturday" in your words.
Share of voice
The percentage of relevant AI answers that name your business, measured across a set of real questions asked repeatedly. If assistants are asked forty local dining questions and your restaurant appears in six answers, your share of voice is fifteen percent. It is the visibility metric closest to revenue, because every mention is a diner-facing recommendation.
Two competitors at thirty percent share of voice while you sit at ten explains a quiet Tuesday better than the weather does.
AI crawlers
The bots AI companies use to read the web: GPTBot (OpenAI), ClaudeBot (Anthropic), PerplexityBot, Google-Extended, and others. They respect robots.txt, so one careless disallow rule can turn every assistant away at the door. Welcoming them costs nothing and is the single cheapest AI visibility decision a website makes.
A robots.txt that blocks GPTBot removes your restaurant from consideration before a single diner asks a single question.
Grounded search
When an AI assistant answers by fetching and reading live web sources rather than relying only on what it memorised in training. Grounded answers cite pages, favour sites that load fast and read cleanly, and reflect current facts like tonight's hours. It is why technical basics, working https, readable HTML, correct structured data, directly shape whether you appear.
A grounded assistant checking Sunday hours reads your site at that moment; if the hours only exist in an image, it moves on.
Memory-mode AI
The opposite of grounded search: the assistant answers from its training data without fetching anything live. Here your visibility depends on what the model absorbed months or years ago, mentions, reviews, articles, consistent facts across the web. You influence memory-mode answers slowly, by being clearly and consistently described everywhere, long before the question is asked.
An assistant answering offline still recommends the trattoria it read about in fifty consistent places; one-off bursts of promotion leave no trace.
Citation
When an AI answer names or links its source. Citations are the currency of grounded search: being cited means the assistant read your page and used it, which both sends visitors directly and reinforces your authority for future answers. Content earns citations by answering real questions plainly, close to the top of the page, on a site machines can parse.
When Perplexity answers "best natural wine bars in town" and links your list of what you pour, that citation is doing unpaid sales work.
Entity
The single, unambiguous "thing" a machine understands your business to be, distinct from any one page about it. Entities are built from consistency: the same name, address, and facts on your site, your map profile, directories, and press. When the details disagree, machines see several weak candidates instead of one strong one, and confidence collapses.
If your website says Osteria Bianchi, your map listing says Bianchi's, and a directory says Osteria da Bianchi, the machine sees three restaurants, none of them recommendable.
Knowledge graph
The structured database of entities and relationships that search engines and AI systems maintain: this restaurant, in this city, serving this cuisine, owned by this person, near this landmark. Getting into the graph correctly, via structured data, consistent listings, and authoritative mentions, means machines reason about you as a fact rather than a string of text.
A restaurant properly placed in the knowledge graph gets recommended for "dinner near the opera house" without ever writing those words.
E-E-A-T
Experience, Expertise, Authoritativeness, Trustworthiness: the qualities search systems look for when deciding which sources deserve to be believed, and by extension quoted in answers. For hospitality it translates concretely: a named owner, a real story, reviews that corroborate the claims, press mentions, and facts that check out everywhere. Machines reward being verifiably real.
A chef's bio with a name and history outranks anonymous marketing prose because it is evidence, and assistants prefer evidence.
Canonical
A tag that declares which address is the definitive version of a page when several addresses show the same content: with and without www, with tracking parameters, printer versions. Without it, crawlers split attention across duplicates and none of them accumulates full authority. It is one line in the page head and one of the cheapest fixes in the discipline.
When your menu lives at three URLs, the canonical tag tells every machine which one to trust, index, and cite.
Sitemap
An XML file listing every page a site wants crawlers to find, usually at /sitemap.xml and referenced from robots.txt. It does not improve any single page; it guarantees discovery, which matters for pages nothing links to yet. Most website platforms generate one automatically, and pointing robots.txt at it finishes the job.
Your seasonal menu page is only useful if machines know it exists; the sitemap is how they find out the day it goes live.