Vegan, gluten-free, oat milk: how AI models actually read your café menu

Why AI often can't read your café menu at all

Most cafés keep their menu as a nicely designed PDF or a photo of the chalkboard posted to Instagram. That looks fine to a human. To a language model it's close to invisible. ChatGPT, Gemini, and Perplexity generate answers from text they can parse cleanly — HTML, structured data, plain descriptions. A photo of your chalkboard, or a menu that lives only inside an ordering app, simply doesn't surface in that process. The information exists, just not in a form a machine can use.

You may already have seen the result: a guest asks the AI about a vegan breakfast nearby and gets three other cafés recommended, even though your vegan bowl is better. Not because your food is worse — because the machine has nothing to read. AI visibility isn't built with beautiful photos. It's built with clear, readable text on your own website.

The good news: you don't need to become a developer. It's enough to get your menu into proper text once, and to name vegan, gluten-free, and lactose-free clearly enough that no model can miss them. What's obvious to a human from context has to be spelled out explicitly for a machine. That's what the rest of this guide covers.

How guests really search for your café today

Search behavior has shifted. People used to type "café downtown" into Google. Now the same person writes a full sentence into ChatGPT: "I'm gluten intolerant — where in Regensburg can I get coffee and cake in a cosy spot?" or "Which café near me has a vegan cappuccino with oat milk and Wi-Fi?" These questions are specific, conditional, and built around one particular need.

That's a real opportunity for your café. Someone who asks such a specific question is close to ready to buy — they're not browsing, they want exactly this. If the AI names your café because you've clearly labelled the gluten-free cake and the oat milk, you've won a guest that a competitor with an unreadable PDF menu never had a chance to reach.

The key is knowing what your guests actually ask. Pay attention at the counter: what comes up again and again? "Is the banana bread vegan?" "Do you have lactose-free milk?" "Does it contain nuts?" These are the same questions people now type into an AI assistant. If your website answers them in plain text, you become the source it quotes.

Vegan, gluten-free, lactose-free: spell out the terms, don't abbreviate

Language models read text remarkably well, but they don't tolerate ambiguity. If your menu only shows a small green leaf next to a dish, the AI can't tell whether that means "vegan," "vegetarian," or "organic." Spell it out. Instead of a symbol, put plain words behind the carrot cake: "vegan, lactose-free, gluten-free on request." Those few words are worth more to a machine than any icon.

Avoid shorthand that only makes sense to your staff. "VG," "GF," or "nL" on a printed menu might be fine at the counter, but to a model they're just noise. Always write "vegan," "gluten-free," and "lactose-free" in full, and add the synonyms guests actually use: "plant-based," "dairy-free," "wheat-free," "oat milk instead of cow's milk." The more natural phrasings you use, the more reliably an AI recognizes what you offer.

Be honest and precise about limitations, too. If your gluten-free cake is baked in a kitchen that also handles gluten, say so: "gluten-free recipe, not a dedicated gluten-free kitchen." That protects guests with allergies and makes you more credible as a source. Models trust precise, responsibly worded claims over vague, confident-sounding ones.

Getting the menu onto your website as real text

The most important technical fix is also the simplest: your menu needs to exist on your website as real text — not only as a PDF download, not as a photo. Build a dedicated menu page with headings like "Coffee & Drinks," "Cakes & Pastries," "Breakfast," and list each item underneath as a short readable paragraph: name, description, price, and dietary labels. This is the HTML text that language models actually read and quote.

Describe each item the way you'd explain it to a guest standing at the counter. Instead of just "Carrot Cake — 4.50," write: "Carrot cake, moist carrot cake with walnuts, vegan and lactose-free, gluten-free on request, 4.50." That one sentence answers three common guest questions at once and hands the AI facts it can use immediately. A photo next to it is nice for people; the text is what does the work for your visibility.

Keep the menu current. If your offering changes with the seasons, update the text online, not just the chalkboard in the shop. A pumpkin latte still listed in July looks unreliable to guests and to models alike. A menu that's clearly kept up to date signals the opposite: this is a source worth quoting.

Structured data: the layer nobody sees but machines read

Beyond the visible text, there's a layer guests never see but that helps search engines and AI systems substantially: structured data using the Schema.org format. For cafés, the relevant types are mainly "Restaurant" or "CafeOrCoffeeShop" and "Menu." With these, you record directly in your page's code — in a form machines can parse — that you're a café, where you're located, your opening hours, and which dishes carry which dietary attributes.

Schema.org even has a property called "suitableForDiet," with values like "VeganDiet," "GlutenFreeDiet," or "LowLactoseDiet." Marking your menu items up this way gives a machine an unambiguous, structured version of your offering to parse alongside your text. It won't guarantee a citation on its own — Google's own guidance says no special markup is required for AI Overviews — but it removes any doubt about what a dish actually is. If you'd rather not code it yourself, ask your web provider to add it; it's a small job.

Pair this with a clean, complete Google Business Profile. Fill in the same attributes Google offers for hospitality listings: "Vegan options," "Gluten-free options," Wi-Fi, outdoor seating. Several AI systems draw on this profile data. When your website, your schema markup, and your profile all say the same thing, you read as a reliable source.

Answer the typical guest questions directly

Language models like content built as question and answer, because that's the pattern they work in. Add a short FAQ section to your website that answers your guests' real questions in plain text. "Do you have oat milk?" — "Yes, barista oat milk is available at no extra charge on all coffee drinks." A block like that is essentially a ready-made answer an AI can quote almost word for word.

Pull these questions from real life at the counter. What are the handful of things guests ask most? For most cafés it's plant-based milk, gluten-free cake, allergens, seating, and dogs. Answer each one honestly and concretely. Skip vague marketing language like "large selection" and write what you actually have: "three vegan cakes daily, at least one of them gluten-free."

This pays off twice over. Your human guests find what they need faster, and the AI gets an answer it can quote directly. The key is that each answer stands on its own and still makes sense if a model pulls out just that one paragraph and reads it back to someone.

Consistency across all channels

Nothing confuses an AI more than contradictions. If your website says you have oat milk, your Google profile doesn't mention it, and an old review claims you only stock cow's milk, the model has no way to know what's actually true. When it's unsure, it will often just leave your café out of the recommendation. Your job is to tell the same story everywhere.

Go through every place your café shows up online, once, systematically: your website, Google Business Profile, Instagram bio, delivery apps, business directories. Check that the name, address, hours, and — above all — the dietary details match everywhere. That consistency is one of the strongest signals you can send an AI, because agreement across many sources reads as reliability.

Reviews reinforce this too. When guests mention your "best gluten-free banana bread" or "great oat milk coffee" in a review, it backs up what your own website already says. Encourage happy regulars to be specific in their reviews about what they enjoyed — that kind of genuine, concrete language is useful confirmation for AI systems.

Start in three steps: your practical plan

Start small, but start. Step one: get your full menu onto your website this week as real text, with vegan, gluten-free, and lactose-free spelled out behind every item they apply to. That single change already puts you ahead of most cafés still linking to a PDF.

Step two: build an FAQ section with the five questions guests ask most, answered honestly. Step three: ask your web provider to add Schema.org structured data, and keep your Google Business Profile updated with the right dietary attributes. After that, check regularly that every channel still tells the same story.

Finally, here's the simplest test: ask ChatGPT or Gemini yourself about a vegan cake or a gluten-free breakfast in your town, and see whether — and how — your café gets mentioned. Repeat this every few weeks. It's a direct way to see whether the work on your machine-readable menu is paying off, and where you still need to tighten things up.

Keep in mind that this work compounds. Every guest who finds you through an AI recommendation because you clearly labelled the oat milk and the gluten-free cake is a guest who would otherwise have ended up with a competitor. Very few cafés have made their menu machine-readable so far — which means you have a real head start available right now, for a few hours of work.