# AI for Restaurants: The Restaurant Marketing Agent Product page: https://www.winstondigitalmarketing.com/products/ai-for-restaurants/ Price: $47 USD // product / restaurants ## AI for restaurants: it will write "gluten free" about a dish it has never seen. A marketing agent for a restaurant that refuses to make an allergen or dietary claim, and refuses to write menu copy at all until your allergy line is in the profile. Twelve workflows, eight lines you do not cross, and a checker that finds the sourcing claims that stopped being true two years ago. **$47** [Buy the kit](/contact/?product=ai-for-restaurants) [Or read the free guide first](/playbooks/restaurant-local-seo/) Instant download after checkout. Fourteen day refund, no questions, including that you read it and decided it was not for you. > General information, not legal advice, and not food safety guidance. Food labelling, allergen disclosure, nutrition posting, alcohol advertising and delivery rules differ by country, state and city, and they change. Confirm anything here with your local health authority, your liquor authority and your own counsel. ## The claim this kit exists to prevent. Ask an unconstrained AI assistant for menu descriptions and it will characterize dishes dietarily, because every menu it has ever read does. It has not seen your kitchen. It does not know whether the fryer is shared, whether the same board is used for the bread and the almonds, whether the stock has chicken in it, whether your bakery supplier also handles nuts. A person with a severe allergy reads a menu description as a statement about whether the food will hurt them. That is a reasonable reading, not a legalistic one. Cross contact does not appear in a recipe and it is exactly what sends people to hospital. Your kitchen can make those claims, from knowledge of its own equipment and its own prep, and it takes responsibility for them. A model has neither the knowledge nor the responsibility, and no phrasing fixes it: "can be made gluten free" reads to somebody at risk as precisely the assurance they were looking for. So this agent writes the dish and its ingredients, plus your allergy line telling guests to speak to the team. And if your profile has no allergy line, it refuses to write menu copy at all until you give it one. That friction is deliberate. ## The quieter problem, which costs you more often. House made pasta that became a supplier product during a staffing crunch. Locally sourced from a farm you stopped using two years ago. Baked daily, on a Monday. These are the most common inaccuracies in restaurant marketing and none of them start as dishonesty. They start as true, and then the kitchen changes and the website does not, because in most restaurants nobody actually owns updating it. The profile asks what you can substantiate this week. The agent uses only that. The included checker finds the ones already live. And there is a workflow to run every time the kitchen changes something, which tells you every place that probably still says the old thing. A guest who notices one loose claim assumes the rest of the menu is equally loose, which is the real cost. ## Where it actually helps. - **Menu descriptions from real ingredient lists.** The highest-volume job, and it works well when the profile has actual ingredients in it. Given a dish name and nothing else it asks, which is the behavior you want, because a description invented around a name is how a restaurant advertises something the kitchen does not serve. - **Making the surfaces agree.** Your website, your Google listing, the delivery platforms and the reservation system disagree constantly, because four people update four systems and nobody owns the whole picture. The agent tells you what disagrees and which version a guest sees first. Dull, and the most common failure in the category. - **Google Business Profile.** More valuable than the website for a restaurant and it gets a fraction of the attention. Hours including the awkward ones, attributes, categories, posts. - **Private dining, catering and large groups.** These convert at far higher value than a table for two and they are usually one paragraph and a contact form. The page exists to answer the ten questions an event planner asks by email. - **Review replies at volume.** Restaurants get more reviews than almost any local business and reply to a fraction. Same job every time. ## Where it does not help, whatever anyone selling it says. - **Allergen and dietary claims.** By design, and it is the point of the product. Anything offering to generate a fully labelled menu in one pass is selling you the one automation in this business that can hurt somebody. - **Photography.** Your photographs sell the restaurant on the surfaces where the decision actually happens, more than any words will. For most restaurants one good shoot beats a year of copy. - **Winning the map result.** Proximity, categories, reviews, photographs and completeness. Content barely moves it, so the kit puts fixing the profile ahead of writing anything. - **The food.** Worth saying out loud. A well-marketed mediocre restaurant just fails faster, because more people find out. - **Telling you your alcohol rules.** They differ by state, platforms restrict alcohol independently of any law, and both change. The agent follows your profile and flags the rest for your liquor authority. ## It is not just prompts. **A menu checker.** Point it at your menu and website copy and it returns every allergen and dietary claim, every nutrition claim, and every sourcing or production claim, with a line number and a replacement. Nearly every restaurant finds at least one of each, and the second category is the one that surprises people. Python 3, no dependencies, no network, nothing leaves your machine. **The surfaces audit.** Paste in what your website says, what Google says and what a delivery platform says, and get back every disagreement about hours, items, prices and address, ranked by which one a guest is most likely to see first. Delivery listings are usually the most wrong and the most seen. **The change workflow.** Run it whenever the kitchen changes something and it tells you every place on your site, menu and profiles that probably still says the old thing, with the corrected wording for each. Ten minutes at the point of change beats finding it in a review eight months later. **A review rule for the hard case.** The agent will not draft a public reply to a review mentioning illness, an allergic reaction or a food safety concern. It flags it instead, because that needs a person first and a fast public reply is usually the wrong opening move. ## What it is not. Not software, not a course, not a list of restaurant AI tools. A folder of markdown files you read and then hand to an assistant you already pay for. It never touches your POS, your reservation data or your guest records. It drafts from the menu and the facts you put in the profile, and a person reads every draft. ## Refunds. Fourteen days, no questions, for any reason including that you read it and decided it was not for you. Email john@winstondigitalmarketing.com. ## Frequently asked questions. ### Can AI write menu descriptions? Yes, and well, on one condition: you give it the real ingredient list for each dish rather than a dish name. Given actual ingredients it writes what the dish is, how it is cooked and what it tastes like, in your voice, at whatever length you want. Given a name alone it will invent a preparation, which is how a restaurant ends up advertising something the kitchen does not serve. The other condition is that it must not characterize dishes dietarily, which is the subject of the next question. ### Why will it not write gluten free or vegan on a menu? Because a guest with a severe allergy reads a menu description as a statement about whether the food will hurt them, and an assistant has never seen your kitchen. It does not know whether the fryer is shared, whether the same board is used for the bread and the almonds, whether the stock has chicken in it, or whether your bakery supplier also handles nuts. Cross contact does not appear in a recipe and it is exactly what puts people in hospital. Your kitchen can make those claims from knowledge of its own equipment and takes responsibility for them. The hedged forms do not solve it either, because "can be made gluten free" reads to somebody at risk as the assurance they wanted. ### What should a restaurant never let AI write? Eight things. Allergen and dietary claims, including hedged ones and including vegan and vegetarian, which usually fail on a stock or a sauce base. Nutrition and health claims, including invented calorie counts. Sourcing and production claims like organic, locally sourced, grass fed and house made. Prices, availability or hours you did not supply. Review requests aimed only at guests who seemed happy, or offering anything in exchange. Alcohol content outside your own jurisdiction's rules. Statements of what a food rule requires. And anything that publishes without a person reading it. ### How do I find the claims on my menu that are no longer true? Run the included checker over your menu and website copy in one command and it flags every sourcing and production claim with a line number, alongside the allergen and nutrition ones. Then check each against what the kitchen actually does this week. House made is the one to be rigorous about, because it is true when written and stops being true when a cook leaves, the kitchen gets busy or a supplier gets cheaper, and nobody updates the website. The kit also includes a workflow to run at the moment something changes, which is what stops the problem recurring. ### Why does my website disagree with my delivery apps? Because four different people update four different systems and in most restaurants nobody owns the whole picture. It is the most common failure in the category and it is invisible from the inside, since you look at your own website and the guest looks at Google or a delivery app. The kit includes a workflow that takes what each surface currently says and returns every disagreement about hours, items, prices and address, ranked by which a guest is most likely to see first. Delivery listings are usually both the most out of date and the most viewed. ### Should I reply to a review about someone getting sick? Not with a drafted public reply as the first move, and this agent will not write one. It flags those reviews instead. A complaint about illness, an allergic reaction or food safety needs a person, needs to reach whoever handles that at your restaurant, and may need more than that before anything is said publicly. A fast public reply written to sound good is a bad opening move in a situation where what you say can matter later. Agree that path in advance, which is what week three of the rollout is for. ### What do I need to run this? An AI assistant you already pay for, or a free tier. The kit is plain markdown and works with Claude, ChatGPT, Gemini or Copilot. The included checker script needs Python 3, which is already on every Mac and most Linux machines. You also need your real ingredient lists and an allergy line, and the agent will not write menu copy without the second one. ## Related reading, free. This kit is built on guides already published here, which cost nothing. Read those first if you would rather not pay for anything. - local SEO for restaurants - GEO for restaurants - Google Business Profile optimization - the rest of the products ## AI for Restaurants The marketing agent, twelve workflows, the menu checker, and the eight lines you do not cross. $47. [Buy the kit](/contact/?product=ai-for-restaurants) [Read the free guide first](/playbooks/restaurant-local-seo/)