Having access to AI is brilliant. It's almost like having a team of dozens of people ready to search, sort, compare or write for you at the click of a button. But let's be honest: haven't we all got a bit too comfortable now that it's always within reach? We fire off quick questions, give it very little context and expect the tool to fill in everything we haven't told it.
Whether we've got slightly lazier is a topic for another article, but it's not the problem we're interested in here. The problem is that quick, overly broad questions leave us a long way from what AI can really do. If we ask it to “analyse my sales” or “make me a campaign for Tuesday”, it might give us something reasonable, sure, but it's working almost blind.
In a restaurant, thinking a dish through properly is almost as important as plating it. AI is much the same. The better it understands what kind of restaurant you are, what's going on, what you want to achieve and the limits it has to work within, the more useful the answer gets. So let's look at the part that tends to get lost because it's so easy to just open a chat and start typing.
Want the guide to hand?
Leave your details and download the full guide, with all the prompts and examples from this article.
AI for restaurants · 2026
The point of the guide isn't to hand you 50 prompts and leave you to it. Before the library, there's a fairly important section on how to work better with these tools without overcomplicating things.
We explain how to build a good prompt, how to separate your restaurant's fixed context from the information for a specific task, what's worth saving in tools like ChatGPT, Claude, Gemini or Perplexity, and which type of AI might make most sense for what you want to do.
Then come the 50 prompts, organised into sales and reports, costs and profitability, purchasing and inventory, delivery and digital menus, marketing, customers and reviews, operations, kitchen, growth and automations. Each example covers when to use it, what to prepare beforehand and what to check once you've got the answer in front of you.
And at the end there's a section that's perhaps less exciting, but very useful once you start working with AI regularly: a checklist for reviewing answers, another for preparing files, a permanent context sheet and a template for saving the prompts that have actually worked for you. Because finding a good way to do something and then losing it in a pile of 200 chats isn't much of a system either.
For a while, we've treated prompts like questions for an oracle, hunting for the one right question that will solve all our problems. As if finding the right words were enough to turn an AI from so-so to brilliant. In reality, it's far less mysterious than that.
A good prompt is a lot like giving someone on your team a good briefing. If you say “have a look at this and tell me what you'd do”, they'll have to fill in too many gaps. If you explain what kind of restaurant you run, what the problem is, what data they have in front of them, what they shouldn't assume and what result you need, the conversation changes completely.
The guide boils it down to six building blocks: context, goal, data, criteria, format and verification. Asking it to “analyse these reviews” is one thing. It's quite another to tell it you run two Italian restaurants, attach the last quarter's reviews, ask it to split the problems between front of house, kitchen and delivery, and limit its suggestions to changes that can be made in under 30 days without adding staff.
You don't need to turn every prompt into a three-page document. What matters is that the AI has the information a person would need to do that job well.
And here's another big step forward: not all the context has to be repeated every time. Your restaurant's concept, audience, sales channels, tone of voice and certain rules can be saved as permanent context. Then you only need to add what's happening today, which data you're using and what you want to solve. Less time explaining who you are in every chat, more time getting straight to the problem.
When we talk about AI, it's easy to end up thinking about text, images or fairly flashy stuff. But in a restaurant, it's probably most useful for tasks that are far less spectacular and far more frequent.
Reading a report nobody has time to read. Comparing this week with last week. Spotting a pattern across a hundred reviews. Checking where your margin is going. Sorting through issues from several shifts. Preparing a briefing before service. Turning a half-formed idea into a procedure someone else can follow.
That doesn't mean handing it the keys to the restaurant. An AI answer can sound completely sure of itself and still be wrong. When it comes to allergens, food safety, tax, employment law or major financial decisions, human review is still a must. The point isn't to let it decide for you, but to take some of the groundwork off your plate so you can make better decisions.
In the guide, we've put together 50 prompts across 10 areas of the restaurant. They're not there for you to copy without thinking. They're there to show you what changes when a request has context, data, limits and a clear output. Here are ten:
Analyse sales for [day] and compare them with [comparable day or average]. Summarise in five points: revenue, orders, average ticket, categories and channels. Only flag variations above [percentage] and separate facts from possible explanations. Finish with three questions the manager should look into tomorrow.
For each product, calculate the total cost, gross margin in euros and margin percentage. Explain the formulas and keep the original values. Flag any products below [target margin]. If a cost is missing, don't estimate it: leave the result incomplete and point out the missing data.
Group wastage by product, cause, shift and day. Calculate frequency, quantity and cost where the data exists. Identify the three most relevant patterns and suggest a small test for each one. Don't mistake more complete record-keeping for a real increase in waste.
Analyse which products appear together most often. Distinguish general popularity from genuine association: don't flag a combination just because both products sell well. Suggest five cross-selling opportunities and when in the ordering process it would make sense to show them.
Design five ways to increase demand for [time slot] without applying a blanket discount. Consider products with spare capacity, cross-selling, bookings, experiences or limited perks. For each option, state the cost, complexity, risk to margin and how to measure whether it works.
Group these reviews by theme: food, service, waiting times, atmosphere, price, delivery and other. For each theme, show volume, overall tone, short examples and how it has changed over time if the period allows. Distinguish recurring problems from one-off cases.
Prepare a three-minute verbal briefing for the [shift] service. Structure it as: situation, goal, changes, risks, responsibilities and final confirmation. Only include information that changes today's work. Finish with three quick questions to check everyone has understood.
Suggest ten dish ideas for [goal]. They must fit [concept], use [ingredients or season], be prepared with [equipment] and stay within [cost or complexity]. For each idea, explain the main element, technique, accompaniment and operational risk.
Create an initial list of areas to research in [city]. Use current sources and cite every relevant claim. Assess audience fit, concentration of restaurants and bars, competition, accessibility and demand signals. Don't invent rents or footfall. Finish with the data that needs to be validated locally.
Design the workflow for generating a [frequency] report from [source]. Define the data needed, transformations, checks, output format and situations that require human review. Include what should happen if data is missing or the schema changes.
None of these prompts contains magic words. What they have is a well-framed task. They tell the AI what to look at, what not to make up, which limits to respect and what kind of answer would actually be useful afterwards.
Pretty much anyone making decisions in a restaurant who feels they could get a lot more out of these tools.
If you run the business, you can use them to understand sales, costs or margins without spending hours sorting through data. If you look after operations, you can turn issues into actions, prepare briefings or document processes. If you're in marketing, you can analyse reviews, plan campaigns or work on content without ending up sounding like the rest of the internet. And in the kitchen, they can help you organise tests, review the menu or explore ideas within real limits.
There's also a more technical section for anyone who wants to get started with automations, connectors or MCP. Here the rule is even simpler: start small. Let it read first, then recommend, and only later take actions, always with clear permissions and someone reviewing what happens.
You don't need to set up a restaurant run by AI agents by next Monday. In fact, it's probably better if you don't.
Find one specific task that's eating up your time right now. A weekly report. Last month's reviews. A costings spreadsheet. The issues from the last few services. Give it context, attach the data it needs and ask for an answer you can check.
If it works, save the approach and reuse it. If it doesn't, check what information it was missing before deciding that “AI just isn't any good for this”.
Because in the end the idea is pretty simple: AI doesn't run your restaurant. But it can save you hours of repetitive work, help you organise information better and get you to better questions. The tool brings the speed. The judgement is still yours.
