AI in restaurants: where it earns its keep today

A restaurant owner asks an AI tool to write descriptions for the new spring menu. In seconds, it produces twelve elegant paragraphs. They read beautifully, and three of them describe ingredients the dishes do not contain, including a nut garnish on a dish the kitchen serves nut-free. The owner catches it, this time. The descriptions were fluent, confident and wrong in exactly the place where being wrong matters most.

That is the shape of AI in restaurants today. It is genuinely useful for some jobs: drafting, summarising, suggesting, forecasting a first estimate. It is unreliable for others, in ways that are hard to spot because the results look so convincing. The venues that get value from it use it where a person who knows the venue can check the result, and keep it away from decisions where a confident mistake could hurt a guest.

This guide covers the three places AI earns its keep in a restaurant now, the two places it quietly fails and the simple rule that keeps it useful. Every venue and figure in it is an example.

In this guide, AI means the tools that have become widely available in the last few years: systems that write, summarise and answer questions in ordinary language, and systems that find patterns in data to make predictions. Much of the recent excitement is about generative AI, which produces text, images and other content from a short instruction.

These tools are good at producing plausible results quickly. They are not good at knowing whether a result is true for your particular venue. That gap explains most of where they help and where they fail.

The single most useful rule for AI in restaurants is to use it where the result can be checked by someone who knows the venue, before it reaches a guest or drives a decision. A draft menu description checked by the chef, a forecast compared with the manager's own judgement, a rota draft adjusted by the person who knows the team.

Where a result cannot be checked, or will reach guests directly without a person in between, AI is a risk. The more confident and fluent its output, the easier it is to trust without checking, and the more harm a mistake can do.

The first place AI earns its keep is writing. Drafting menu descriptions, social media posts, job adverts, emails to guests, replies to reviews, procedures and notes for the team: AI tools can produce a decent first draft in seconds, which a person then edits.

The value is in the speed of the first draft, not in the final text. A manager who used to spend an hour writing a job advert can spend ten minutes editing a draft. The editing is essential: every fact, every price, every allergen and every promise must be checked.

Some writing must always be checked line by line: anything about allergens, ingredients, prices, opening times, policies or legal matters. AI tools will fill gaps with plausible guesses, and a plausible guess about an allergen is dangerous.

Make it a rule that allergen and ingredient information comes only from the venue's own records, never from an AI draft. Let AI help with the words around the facts, and never with the facts themselves.

The second place AI helps is forecasting. AI forecasting tools look at past covers, bookings, weather, events and seasons, and produce an estimate of how busy the coming days will be. For a venue that forecasts by gut feel, a reasonable first estimate is useful.

AI forecasting works best as a starting point that a manager adjusts, not as an answer to follow blindly. The manager knows about the local event, the new competitor or the regular's party of twenty that the data does not show.

Demand forecasting does not need AI to be useful. A simple method, last year's matching weeks adjusted for this year's bookings and known events, gets most venues most of the way. AI tools can refine it, but they are not required to start.

Our guide to demand forecasting without a data scientist covers the simple method, and our guide to a covers forecast your kitchen will believe covers how to turn it into a plan the kitchen trusts.

Check any forecast against what actually happens. Each week, compare the forecast covers with the actual covers, service by service. A tool whose forecasts are consistently close is worth trusting more; one whose forecasts are often far off needs adjusting or replacing.

Keep the comparison simple: a note each week of forecast against actual, with a word on any big miss. After a few months, you will know how far to rely on the forecast and when to override it.

The third place AI helps is drafting rotas. Given the forecast, the team's availability and the rules, such as minimum rest between shifts, some tools can produce a draft rota quickly, which the manager then adjusts.

The draft saves time; the adjustment keeps it fair. The manager knows who needs a particular day off, who works well together and who has carried the weekends lately. A rota published straight from a tool, without that judgement, can undo a team's trust in a week.

Rotas affect people's lives, and fairness is something the team watches closely. AI can help with the arithmetic of covering the forecast with the right number of people; it cannot judge what feels fair to a team.

Use the draft as a starting point, and make the final decisions yourself. Our guide to a staff rota that survives first contact covers how to build one the team can rely on.

AI tools are good at summarising large amounts of text, which makes them useful for reading guest feedback in bulk. A month of comments can be summarised into the main themes, praise and complaints, in seconds.

Check the summary against the comments themselves, at least now and then. Summaries can miss a serious complaint buried among the praise, or exaggerate a theme that appeared only twice. Use the summary to know where to look, then read the comments that matter.

AI tools can also help with ideas: names for a new dish, ways to describe a set menu, themes for a seasonal event, questions to ask in an interview. The value is in having more starting points than one person would think of alone.

Treat the ideas as raw material. The chef decides the dish; the manager decides the event; the person who knows the venue chooses what fits. AI is a useful colleague in a brainstorm and a poor one in a decision.

The first place AI quietly fails is in confident mistakes. AI tools produce fluent answers whether or not they are right, and they rarely say when they are guessing. A wrong ingredient, a made-up regulation, an invented statistic or a misremembered price all look just as convincing as the truth.

These mistakes are dangerous because they are hard to spot. A person reading a fluent paragraph assumes it is correct. The fix is not to trust less in general, but to check every fact that matters, every time.

The second place AI fails is talking to guests without a person in between. A chatbot on the booking page or an automatic reply to reviews can make promises the venue cannot keep, give wrong information about allergens or opening times, or respond to an upset guest in a tone that makes things worse.

Guest conversations need judgement and care. Keep a person between AI and the guest: let AI draft a reply, and have someone read and edit it before it is sent. A wrong answer sent automatically to a guest is far harder to undo than a slow answer.

AI tools can now create images of food that look like photographs. For a restaurant, using them is a risk: a picture of a dish the kitchen does not make, or one that looks better than the real plate, sets expectations the venue cannot meet.

Use real photographs of your own food, your own room and your own team. Guests trust what they can recognise when they arrive, and an image that turns out to be invented damages that trust quickly.

Some tools offer to suggest prices, or to change them automatically with demand. Pricing a menu needs judgement about guests, competitors, costs and the venue's reputation, and changing prices automatically can feel unfair to guests who notice.

Use any pricing suggestion as one input among several, checked against your costs and your guests. Our view is that a menu's prices should be decided by people who can explain them to a guest.

Restaurant automation, systems that do routine tasks without someone pressing a button, is often confused with AI. Booking reminders sent at a set time, an email to guests who have not visited in two months and a report produced every morning are automation, and they are useful, reliable and checkable because they follow simple rules.

AI is different: it makes judgements rather than following rules. Use simple automation wherever a task can be described in a sentence. Use AI where judgement is needed and a person can check it.

Be careful about what guest data goes into AI tools. Pasting a guest list, booking notes or feedback into a general-purpose AI tool may send personal information to a third party, which data protection law may not allow.

Check what each tool does with the information you give it, and keep personal guest data out of tools that do not protect it properly. When in doubt, describe the situation without names.

Some of the most useful AI help is in the office rather than on the floor: drafting an email to a supplier, summarising a long contract into its main points, turning notes from a meeting into a short list of actions, or checking a letter for tone.

Each is checkable by the person who asked for it, and each saves a little time. Keep the same rule as everywhere else: read the result before acting on it, and never rely on an AI summary of a contract instead of reading the parts that matter.

Decide who checks each kind of AI output. The chef checks anything about dishes and ingredients; the manager checks guest-facing writing; the owner checks anything about money, contracts or the law. A draft that nobody is responsible for checking is a draft that will eventually go out wrong.

Write the responsibilities down with the team's AI rules. Clear ownership is what turns a risky tool into a useful one.

Try any AI tool as a small trial before committing to it: a month, one task, a clear measure of whether it helped. Did it save time? Did its results need much correction? Would you miss it if it went?

Many tools look impressive in a demonstration and save little in practice. A month's trial on a real task tells you which kind each one is, at very little cost.

AI tools are often sold as add-ons with their own subscriptions or usage fees. Before paying, ask what problem the tool solves, how much time it saves each week and whether a simpler method would do the same job.

A writing tool that saves a manager two hours a week may be worth it. A forecasting tool that produces the same answer as last year's figures plus bookings may not.

AI is the loudest of the current restaurant technology trends, and the one most likely to be overpromised. Some of what is promised will arrive and become ordinary; some will not. The venues that do best adopt the useful parts early and carefully, and ignore the rest.

Our guide to restaurant technology trends worth your attention sets AI alongside the other developments that matter, and those that are noise for now.

If the team uses AI tools, set a few clear rules: what the tools may be used for, what must always be checked and what must never be put into them, such as guest details or allergen information. Write the rules on one page and share them.

Most problems with AI at work come from people using it in ways nobody discussed. A short conversation and a one-page rule set prevent most of them.

Guests value the human voice of a venue: a reply to a review that sounds like the owner, a menu description written by the chef, an email that sounds like the people who will serve them. AI drafts tend towards the same polished, general tone, and guests can often tell.

Edit AI drafts until they sound like you. Keep the phrases your team actually uses, the details only your venue has and the warmth that makes guests come back. A draft that saves time is useful; one that makes the venue sound like everyone else is not.

In the kitchen, AI tools can help scale a recipe, suggest uses for surplus ingredients or draft a prep list from a forecast. Each is checkable by a chef who knows the kitchen, which makes them reasonable uses.

They should not decide food safety matters: cooking temperatures, cooling times, allergen handling or storage. Those come from the venue's food safety plan and the local guidance, and an AI answer that sounds right but is wrong could make a guest ill.

If you want to try AI, start with the safest, most checkable task: drafting a social media post, a job advert or a procedure, which a manager then edits. The risk is low, the time saved is real and the team learns how the tools behave.

Then, if it helps, try a forecast alongside your own, compared each week with the actual covers. Leave guest-facing and safety-related uses until you are confident you understand where the tools go wrong.

An invented brasserie's manager uses an AI tool to draft the weekly social posts and job adverts, editing each before it goes out, and saves several hours a month. They try an AI forecast for a quarter, comparing it each week with the actual covers, and find it useful for normal weeks and poor for weeks with local events, which the manager now adjusts by hand.

A supplier offers an AI chatbot for the booking page. The manager tests it and finds it confidently tells a guest that a dish is gluten-free when it is not. The chatbot is declined, and the brasserie adds a rule: allergen information comes only from the kitchen's own records, never from an AI draft.

servearound has no AI forecasting, rota drafting or writing assistant today, and no chatbot answers guests on its behalf. What it holds are the figures a forecast is built from and checked against: reservation analytics show covers by day and hour and by shift, the reservations calendar shows the bookings already made for the days ahead and payment analytics show revenue by hour, all on the venue's own clock.

Its simple automation is rule-based and checkable: booking reminders go out a set number of hours before each booking, and email automations to chosen audiences run weekly or monthly, reaching each guest at most once. The Digital Twin makes your own chairs, tables and lamps into 3D models from photos for the floor plan. See Analytics.

Drafting writing, giving a first forecast and drafting rotas, each checked by a person who knows the venue.

In confident mistakes and in unsupervised conversations with guests.

It can draft the words. Ingredients and allergens must come from your own records.

As a starting point, compared each week with actual covers and adjusted for local knowledge.

It can draft one. The manager should make the final decisions, for fairness.

Not without a person checking what it says.

No. Automation follows simple rules; AI makes judgements.

Drafting social posts or job adverts, edited before they go out.

AI in restaurants earns its keep today in three places: drafting writing, giving a first forecast and drafting rotas, each checked by someone who knows the venue. It quietly fails in two: confident mistakes that look like facts, and conversations with guests that no person checks. The rule that keeps it useful is simple: use it where you can check it.

Keep allergen and ingredient information out of AI drafts, compare AI forecasting with actual covers every week, keep the manager's judgement in every rota and keep a person between AI and the guest. Prefer simple restaurant automation where a rule will do, treat AI as one of the restaurant technology trends to adopt carefully, and remember that demand forecasting works well without it.