GuidesUpdated

AI for cafés, pubs and restaurants: what it can do with your own numbers

AI can already do six jobs for an independent café, pub, bar or restaurant. Free ChatGPT handles the writing half of each one; the other half needs your own till, rota and review data. Here is which is which, job by job, with a guide for each.

Alex Riesenkampff

6 October 2026 · Updated 15 October 2026 · 12 min read · Markdown

AI can already do six jobs for an independent café, pub, bar or restaurant: write and design menus, draft the rota, reply to Google reviews, read your sales, run your marketing and Google listing, and clear the admin writing. Free ChatGPT does the writing half of every one of those jobs, and text generation is already the most adopted AI technology in UK business, at 17% of firms in June 2026 according to the ONS. The other half of each job needs your own numbers: which dishes sell, who is free on Saturday, what guests keep complaining about. A general chatbot sees none of that until you export it and paste it in. Below is each job, what the free tools do well, what needs your data, and the guide that goes deeper.

That split matters more than which assistant you pick. Much of what gets sold as restaurant AI is phone receptionists, drive-thrus and robots. If a ringing phone is the problem, our guide to AI phone answering prices the receptionists and shows when a booking link does the job. Very few independent venues need any of those. They need the menu written by Thursday, the rota done by Friday and the reviews answered before the weekend.

What can AI do for a restaurant, café or pub today?

AI is useful for six jobs in an independent venue, and the free tools cover the writing in all of them. In a survey of 1,000 UK small and medium firms run by OpenAI between February and March 2026, 30% of hospitality businesses used no AI at work at all, close to one in three. If you are starting, start with one job rather than all six; the table below shows which half of each one a free tool can finish.

The table is the short version of this page. Read across a row to see where the free tools stop.

Dom Jacobs, managing director of The Cadogan Arms, put the test for all six plainly at a Restaurant magazine roundtable:

Trust comes down to what the tool was given. Each section below says what to give it.

Menus: descriptions, design and the prices behind them

AI writes a good menu description in seconds, provided you give it the real ingredients and preparation. "Chicken sandwich" gets you filler. "Crispy buttermilk chicken, hot honey, house pickles, toasted brioche" gets you something a guest wants to order. The same goes for a bakery's counter cards or a pub's Sunday roast board: the model can only sell what you tell it is on the plate.

UK law requires a food business to give allergen information for any loose item containing one of 14 allergens, so every AI-written description gets checked against the recipe before it is printed. Models add plausible details. A scatter of toasted sesame or a "nutty" brown butter that is not on the plate is an allergen statement you never meant to make. Add one line to every menu prompt: do not add any ingredient I have not listed.

Design is the newer part. Image tools can now lay out a full menu, a chalkboard-style special or a poster, and they work best from your own product photos rather than invented food. Expect to correct prices and spellings on the image itself, every time.

What free tools cannot tell you is which dishes deserve the space. That needs item sales and cost per dish. Our menu pricing calculator does the contribution sum per dish, and menu engineering beyond the matrix covers what to do with the answer. The full walkthrough, with copy-ready prompts for dish descriptions, is in our AI menu maker guide.

Rotas: drafting the week, and what a chatbot cannot see

Give a chatbot your constraints in plain English, the days each person cannot work and the cover you need on a Friday, and it lays out a week faster than you can open a spreadsheet. For a café or bar with a small team on settled shifts, that and your existing spreadsheet are enough. Past about eight people, the work stops being the drafting and becomes chasing availability, swaps and sick calls.

Workers are owed 11 hours' rest between working days and a 20-minute break when they work more than 6 hours, and a chatbot drafting from a pasted list checks neither unless the prompt tells it to. Total each person's hours and check the gap between a late close and an early open before you publish anything. Under-18s are owed more.

The part a chat window cannot do is know what changed since you last pasted. It does not know who swapped on Tuesday, or that Saturday lunch has been busier for a month. Heath Ball, owner of The Red Lion & Sun, said at the same roundtable that his managers "would be very happy if the computer just created the rota", because it would stop them being leaned on by staff who want or avoid certain shifts. That only works if the draft starts from current availability and real demand.

Our comparison of AI rota tools prices seven options for a UK team of twelve, from free ChatGPT to apps that draft from your till data, and says which suits which size of venue.

Google reviews: replies in your own voice

AI is good at the hardest part of a review reply, which is staying calm and specific at eleven at night. Paste the review, say what happened on your side and how you talk, and ask for 60 words. Google's own guidance asks owners to "Keep it short and simple" and to reply where they have something useful to say rather than sending the same thank-you to everyone.

60% of British consumers in a CGA by NIQ and Reputation survey said they trust AI-generated review summaries, so the themes in your reviews now reach guests who never read the reviews themselves. That makes the pattern more important than any single reply. If six reviews in a month mention the wait for drinks, a chatbot shown one review at a time will never tell you.

Keep a person in the loop for the first replies at least. Mark Selby, co-founder of Wahaca, said in the Restaurant discussion: "I’d hesitate to let AI reply to a customer without a human sense-checking it." The free-tool limit is the routine: every new review means copying it into a chat, then copying the reply back to Google.

Our guide to replying to Google reviews with AI has copy-ready reply prompts and covers when automatic replies are sensible. For how many reviews you need in the first place, see how many Google reviews a restaurant needs.

Reading your till: the job most venues never get to

This is the job with the most money in it and the one free AI does least well, because it starts with data the chatbot does not have. You can export a month from your till, remove names, upload it and ask good questions. Our prompt library for reports, rotas and reviews has a margin prompt that makes the model show every formula. It works. Then next month the export, clean-up and upload start again, and most owners quietly stop.

When Super44 compared an East London café's own trading days head to head, closing at 4pm capped takings at about £804 a day, while staying open to about 8pm produced £1,062 or more. No benchmark and no general chatbot could have said that. It came from that café's own transactions, and our dead season playbook carries the rest of the spread, down to the days that closed at 3pm.

The questions worth asking your till are rarely complicated. Which hour of the week loses money. Which dish sells well but earns little. Whether the new lunch offer works on every weekday or only one. A bar that launches a food menu, or a bakery that drops two loaves, wants the same thing: a straight answer from its own numbers, a few weeks after the change.

Marketing, your Google profile and being recommended by AI

AI writes a decent Instagram caption, a Google post or a profile description in seconds, and it gets better the more specific the fact you feed it. "Blood orange tart, this week only" beats "come and try our cakes". What it cannot choose for you is which dish to promote, which is a question for your sales data again.

26% of British consumers use AI tools like ChatGPT to learn about venues, close to the 27% who use Google Maps, according to a CGA by NIQ and Reputation survey of 755 people. So AI also matters from the other side of the counter. Assistants answer "where should I get brunch near me" from what they can find about you, and a listing with wrong hours, no menu and unanswered reviews gives them little to go on.

The practical work is unglamorous: hours that match reality, including bank holidays; a description that says what you serve; recent posts; replies to reviews. Our guide on how to get ChatGPT to recommend your café sets out what assistants look for, and ranking higher on Google Maps covers the listing itself.

Admin and prompts: the writing that eats your evenings

For pure writing, free AI is close to all you need. A firm email to a supplier who raised a price without notice, a job advert for a weekend barista, an allergen sheet translated for tourists, a twelve-page lease read through for the clauses that bite: all of it works on a free tier this afternoon.

In OpenAI's survey of 1,000 UK small and medium firms, the most common uses of AI were research and summarisation (47%), emails and business communications (42%) and brainstorming (39%). Those are the jobs that land on an owner after close. A good prompt gives the model a role, the real facts and the format you want back; our prompt library has copy-ready versions for menus, reviews, reports, rotas, captions, supplier emails and translations.

Where admin turns into money, the free tools hit the same wall. "Is our dairy supplier creeping up?" needs a year of invoices. To choose between ChatGPT, Claude, Gemini and the connected tools, our shortlist of AI tools for restaurant owners compares them on what each can see and what each costs.

What a general chatbot cannot do

A general chatbot cannot see your till, your invoices or your rota. It answers from whatever you last pasted in, and it does nothing until you ask. Out of the box, ChatGPT, Claude and Gemini read none of your venue's data, and 51% of UK and Irish operators already name data security and privacy among their biggest concerns about AI, according to Access Hospitality.

There are partial exceptions, and they are worth knowing:

• Square: Square runs a connection that lets Claude reach a Square account. It is in beta, you add it by pasting a server address into Claude's settings, and Square recommends testing it on a sandbox account first. For most other tills, the big assistants document no such connection.
• Google Gemini: Gemini can read Business Profile reviews and draft replies, but Google offers it "wherever the Gemini web app is supported except the European Economic Area and the United Kingdom". For a UK venue, that feature does not exist.
• Your own uploads: any assistant will analyse a file you attach. The answer is as current as the file.

Three things follow. The answer about your venue is only as fresh as your last export. Nothing joins the till to the rota to the reviews, so "were we short-staffed on the nights the reviews dipped?" stays unasked. And the follow-up stays with you: the reply still needs posting, the listing still needs correcting, the rota still needs sending to staff.

None of this makes the free tools a poor choice. For a first-year café with no budget, a free chatbot and an hour a month in the till's back office beats what most venues do now. It just means the jobs above split cleanly in two.

How Super44 does it

Super44 is an AI assistant for independent venues that connects to the data the free tools cannot see. The Starter plan is free with no credit card and covers Google review monitoring and reply support; the live till connection starts on Core at £19 a month, and staff management is on Plus at £49, prices including VAT.

What that connection does, job by job:

• Your till: it connects to Square, Lightspeed, SumUp, Epos Now and other tills, and answers questions about your own sales in chat or out loud by voice: revenue and units per dish, quiet hours, what each dish contributes. If your till is widely used, has a cloud back office and is not yet supported, and you give us access, we normally have it connected within a few working days.
• Reviews: it reads your Google reviews, finds the themes and drafts replies in your own voice. You approve the first replies before anything posts. After five approved replies you can let it answer chosen star ratings automatically. Drafting in your voice needs the Core plan, and a trial counts; publishing also needs your own Google connection.
• Menus and marketing: it creates and edits menus, posters and social images from your saved product photos, with an allowance that depends on the plan. Check every image before it goes out. With your Google connection and the Core plan it can publish posts and update your hours and description.
• Rota: its staff side drafts next week's rota from roles, availability, approved leave, recent weeks and demand in your till data. You review it and publish it yourself. Staff swap shifts, request leave and clock in from their phones or a kiosk in the venue. If you already use Deputy or RotaCloud, it can read that data for analysis without changing your rota there.

What it is not: an ordering system. And if all you want is writing, the free assistants above do that as well as we do.

Start with the job that costs you most time this week. If that is writing, open a free chatbot and use the prompt library. If it is the question you keep meaning to ask your till, that is the job a connected tool exists for.