AI in restaurant marketing: what it actually does for a small restaurant

AI in restaurant marketing is mostly a fast, tireless assistant: it picks up calls your team can’t get to, writes first drafts of emails and texts, spots slow hours, suggests offers and reads paperwork. It does not know your restaurant unless you tell it, and it has no taste of its own. Used well, it gives a small team back hours each week. Used carelessly, it says wrong things in your name.

Are restaurants actually using AI yet?

Some are, and most are careful. The National Restaurant Association’s State of the Restaurant Industry 2026 report found 26% of operators using AI-related tools, and marketing was the top use. A later Association survey of 212 operators, reported in November 2026, put use at nearly 39%. Among those users, 58% used it for personalized marketing and outreach, the most common job.

The same survey shows the doubts: 42% of operators called their approach “cautious,” and the biggest worries were complexity (46%), cost (43%), whether guests will accept it (42%) and data security (35%). Those are fair worries. This guide takes them one at a time.

What can AI do for a small restaurant today?

Here is a plain map of the jobs AI handles well now, and what you still need to check.

Job	What AI does well	What you still check
Phone calls in the rush	Picks up, answers hours and menu questions, takes pickup orders and bookings	That its facts are current; how it handles odd requests
Emails, texts, WhatsApp	Writes a first draft in seconds, in several lengths	Tone, prices, dates, and that guests opted in
Slow hours and offers	Reads sales by hour and suggests a deal for the quiet block	Whether the deal makes money and the kitchen can do it
Offer pictures	Makes a clean image for a deal or a post	That it looks like your real food, not a stock fantasy
Menu suggestions	Suggests add-ons and pairings guests often buy together	That suggestions fit your menu and margins
Catering quotes	Turns a head count and budget into a tray plan and price	Portions, dates and anything custom
Supplier bills	Reads a photo of an invoice into line items and totals	Totals against the paper, and price jumps

Picking up the phone when your team can’t

Many restaurant calls come in during the lunch and dinner rush, which is exactly when nobody can answer. An AI phone host can pick up those calls, answer simple questions (hours, parking, “do you have vegan dishes?”) and take pickup orders and table bookings. That turns a missed call into a sale.

Where it struggles: noisy lines, strong accents, unusual requests and anything that needs judgment. Large chains have found this out in public. McDonald’s tested AI ordering at about 100 drive-thrus with IBM and ended the test in 2024. It had wanted better than 95% order accuracy before a wider rollout. The lesson for a small restaurant is to use AI as the backup, not the only way in, and to test it yourself. More on this in our restaurant phone answering guide.

Writing campaigns, emails and texts

This is the easiest win. Give AI the facts (the dish, the price, the day, who it’s for) and it gives you a draft for email, text and WhatsApp in a minute. The blank page is gone, which is often the reason the weekly email never gets sent.

Two cautions. First, AI fills gaps with guesses, so a draft may invent a discount or a closing time. Read every number. Second, the law does not change because a machine wrote it: marketing texts need consent and emails need an unsubscribe. See SMS and email rules for restaurants.

Spotting slow hours and drafting offers

AI is good at reading weeks of sales by hour and pointing out, for example, that Tuesdays from 3 to 5 p.m. run well below your average. It can then suggest an offer for that window: a bundle, a free add-on with a minimum spend, bonus points.

What it can’t do alone is the business judgment. A deal that looks clever can lose money once it reaches guests who were coming anyway. Do the math in our slow-day promotions guide before you approve. And make sure an offer stops when the kitchen gets slammed, so a deal never makes a rush worse.

Offer pictures and menu suggestions

Image tools can make a clean picture for a deal, a holiday special or a social post in seconds. That helps when you have no photographer. Keep the picture honest. If the image shows a dish that looks nothing like what arrives, you trade a click for a bad review. Real photos of your real food are still best for the menu itself.

Menu suggestions (“guests who order pasta often add garlic bread”) are a quieter use. They can lift the average check a little on every order without a discount. Look at the suggestions it makes and remove any that push low-margin items.

Catering quotes and supplier bills

Two jobs that eat owner time:

• Catering quotes. A guest says “60 people, mostly vegetarian, $1,200.” AI can turn that into a tray plan with sizes and a price, or read a list the guest sends and turn it into a quote. You check portions and dates, then send. Speed matters in catering, because a clear quote sent the same day is easier to say yes to. For pricing basics, see how to price catering.
• Supplier bills. Snap a photo of an invoice and AI pulls out the items, quantities and prices. It can also flag, say, onions costing 30% more than last month. That’s not marketing, but the time saved goes back into marketing, and catching price creep protects the margin your offers depend on.

Where does AI fall short?

Be clear-eyed about the limits:

• It doesn’t know what it hasn’t been told. If your hours changed last week and nobody updated the AI, it will give the old hours with full confidence.
• It sounds sure when it is wrong. AI rarely says “I’m not sure.” That is the most dangerous habit it has.
• It writes generic copy by default. Without your stories, words and details, every restaurant’s email sounds the same.
• It has no sense of tonight. It doesn’t know the walk-in is broken, two cooks called in sick or a bus tour just arrived.
• It can’t build relationships. A catering client, a local school, an office manager who orders every Friday: those come from people.

What should stay human?

Two things, firmly.

1. Approving offers and prices. AI can draft. A person should decide what discount goes out, to whom and when. It’s your margin and your kitchen. A good rule: nothing that changes what a guest pays goes live without an owner or manager tapping “approve.”

2. The voice of the restaurant. Your guests come for you, not for a template. Write down how you talk: a few words you use, words you never use, how formal you are, a couple of example messages you love. Give that to the AI and edit its drafts toward it. If a message would sound odd coming out of your mouth at the host stand, don’t send it.

Also keep human: replies to angry reviews, anything about allergies beyond the facts on your menu, and conversations with big clients.

How do you start with AI marketing?

Start small and pick the job that hurts most. A simple 30-day plan:

1. Week 1: Pick one job. Missed calls in the rush, no weekly email, or slow afternoons. Just one.
2. Week 1: Write your fact sheet. Hours, address, parking, menu with prices, allergens, policies, catering minimums, your voice notes. This is what the AI will rely on.
3. Week 2: Run it with checks. Read every draft before it goes. Listen to a few AI calls. Fix the fact sheet each time it gets something wrong.
4. Weeks 3–4: Measure. Compare with the four weeks before (see below).
5. Day 30: Keep, fix or drop. Then add the next job.

What should you measure?

Measure money and time, not novelty.

If AI is doing…	Measure	Compare against
Phone backup	Orders and bookings taken when staff didn’t pick up	Missed calls in the four weeks before
Campaigns	Orders or bookings from each send, unsubscribes	Weeks with no send, or your old emails
Slow-hour offers	Checks and profit in that hour block	The same weekday and hours before the offer
Catering quotes	Hours from enquiry to quote, quotes won	Your last ten quotes done by hand
Bill reading	Time spent on invoices, price jumps caught	A normal month of manual entry

If a tool can’t show you these numbers, ask why.

What are the risks, and how do you manage them?

Wrong information

You are responsible for what your AI tells guests. In a 2024 Canadian case, Air Canada’s website chatbot told a grieving passenger he could claim a bereavement discount after his trip, which was not the airline’s policy. Air Canada argued the chatbot was responsible for its own words. The tribunal disagreed, said the chatbot was part of the airline’s website, and ordered Air Canada to pay. A restaurant giving the wrong price, hours or allergen answer is in the same spot.

How to manage it: keep one up-to-date fact sheet, tell the AI to say “let me have the team call you back” when it doesn’t know, and never let it make medical or allergy promises beyond your listed ingredients.

Tone

A wrong word can do more damage than a wrong date. Watch for over-eager copy, made-up superlatives (“best in the state”), and jokes that don’t fit your crowd. Keep your voice notes short and specific, and read before you send.

Privacy

Guest names, phone numbers and order history are personal data. Don’t paste your customer list into free chatbots. Choose tools that state how they store and use your data, limit who on staff can export lists, and only message guests who agreed to hear from you. A third of operators in the Association’s survey named data security as a top worry, and they are right to.

How WhatsTable helps

WhatsTable is built to give a small restaurant an AI team with the owner in charge. Maya, the AI phone host, picks up when your team can’t and takes pickup orders, table bookings and catering orders. The AI offer planner finds slow hours and drafts a deal with a picture; nothing goes live until you approve it, and offers pause when the kitchen is busy. AI campaign writing drafts email, WhatsApp and SMS messages for you to edit and send, and the AI growth plan suggests what to try next.

On the back of house, the AI catering planner builds tray plans from a budget or a scanned catering list, the website learns which dishes sell and makes smart menu picks, and purchasing reads supplier bills from a photo. It all sits alongside online ordering, group orders, bookings and rewards, and works with Clover, Toast, Square or WhatsTable’s own POS. Pricing is $0 a month, plus 5–10% of the sales that come through WhatsTable. Meet the AI team.