Facebook ads for restaurants, cafés and bakeries

On Facebook a restaurant is a set of objects Meta can sell ads against: an address, your own phone number, a calendar of events and, for a group, a family of location Pages. Each one has its own ad mechanism, and that is the difference this platform makes for the trade.

The address becomes a button. Meta's local-ad example, written for a store, runs under Awareness with Reach as the performance goal (the thing Meta's delivery system optimises for; here, the number of different people who see the ad) and puts a Get directions button on it, linked to the business's coordinates, so that tapping it presents the location "with a map and directions". Meta's own list of Ads Manager buttons offers Get directions under the Awareness objective and under none of the others. Get directions exists only in the objective this page recommends for dine-in.

Your phone becomes a destination, and for a bakery taking a cake order or a restaurant taking a booking, the call is the sale. Call ads (click-to-call) run only on mobile placements: Facebook Feed, Marketplace, Reels and Stories, and Instagram's feed and Stories. Under Traffic, Engagement, Leads and Sales they optimise for calls lasting sixty seconds. Under Awareness they optimise for the goal you picked, and the button rides along. Ads Manager reports calls placed, twenty-second calls, sixty-second calls and callback requests, which are messages people send through the ad's Send message button asking you to ring back. That last metric matters to a trade whose phone rings loudest while nobody can answer it.

The calendar becomes an objective. A Facebook event is an object Meta will promote directly: the Engagement objective, the event itself as the destination, optimised for event responses. A restaurant produces dated events all year (a supper club, a quiz night, a Thanksgiving seating, a New Year's Eve that sells out or does not), and each is an ad whose job is "come on Thursday".

The branches become a set, and a single restaurant can skip this paragraph. A group whose Page has location Pages under it can build what the Marketing API calls a place page set from the parent Page and point an ad's creative at it. The field description refers to a "Local Awareness guide" that we could not reach when this page was checked, so confirm what it offers in Ads Manager before planning around it.

Under all four sits the audience. Pew's November 2025 survey puts Facebook at 71% of American adults against about half for Instagram. That is a national figure, not your street, but it is why the budget section below can treat your radius as a population to reach rather than a slice to find.

Ads Manager offers six objectives: Awareness, Traffic, Engagement, Leads, App promotion and Sales. The objective you pick is an instruction about whom Meta shows the ad to, so pick it by asking where the sale actually happens. At a table or across a counter: Awareness, because nothing there reaches Meta to optimise for. On a page you own: Sales, because a purchase there is a real event with a real value. In a conversation about a date and a headcount: Leads. At a Facebook event: Engagement, optimised for event responses. On a page the reader has to see before they will come: Traffic. Nearly every restaurant ad fits one of those five sentences.

Two choices look reasonable and are not. Sales pointed at a dine-in offer asks Meta to find people who will complete a purchase on a page where no purchase ever happens, so it has nothing to learn from and will report nothing while it tries. And boosting a post of tonight's special is the quickest ad there is, but most of what this page relies on (a Reach goal with a cap, more than one radius, a 21+ split, exclusions, a Get directions button) lives in Ads Manager. When the Boost screen does not offer the setting you need, build the ad there.

The losing restaurant account runs one photograph of tonight's special, boosted to everyone within ten miles, every week. It tells a neighbour that a restaurant has food and asks them to act on a Tuesday afternoon when they are not hungry. The rows below are offers with an argument attached, and they differ in the one way that matters here: where the sale happens, and whether it leaves anything behind that Meta can count.

Each row pins four decisions: who the offer is for, which angle family argues it (each is a page in our library of ad angles), which objective buys it, and, where the examples page has one, a finished ad. Thirteen rows is not a campaign plan. Pick the two or three that sell capacity you actually have spare: an empty Tuesday, a kitchen idle before eleven, an oven that could bake another hundred pies.

The rows run from the half Meta cannot see to the half it can. The first four leave nothing behind, and you will judge them on the till. From the waitlist down, each leaves contact details, an event response or a purchase, and that is the part you will be able to count. Run both halves. Just never compare them.

Food is not a special ad category, so a restaurant campaign keeps every lever: age, lookalikes, exclusions, custom audiences, and a pinned radius as small as 0.63 miles or 1 kilometre. What you cannot choose is who inside the circle counts. The location type is fixed at people who live there and people recently there, together, and for a restaurant that is the right answer anyway: the office worker in your street at lunchtime is a customer wherever they sleep.

A restaurant has at least three audiences, each a different shape: the dine-in radius, the order radius and the catering pins in the table below. Your competition sets the first, your delivery zone sets the second, and the third is not centred on you at all. Do not guess any of them: your pickup addresses, catering deliveries and pre-order checkouts are already a map of where your customers come from. Give each shape its own ad set, because each offer is only true inside one of them.

Advantage+ audience is the default to check before it quietly changes the plan. Since version 23.0 of Meta's API, a new ad set made through the API either defaults to Advantage+ audience or must set it explicitly, so look at the audience section of every new ad set in Ads Manager. Meta describes Advantage+ audience as widening the pool its delivery system can search, subject to a short list of constraints it will not expand: location, minimum age, language and custom-audience exclusions. On a radius campaign that costs you nothing, because the radius holds. On a regulars campaign it matters. An included customer list is not among those constraints, so treat it as a starting point rather than a boundary, and switch Advantage+ audience off for any ad set that must reach only the people on it.

Minimum age holds as well, and under Advantage+ audience it can be set anywhere from 18 to 25. So a 21+ ad set costs nothing but the split. In the United States, any ad that shows drinking or the sale of alcohol (the bar, a wine list, a glass on a table in a dinner photograph) goes in its own 21+ ad set, and everything else stays open to adults. Outside the United States, Meta's alcohol standard defers to local law.

One restaurant ad does carry a special ad category: the one that says "we're hiring". An ad offering a job must declare the employment category, which brings a minimum radius of 15 miles or 25 kilometres in the US and Canada, ages fixed at 18 to 65 and over, no ZIP codes and no lookalikes. Meta warns that with the wrong category "there is a risk your ads will be paused until the campaign is adjusted". Hire in a campaign of its own.

Last, the lists you already own. The waitlist, the reservation book, pickup orders and club members are the audience closest to a sale, because every name on them has eaten your food, and seated parties you upload as offline events can become a custom audience of their own. Uploading a customer list is a promise to Meta: its Custom Audiences terms have you warrant that you hold "all necessary rights, permissions, and a lawful basis" to use the data. Read your privacy notice and your reservation provider's terms before the first upload.

The usual question is lead form or landing page. A restaurant has an earlier one: should there be a click at all? For a dine-in ad, usually not. The reader has seen the dish and wants either to get there or to ask about a table, and Meta has a button for each. Where your dine-in radius is a drive, use Get directions, which opens a map. Where the call is the booking (a table, a cake), use Call now in the hours before service. For a lunch counter inside a walk, the name and the street in the picture are the instruction and neither button is needed. For people who want to read first, your Facebook Page is the page. A call ad's number must be in the same country as the audience.

Anything sold on a screen goes to your own screen. Pickup, the club, a pre-order or a gift card sold on your own site can carry the Meta pixel and report a purchase with its value, which is the whole reason to buy Sales. Ask your ordering provider two questions before choosing the objective: can I add my Meta pixel, and does a completed order send a Purchase event with the order value? If the answer is no, run the ad as Traffic and count the orders yourself. A listing on a delivery app is somebody else's page. Your pixel is not on it, and the commission comes out of the order the ad just paid for.

A restaurant that takes reservations can run a Book now ad to its booking page as Traffic, optimised for landing page views. Count the bookings in your reservation system, ask the provider the same pixel question, and upload the seated parties afterwards.

Catering, private dining and cake orders go to an Instant Form, and the first thing to know about that form is when its leads disappear. Meta keeps them downloadable for 90 days from submission, so a Christmas party inquiry sent in September can vanish around the week of the party. Connect the form to your email or booking system, or download the leads every week.

Then build the form around the three facts every such sale starts with. There is a date-and-time question type, and a custom question can offer a list of choices, so the form can ask for the date, a headcount band that matches your rooms, and dietary requirements by allergen, which is the question an office manager otherwise answers twelve times on the day. Choose the Higher intent form type, which adds a step where people review their answers before submitting, and consider the separate setting that verifies the phone number by text.

Leads has two other destinations that suit this trade. It can send the reader into Messenger instead of a form, which fits private dining, where the useful answer is a conversation with a manager about a room. And it can send them to a phone call, optimised for sixty-second calls, which fits catering and cakes, where a buyer may want to ask one question and order. Both only work if somebody answers.

Anything with a date goes to the event. If the supper club or quiz exists as a Facebook event, the Engagement objective can use the event itself as the destination and optimise for responses. If seats are sold on a ticket page instead, that is a Traffic or Sales ad to the ticket page, and the number of seats left is the scarcity.

Start with what a neighbourhood buy does to an ad: the same people see it again, and you cannot choose which of your ads a neighbour sees first. So every ad has to say what the place is and where it is, and the reasons to come this week should vary across them. Rotate a few creatives inside the cap, and let a changing menu do the rotating for you; a fixed menu has to manufacture its variety. Put the night and the hours in the frame whenever the offer has them. The schedule decides when an ad is seen; the copy, not the schedule, names the night. And for people who have walked past for years, show the front they walk past as well as the dish, because a Get directions ad has no page to show what your front looks like.

In a restaurant ad the picture of the food is a statement about the menu. A customer who orders what the ad showed and gets something else has been misled, whoever drew it. Most of the ten ads on the examples page show dishes their one-line briefs never mentioned, but most of those came from the menu a model wrote for each invented restaurant in place of a website, and the ads copied them faithfully. The failures that would reach your own ads happened later, in the picture: the pastrami came back with a layer of loose peppercorns that no sandwich has. Check a generated image dish by dish against your menu, or photograph your own.

Then look for a glass. None of the ten briefs mentioned a drink, and both ads that came back with wine were for restaurants whose invented menus sold it: the picture took the drink off the menu and put it on the table. A glass on the table puts the whole ad set at 21+ in the United States.

Some claims a kitchen has to be able to prove. Gluten-free has a federal definition, under 20 parts per million, and the FDA expects a restaurant's menu claim to be consistent with it, so the word carries a standard whether or not you print a number. Sesame has been the ninth major allergen since 1 January 2023. An allergen card that lists what a dish does not contain and leaves out what it does is worse than no card.

Describe the room, never the reader. Meta's personal attributes standard prohibits content that asserts or implies a person's health, including medical conditions, or implies you know "the personal attributes of their family". Dietary need is where a restaurant is most likely to meet the rule, and "when your child can't eat gluten", the gluten-free ad's first line, is the natural sentence and the prohibited one. A line about a room that wheat flour never enters sells the same cake and asserts nothing about anybody.

Never let a model compute the number an ad rests on. A break-even, a price per head, a saving against the app: each is a sum you can do on paper in ten seconds, and a model can get the same one right and wrong in a single run. Leave a line for margin in any price breakdown. And when your app prices are higher, separate the fees the app charges your customer from the menu prices you set, and explain the second yourself.

Shape and length: Facebook Feed wants 4:5 at 1440 × 1800, primary text of 50–150 characters and a 27-character headline; Marketplace wants a square with a 30-character description, which all ten ads in this batch met. Most of the batch's primary texts and headlines run past Feed's figures, so measure every ad before it runs. The counts are on the examples page.

A dinner is decided close to the meal, and a private party or a holiday order is decided weeks ahead, so the structure is not a sequence of stages one person walks through. It is four lines running side by side, one for each place the sale happens. Each is judged by its own instrument and sized by its own arithmetic, not by a percentage of a budget.

The neighbourhood line will show almost nothing in the conversions column, and the screen line will show purchases. That tells you what Meta can see, not which line is working. A café with no catering and no private room runs the first two lines and a small fourth, and that is a complete account.

Start with the neighbourhood line, because it is arithmetic about a population rather than a guess about who will respond. The reach sum: the accounts in the audience estimate Ads Manager shows for your dine-in radius, × the sightings a week you will pay for each of them, ÷ 1,000, × your CPM (cost per thousand impressions) = roughly the weekly cost.

Treat it as an upper-end planning figure rather than a promise: a frequency cap is a maximum, and the auction does not guarantee that everyone reaches it. If you have never run Reach you do not have a CPM yet, so run the ad set for one week at a spend you choose, read the CPM off the report, then do the sum, remembering that the CPM at a larger spend may differ from the test week's.

Then work out what the line has to produce, on the service it is meant to fill. An extra cover on a quiet Tuesday is worth roughly the bill minus the cost of the food, because the staff are already rostered and the rent is already paid. An extra cover at seven on a full Saturday is worth close to nothing, because the table would have sold anyway. So: weekly cost ÷ (the bill − its food cost) = the extra quiet-night covers a week the line must produce to pay for itself. And weekly cost ÷ (average bill × the visits a regular makes in a year on services you could not otherwise fill × your margin) = the new regulars a week it must find. If either number is larger than you would ever notice at the door, shrink the radius or lower the cap before you spend.

Set the cap per offer, over the interval the decision takes. A lunch counter's regulars decide every day, so their cap runs over seven days; a tasting menu or a birthday dinner is decided over weeks, so its cap is lower and runs over thirty. The cap can cover anywhere from 1 to 90 impressions over 1 to 90 days. Pair it with the schedule so the sightings land before the decision: Friday afternoon for Saturday dinner, late morning for lunch.

You will buy that reach in the auction. Meta's other buying type, Reach and Frequency, is available only to some ad accounts and needs a target audience of at least 300,000 accounts and a minimum reach of 200,000 in a single country. If your radius is smaller than that, and one restaurant's usually is, the auction is where you buy. The frequency cap on an auction Reach ad set, available only with the Reach and ThruPlay goals, is a restaurant's media plan.

The screen line's problem is volume. Meta's delivery system needs roughly 50 purchases a week per ad set to finish its learning phase, the stretch after launch when it is still working out whom to show the ad to. Multiply what Meta charges you per online order by 50 and compare it with your weekly budget. If it is larger, put the screen offers that share a radius in one Sales ad set rather than four that each stay stuck, and switch its radius and offers for November and December. A regulars-only pre-order or a lookalike test is a second Sales ad set, so run one only if 50 × your cost per purchase covers both.

Then measurement, which you build yourself from three instruments. The first is the till: compare the weeks an ad runs with the weeks it does not, on the same days of the week, and accept that the answer is a range. It is the only instrument that measures what the ads caused. The second is a code, word or named offer redeemed at the counter and counted by you. Meta's own promo-code feature applies codes at checkout on a website, so an in-store redemption is yours to count.

The third is the bridge, and it needs no developer. In Meta's Events Manager, create a dataset and assign it to your ad account. Upload a CSV file of the parties you seated: a phone number or email, the time, and the bill if you want value reported. Upload within 62 days of the visit; weekly is a sensible rhythm. The matched parties appear in Ads Manager under the offline conversions columns, and the same uploads can build a custom audience.

Read them carefully. Waitlists exist on full nights, so the dinners you can upload lean toward the services an ad cannot improve: count with them, but do not move money by them. Each upload is a table, not a head count. The number worth watching is first visits: phone numbers your waitlist or reservation book has never seen before. Your own records can pick those out, and their count in the weeks the ads run against the weeks they do not is as close as this data gets to new guests the ads found.

Some point-of-sale companies sell a shortcut. Toast runs its own ad product and says the sales it reports for Meta campaigns combine online, catering and on-premise sales "attributed to your campaign" using "Meta's attribution technology". Its help page does not say how a guest at a table is matched, and that answer decides what the number means, so ask before you let it judge your ads. Square's Meta help, when we checked, covered its online store rather than in-store sales.

Last, spend where the capacity is. A Saturday at seven sells itself and a Tuesday at six does not, and the arithmetic above says the Tuesday cover is the one worth paying for. Point the neighbourhood line at the services you cannot fill, with copy that names them. Run pre-orders from the start of November, and open private-dining ads in September for December: the calendar a guest plans on runs weeks ahead of the one the kitchen works to.

The first two are one mistake seen from opposite ends: paying to fill the night that fills itself, and then judging the line meant to fill the others by a column that cannot see it. The rest are ordinary, expensive and specific to kitchens.