How Meta ads work in 2026
Why I wrote this?
If you open Meta Ads Manager today, it still looks familiar. Campaign, ad set, ad. Budget, audience, creative. But behind those screens, most of the decisions that advertisers used to make by hand are now made by Meta’s AI.
I wanted to understand how Meta ads work today, not how they worked three years ago, so I went through what’s available online and put what I learned into one place. Meta changes something almost every month, and a lot of what you can find about it is already outdated, so I tried to keep only what’s still true as of October 2026.
This guide is for anyone who’s starting out with Meta ads, and it’s also a reference for myself.
How Meta chooses which ad to show?
Every time someone opens Facebook or Instagram, Meta has a fraction of a second to decide which ads to show them, out of tens of millions of ads that are running at that moment. It doesn’t compare all of them at once but works in stages, and the easiest way I found to understand those stages is to compare them to hiring someone for a job.
The first stage is a shortlist. A system called Andromeda, which Meta introduced at the end of 2024, works like a recruiter going through a huge pile of CVs, and it narrows all those ads down to a few thousand that could be relevant for this one person.
The second stage is ranking. Here a group of models works like hiring managers, going through the shortlist and estimating, for each ad, how likely this person is to click, buy or sign up.
Behind these models there’s GEM, short for Generative Ads Model, which is Meta’s biggest AI model for ads. It learns from what billions of people do across Facebook, Instagram and Meta’s other apps, so it’s very good at predicting what someone will respond to. The problem is that a model this big is also slow and expensive to run, while Meta has to make this decision almost instantly, every time anyone scrolls, so GEM doesn’t choose the ads itself. It works more like an experienced head of HR who doesn’t interview candidates personally but trains the hiring managers, passing what it has learned to the smaller and faster models that make the actual decision.
The third stage is the newest one, called the Generative Recommender. It uses large language models, the same kind of AI that’s behind chatbots, to understand what an ad is about and what a person is interested in, and to look at both together.
After that comes the auction. The ad with the highest bid doesn’t automatically win, because Meta combines the bid with how likely the person is to take the action and with how good it thinks the ad is.
Meta reports that GEM alone increased conversions by 18.9% and the click-through rate by 12% across its apps. In its results for the third quarter of 2026, it says each of the three systems raised conversions by roughly another 1 to 2% where it measured them. These are Meta’s own numbers about its own product, so I take them as a direction rather than a fact.
What I found most important here is that Meta doesn’t only look at the audience you set up, but also at the ad itself, including what’s in the image or the video and what the text says, and it uses that to work out who might be interested in it. For example, if your ad shows a trainer working with a puppy, Meta can figure out that it’s probably interesting for people who’ve just gotten a dog, even if you never told it that. I’ll come back to this later, because it changes how ads should be planned.
What Advantage+ means today?
Advantage+ is a name you see almost everywhere in Ads Manager, and it’s easy to think it’s a special type of campaign, but it’s not anymore. Today it’s Meta’s name for letting its AI make a decision instead of you.
A sales, leads or app campaign counts as Advantage+ when three settings are turned on. The first is Advantage+ campaign budget. You set one budget for the whole campaign, and Meta decides how to split it between the ad sets. The second is Advantage+ audience. You describe who you want to reach, but Meta treats most of it as a suggestion and can show your ads to other people if it thinks they’re more likely to convert. The third is Advantage+ placements. Your ads can appear anywhere Meta sells ad space: Facebook, Instagram, Messenger, Threads, and other apps and websites in Meta’s network.
Since early 2026, there’s one setup flow for all campaigns, and the older Advantage+ shopping campaigns were merged into it. If you turn one of the three settings off, the campaign still runs, but it won’t be called Advantage+ anymore. By the middle of 2026, Meta said these campaigns were bringing in more than $75 billion a year.
What stays in your hands is still quite a lot. Some audience settings are hard limits that Meta won’t go beyond: the location, the language, a minimum age of up to 25, and the audiences you exclude. You can block placements for your whole ad account. You choose the budget, the result you want, and every ad you upload. And there’s a separate group of settings called Advantage+ creative, where Meta can crop your image, add music or rewrite your text, and each of them can be turned off.
In other words, you’re responsible for the goal and for what goes into the ads, and Meta is responsible for delivering them. You decide what the campaign should achieve, like sales or sign-ups, and you create the offer and the ads. Meta then decides which people see them, where they see them, for example in the Instagram feed or in Facebook Stories, at what time, and, if you uploaded several ads, which one each person gets.
A walkthrough: the campaign, the ad set and the ad
Every Meta ad still lives on three levels. The campaign holds the goal, the ad set holds the rules about the audience and the budget, and the ad is what people see. Below I go through each level, looking at what you set yourself, what Meta takes over, and what seems to work best right now.
The campaign
At the campaign level, the first decision is the objective. There are six of them: Awareness, Traffic, Engagement, Leads, App promotion and Sales. This choice matters more than it seems, because Meta optimizes for exactly what you choose. If you choose Traffic, it’ll find people who like to click, not people who are likely to buy.
You also set the budget, which can be one budget for the whole campaign or a separate one for each ad set, and either daily or for the whole run.
Then there’s the bid strategy, which tells Meta how much it’s allowed to pay for a result. The default is Highest volume, where Meta tries to get as many results as possible for your budget without a price limit on each one. The other options are for when you already know your numbers. With a cost per result goal, Meta tries to keep the average cost around the amount you set. With a ROAS goal, it aims for a certain return on ad spend. A bid cap is a hard limit on what Meta can bid in each auction. The usual advice is to start with Highest volume, because if you set a target that’s too low, Meta may simply stop spending. You also need some results first to know what a realistic target even is.
And if you advertise credit, jobs, housing, or social and political topics, you have to declare a special ad category. It limits your targeting options, but ads that should have been declared and weren’t can be rejected.
Meta’s part here is to decide how the budget is spread between your ad sets and across the days.
As for what works best, the usual advice is to choose Sales if people buy on your website and you can track the purchase, and Leads if there’s a step in between, like a call, a quote or a sign-up. A daily budget is usually recommended over a lifetime one, because it’s easier to change and the campaign can keep running for as long as it works. Most of what I read also agrees that with a small budget, one campaign for one offer works better than several, because if the budget is spread across many campaigns, none of them gets enough results to learn from.
The ad set
The ad set is where most of the settings are. First you choose where the conversion happens: on your website, in an instant form inside Facebook or Instagram, in a chat on WhatsApp or Messenger, in a phone call, or in an app. Then you choose the performance goal, which is usually either the number of conversions or their value. Some goals, like link clicks or landing page views, quietly turn the campaign back into a traffic campaign. Then you choose the conversion event, which is the action Meta will count as a result, for example a Purchase or a Lead. For this, Meta needs the Meta Pixel on your website, and ideally also the Conversions API, which sends the same events from your server so that fewer of them get lost.
The audience is the part that’s changed the most. Location, language, minimum age and exclusions are still hard limits, but age, gender, interests and your own customer lists are now only suggestions, and Meta can go beyond them.
There are two more settings worth knowing about. Value rules let you tell Meta that some groups of people are worth more to you than others, for example people in a certain region or on a certain device, and Meta will then bid up to 1,000% more or up to 90% less for them. The other one is placements. Since late August 2026, you can’t remove placements inside an ad set anymore, but you can still block them for the whole ad account, or use a value rule to lower your bid on a placement by up to 90%.
Meta then decides who sees your ads within your limits, on which placement, at what time and how often. It also retargets on its own, because your customer lists are only suggestions, so an ad set built for new people will still spend part of its budget on people who already know you, like past visitors and customers.
From what I read, a few things matter most at this level.
The first one is to optimize for the result you really want. Meta says it needs about 50 of those results a week to learn properly. If that’s not realistic, you can test a step earlier in the funnel, like add to cart, but then you should expect Meta to bring you exactly that step and nothing more.
This matters a lot with leads, because Meta counts every form someone sends as a success, even if the person left a fake number or can’t afford what you sell, and then it goes looking for more people who behave the same way. A common fix is to filter people out inside the form itself. For example, if you only work with clients in Germany, you can ask where they’re based and set the form to close for anyone who picks another country. You never receive their contact details, so they don’t count as a lead, and Meta won’t look for more people like them.
Another common recommendation is the form type called Higher intent, where people see their details once more and confirm them before sending. You can also add a step where people have to type in a code that’s sent to their phone by SMS, and only those who do it end up as leads. You’ll probably get fewer leads this way, but more of them will be real people who are seriously interested. And since June 2026, a lead form on Facebook can also show a booking calendar from Calendly or HighLevel right after it’s sent, so people can book a call with you straight away without leaving the app.
For sales where some customers spend much more than others, most guides suggest optimizing for value instead of the number of purchases, so that Meta looks for people who spend more and not only for people who buy.
It’s also often recommended to use a broad audience and a single ad set while the budget is small, rather than several narrow ones, for the same reason as with campaigns. It’s similar with retargeting, which means showing ads again to people who already know you, like those who’ve visited your website or bought from you before. Advertisers used to run a separate campaign just for them, but as I mentioned above, Meta now shows your main ads to these people anyway, so a separate retargeting campaign mostly reaches the same people a second time and splits your budget between two campaigns that compete for them.
It also helps to tell Meta who your existing customers are. In the advertising settings, you can define your existing customers and your engaged audience, and after that you can see how much of your budget goes to each group. If too much of it goes to people who have already bought, a setting called Customer Lifecycle Strategy, which became available to all advertisers in October 2026, can focus the ad set on new customers.
Value rules need a bit more explanation. Meta only counts results that happen within a few days after someone clicks or sees your ad, like a purchase or a sign-up, and it knows nothing about what happens after that, for example when a customer comes back and buys again three months later, returns the product, or when a lead fills in your form but never becomes a client. You might know from your own records that, say, customers over 45 come back much more often than younger ones, while for Meta a first purchase from both groups looks the same. A value rule is how you pass this on, so you could tell Meta that people over 45 are worth 30% more to you, and it’ll then bid more to reach them. The common advice is to use value rules only for this kind of information that Meta can’t see on its own. A rule based on something Meta already sees, like which people buy in the first place, adds nothing, and Meta also warns that value rules can raise your average cost per result, because you’re asking it to pay more for some people. So this is more of a tool for later, once you’ve collected enough of your own data to see such differences.
Finally, there’s a setting called incremental attribution, which tries to count only the conversions your ads caused. It makes your results look worse, and it’s usually recommended for bigger advertisers. In one test with about $1 million in ad spend, Meta counted 87% of conversions as caused by the ads, while Google Analytics put it at 67%, so I’d see it as a rough guide rather than a final answer.
The ad
At the ad level, you choose the identity, which is your Facebook Page and Instagram account, or a partnership ad that also runs from a creator’s account. You choose the format, a single image or video or a carousel. It helps to upload a vertical 9:16 version for Stories and Reels and a square or 4:5 version for the feeds. You write up to five primary texts and five headlines, and Meta can also suggest versions written by AI, which you approve one by one. You set the destination, which can be a page on your website, an instant form or a chat.
Then there’s a list of creative enhancements, which are edits Meta is allowed to make to your ad, like cropping it, adding text on top of the image, animating it, adding music or rewriting the text. There are also two tools for testing new ads, which I’ll come back to below.
Meta decides which ad each person sees, which text goes with it, and, if you allow it, how the ad is cropped, animated or rewritten. In sales campaigns that use a product catalog, there’s also an option called Optimize website destination. When it’s on, Meta can send people to your homepage or to one of your product pages instead of the link you entered, if it thinks that page is more likely to lead to a sale.
The most common advice at this level is to check the enhancements one by one. The ones that only resize your ad for each placement are usually safe, but it’s better to turn off things like rewriting the text on an image full of customer reviews, AI-made versions of a real person, or music that doesn’t fit the brand.
And it’s a good idea to check them again after every change, because duplicating or editing an ad can switch some of them back on. This matters even more from October 30, 2026, when Meta’s new terms put all the responsibility on the advertiser, even for things that Meta’s AI added to an ad or changed in it.
Optimize website destination is worth checking too. If you made a page specifically for an ad, for example with a discount that’s only for people who come from that ad, it’s usually recommended to turn this option off, so that people land on the page you prepared and not somewhere else on your website.
Testing new ads needs some planning too. When you add new ads to an ad set that’s already running, they often get almost no budget, because Meta prefers to keep spending on the ads that already bring results rather than risk money on unknown ones, so you never find out whether a new ad is any good. This is what the two testing tools are for. The creative testing tool works like a short, fair contest. You choose 2 to 5 new ads (some accounts can now test up to 10), how much of the budget they get, how many days the test runs and which result decides the winner, and Meta divides the audience between the ads, so nobody sees more than one of them. When the test ends, you see the results side by side, and Meta goes back to spending the budget as usual. The other tool, called push delivery to this ad, is simpler and only available in some accounts. You pick one new ad and give it a fixed share of the ad set’s budget for 1 to 30 days, which is enough to see how it does without setting up a whole test.
From what I read, the testing tool makes sense when you have a few new ads ready at once, and push delivery when you want a quick look at just one. Meta’s own recommendation is to keep a test at 20% of the budget or less, so your best ads keep working in the meantime, and they should stay on while you test. After the test, you keep the new ads that did as well as your current best one or better and turn off the rest. Most of them will still do worse, and that’s normal, because what you’re looking for is the one ad that does better.
But the most important thing at this level is that the ads should be genuinely different from each other. I think this is the biggest change of all, so it gets its own section.
Why the ads matter more than the audience settings?
A few years ago, the way to reach the right people was through the audience settings, like interests, lookalike audiences and carefully built groups, and the ad was just the message. Today most of those settings only guide Meta rather than limit it, and Meta reads the ad itself to decide who should see it. That’s why people in the industry often say that creative, which is what marketers call the ads themselves, is the new targeting.
This has a side effect that I didn’t expect. If ten of your ads look and sound almost the same, Meta doesn’t learn anything new from them, and agencies report that it treats such ads as one, so they compete for the same people instead of reaching new ones. In August 2026, Meta even added a Creative diversity column to Ads Manager that rates each ad set as Low, Medium or High, although Meta describes this score as an estimate that’s still being developed.
So the goal isn’t to find one perfect ad and put everything behind it, but to give Meta several clearly different ads, so it can show each person the one that fits them best. As far as I understand it, a good way to do that looks like this.
First, decide who the ads are for, not everyone who could buy, but the customers who are best for your business.
Second, write down the reasons why they’d buy. Usually there’s more than one, like saving time, better results, price, peace of mind, or how it makes them look to others.
Third, give each reason its own ad. An ad that lists five benefits is hard to follow, while five ads with one benefit each are easy to understand. At this step you’re testing what you say, and the results show which reason matters most to people, which is often not the one you expected.
Fourth, once you know which reasons work best, test how you show them. The same reason can be shown as a short video, the founder talking to the camera, a customer testimonial, a product demo, a carousel, or a simple image with one clear message, and it often works much better in one of these forms than in the others. After that, you can move on to small details, like the first few seconds of a video or the headline.
Many sources also say that specific messages work better than general ones. An ad for running shorts doesn’t speak to anyone in particular, but an ad for running shorts that don’t rub on long winter runs makes the right person stop scrolling, and that person is exactly who Meta is trying to find.
Creators are the other big lever. A partnership ad runs from your account and from the creator’s account at the same time, so it comes with their face and with the trust of their audience. In Meta’s own studies, when at least 30% of the ads in a test were partnership ads, each additional sale cost 45% less. Again, this is Meta measuring its own product, but the fact that advertisers keep spending more on them says the same thing, since partnership ads now bring Meta more than $15 billion a year, almost twice as much as a year before. To find creators, Meta has a tool in Ads Manager called the Creator Marketing Hub. The advice I came across most often is to choose creators whose audience matches yours, rather than the biggest name you can afford.
AI can help you produce more ads, and Meta now has its own tools for this, which nearly 10 million advertisers used in the third quarter of 2026. But there’s one warning I kept seeing, which is that ads that clearly look like they were made by AI tend to perform badly. Placing a real photo of your product in a new setting is one thing, but showing a person or a result that doesn’t exist is another.
What hasn’t changed?
With so much automated, it’s easy to think the basics of advertising don’t matter anymore, but from what I read, they matter even more, because Meta’s AI still needs time to learn, correct data and something worth selling, and it can’t create any of these on its own. Some of these basics are about how you work in Ads Manager, and others are about things outside it, like your offer, your website and how much you can afford to pay. Below are the five that came up most often.
The first one is the learning phase. After you launch an ad set or make a big change to it, Meta experiments to find out what works, and the results jump around a lot. Meta says an ad set leaves this phase after about 50 results within 7 days, and large budget changes, new targeting, a new conversion event or new ads can start it again. If an ad set shows Learning limited, it isn’t a penalty but only means that it’s not getting enough results to learn fully, and it can still be profitable.
The second one is patience. The usual advice is to make changes in batches, about once a week, rather than every day, because Meta plans how often to show an ad to each person, and constant edits break that plan. How long to wait before judging an ad depends on the number of results, not on the calendar, so ten sales in one day tell you as much as ten sales in one month.
The third one is tracking. Meta can only optimize for what it can see, so you need the Meta Pixel and the Conversions API, and you need to check that each event fires on the right page. If a Purchase or Lead event fires as soon as the landing page loads, Meta thinks every visitor converted and starts looking for more of the wrong people. Meta’s free Pixel Helper extension for Chrome shows which events fire on each page.
The fourth one is the offer and the landing page, because no setting can fix an offer that people don’t want, or a page that doesn’t convince them.
And the fifth one is realistic costs. Advertising on Meta keeps getting more expensive. In the third quarter of 2026, the average price per ad was 14% higher than a year earlier worldwide, 23% higher in the US and Canada, and 10% higher in Europe. But what matters in the end is the cost per result and the return, not the price of the impressions, so a higher price can be fine if the people you reach are more likely to buy.
When the results are disappointing, a common way to find the problem is to look at where people drop off. If there are no conversions at all, the problem is most likely the offer. If there are a few conversions but people rarely click, it’s most likely the ads. If people click a lot but few of them convert, it’s most likely the landing page or whatever happens after the click. If there are many conversions but no profit, it’s the business model itself, like the price, the margins, or the lack of a second sale. And if there are many conversions and a profit, there’s nothing to fix, and that’s the moment to spend more, step by step.
The changes in 2026 worth knowing
These are the changes from this year that affect how ads are set up, paid for or measured, starting from the most recent.
October 30, 2026. Meta’s new Commercial Terms take effect. If Meta’s AI adds something to an ad or changes it, the advertiser is responsible for the result, even if Meta turned that feature on by default.
October 2026. Meta announced that its AI assistant in Ads Manager will be able to make changes from a chat, like creating ads or changing budgets, although these features were still being tested. Customer Lifecycle Strategy, a setting that lets an ad set focus on new customers instead of people who’ve already bought from you, became available to everyone.
August 26, 2026. The Creative diversity score appeared in Ads Manager.
August 25, 2026. Placement choices disappeared from ad sets.
July 1, 2026. Meta started adding location fees to ads shown in some countries: 2% in the UK, 3% in France, Italy and Spain, and 5% in Austria and Turkey. What counts is the country of the people who see the ad, so a business from Germany pays the fee when it shows ads in France, while ads shown in Germany have no fee.
June 24, 2026. Lead forms on Facebook started showing a booking calendar from Calendly and HighLevel.
March 2026. A click now means only a click on the link, while likes, comments, shares and video views of at least 5 seconds are counted separately, as engage-through. Because of this, fewer conversions show up as click-through, so older and newer campaigns can’t be compared directly.
February 2026. There’s now one setup flow for all campaigns, with the old Advantage+ shopping campaigns merged into it.
January 2026. People in the EU can choose to see less personalized ads, so there’s less targeting data for part of a European audience.
January 21, 2026. Ads on Threads expanded to all users worldwide.
What we can take from all this?
The settings matter less than they used to, and what matters more is everything Meta can’t do for you, like knowing exactly who the customer is, having an offer they really want, making ads that differ from each other, measuring the right result, and leaving the system alone long enough to learn.
Meta’s AI reminds me of a very fast new colleague. It’s great at finding people, but it only knows what you tell it. If you give it a clear goal and honest data, it works for you, but if you give it the wrong goal, it’ll chase that just as hard.
One more thing to keep in mind is that Meta changes things quickly, so some of what’s written here may already work differently a few weeks from now.