Picking Interests on Meta Doesn't Work Like It Used To: How Should You Think About Targeting?

There was a time when setting up a Meta ad started with a long list of interests: coffee, yoga, entrepreneurship, online shopping. The narrower and more specific the list, the smarter it looked. In 2026 the picture is different. For sales, app and lead campaigns, Meta now starts the setup with Advantage+ (Meta's AI system), and the system picks most of the audience, the placements and small creative adjustments. At first glance this looks like losing control. In fact the real job is to understand what the system bases its decisions on and give it the right material. Below we walk through it step by step without drowning in technical detail.

What changed, in short

For these campaign types, audience, placements and creative settings are now left to AI by default. Some detailed targeting options (picking interests and behaviors) were also removed or narrowed. So the habit of stacking fifteen interests to reach exactly those people is no longer as strong a lever. Piles of narrow interests and very narrow lookalike audiences (people who resemble your existing customers) leave the system little room to learn, so in most accounts they no longer work the way they used to. In practice: for a small business, a tightly drawn frame like "only ages 25 to 35, this interest, this city" can stop the system from finding people who could genuinely buy but would never occur to you. An overly narrow description sometimes limits the opportunity.

So is control completely gone?

No. Control has moved. The deciding question is less "who should see this" and more "what do I feed the system". There are four things a small business still controls. None of these four needs a technical team; most are things a business owner can start within a week.

• Ad creatives (images and videos): the system chooses who sees it, but what they see is your job. Different messages and formats (vertical video, single image, short explainer) give the system more ways to reach different people.
• Conversion signal: the AI learns from what you give it. If the tracking code on your site (the pixel) and the server-side connection (Conversions API) are set up correctly, the system recognizes people who actually bring a sale or a form.
• Audience suggestions: you can offer your customer list or site visitors as a suggestion. It tells the system "our customers look like this". It is a hint, not a hard limit.
• Exclusions: options like leaving existing customers out of a new-customer campaign or setting a minimum age are still available. They may not work as a perfect wall, but they still cut wasted budget.

Fix measurement first

In a system handed over to AI, the biggest risk is teaching it the wrong thing. If you pick a weak action like "viewed a page" as the goal, the system will find more people who only look and never buy. So track a real business result: a purchase, a form submission or a click to WhatsApp. Having both the pixel and the Conversions API working helps make up for data that browsers lose. An example: if an e-commerce site does not report its sales, the system cannot see which ad actually brought an order and spreads budget by guesswork. When measurement is set up correctly, the same system works in your favor.

Varied creative is the new targeting

You used to open separate ad sets for separate audiences. Now the same job is done with different messages. An image that leads with price, a video showing the product in use and a card built on a customer review speak to three different groups of people. The system works out which one lands with whom. Preparing three to five ads that are truly different is more useful than copying one image in different colors. The difference between messages should be real: if one talks about price, one about quality and one about fast delivery, the system gathers meaningful data for you. When reading results, look at which message led to more sales or forms, not only clicks.

What happened to lookalike audiences?

Lookalike audiences have not disappeared; their role has changed. You used to build a very narrow audience of people resembling your top customers and show the ad only to them. Now the same list can be offered to the system as a hint, and the system will step outside the hint if needed. For small accounts this is actually a relief: even without thousands of customers, the system can widen its own audience as enough signal arrives. Still, the cleaner and fresher the list, the more useful the hint. A list full of people who bought years ago and are no longer interested paints the wrong picture for the system. Refreshing the list at regular intervals, for example every quarter, is worthwhile.

One more note for local businesses: some limits, such as location and age, are still in your hands. For a dentist or a restaurant, there is no point showing ads to people outside the area you serve, so limiting that through audience settings makes sense. Not every door has closed; you simply need to reweigh which doors actually matter.

A short list you can do this week

• Check that the pixel and the Conversions API send the same goal action (purchase, form, message).
• Refresh your customer list and add it as an exclusion in new-customer campaigns.
• Put at least three genuinely different images or videos into one campaign.
• Do not squeeze the audience with narrow interests; offer your customer list as a suggestion instead.
• Judge results over several weeks that let the system learn, not over a day or two.