Facebook ad targeting in 2026: what still works after the detailed targeting cuts

Most targeting advice written for Facebook ads describes an interface that is no longer there. Stack interests, exclude the ones wasting money, narrow until the estimated reach looks sensible: of those three moves, one was deleted outright, one now does something different from what it says, and one was never the lever people thought it was. Everything below was read off Meta's own help pages, developer blog and API reference on August 26, 2026, and every date in it is Meta's.

What Meta removed, and when

Meta announces these cuts in three separate places and never in one list: a newsroom post, a developer blog, and a single help page that quietly carries the dates. Put end to end, the record reads like this.

1. 1.August 11, 2020. Meta retires multicultural affinity segments and merges duplicated options, and notes it had already removed thousands of exclusion categories tied to race, ethnicity, sexual orientation and religion over the previous two years.
2. 2.November 9, 2021. Meta announces that from January 19, 2022 it will remove detailed targeting options relating to topics people may perceive as sensitive: health, race or ethnicity, political affiliation, religion and sexual orientation. The specimens it names are worth reading, because they show how wide the net was cast. Lung cancer awareness went. So did same-sex marriage and Catholic Church.
3. 3.December 8, 2021. The developer blog attaches a deadline to it. Affected campaigns keep delivering until March 17, 2022, after which no edit is accepted at any level and the Marketing API answers error 100, subcode 18157520.
4. 4.March 31, 2025. Detailed targeting exclusions stop working in Ads Manager, and are stripped out of campaigns that were already live rather than left to expire. The same removal reaches boosted posts on June 10, 2025. Meta's stated reason is an advertiser test in which median cost per conversion came out 22.6% lower without exclusions than with them.
5. 5.June 23, 2025. Consolidation begins. Interests inside the same sports category, film or music genre, car model, or food and drink type get folded into broader groupings, and Ads Manager offers alternatives when you reach for one that is gone.
6. 6.January 15, 2026. Campaigns created before June 23, 2025 stop delivering if the affected options were never removed or replaced. This one ended campaigns rather than degrading them, which is why it is worth opening ad sets nobody has touched in two years.

Meta has never published a list of the interests it removed. What it publishes is a way to find out which of your own ad sets are affected: a Deprecated Targeting Terms endpoint in the Marketing API, and a targeting state that reads normal, deprecating, delivery affected or delivery paused. Those return ad sets, not terms. Any article handing you a list of dead interests assembled that list by guesswork.

What is left, and what moved underneath it

Seven things can still shape who sees a Meta ad. They are not equally durable, and three of them changed meaning without changing name, which is the main reason old guides read as though they are describing your account when they are not.

Broad targeting stopped being the lazy option

For years, broad meant you had not done the work. The reversal is not a fashion, and it does not require believing that Meta's AI is clever. It follows from what the other settings now do. If detailed targeting is a suggestion, then a hand-built interest stack is not a fence around your audience. It is a first guess handed to a system that is free to look elsewhere, and on several optimization goals it will look elsewhere whether or not you agree.

The honest version is narrower than the slogan. Broad works when the optimization event arrives often enough, and truthfully enough, for the delivery system to learn from it. A local service business whose sales close on the phone is handing that system a partial record of who actually bought, and broad targeting on a partial record finds more of whatever was visible, faster. That is a measurement problem wearing a targeting costume, and no setting in the audience section repairs it.

One reason the old instinct misfires: detailed targeting selections are not mutually inclusive. Add films, books and TV and you reach people who match any one of the three, not all three. Every interest you add widens the pool. Narrowing takes a separate control, called Narrow audience or Define further, which most small accounts never open.

Advantage+ audience and Advantage+ detailed targeting are two different settings

These get treated as one thing in most write-ups, and telling them apart is the practical half of this whole subject. One is a lever. The other is the room the lever sits in.

• –Advantage+ detailed targeting is the lever. It says Meta may deliver beyond the interests you selected. Meta states it applies only to detailed targeting and not to location, age or gender, and that exclusions and every non-interest selection keep applying as before. This is the setting older articles call detailed targeting expansion. Meta's current page carries no formerly known as label for it, even though its pages do carry one for Advantage+ placements and Advantage+ campaign budget, so both names are worth recognizing.
• –Advantage+ audience is the room. It converts the entire audience section into suggestions at once: age, gender, detailed targeting, and the custom audiences you include. What stays binding is a short list, and it is the list worth memorizing: locations, minimum age, languages, and the custom audiences you exclude. Meta puts it plainly on that page, saying suggestions do not always constrain your audience.
• –You can still make interests binding, but not everywhere. Meta's instruction is to choose Further limit your audience, find detailed targeting again, and untick the Advantage+ detailed targeting box. Then comes the sentence that matters for small accounts: if you optimize for link clicks, landing page views, conversations, conversion, app events, app installs or value, expansion is applied automatically with no option to opt out. For anyone optimizing for leads or purchases, that covers the campaign they were going to run.

The asymmetry between including and excluding is the part worth carrying around. A custom audience you exclude is always treated as a control, even when the suggestion box is ticked. A custom audience you include is a suggestion. So under Advantage+ audience, keeping existing customers out of a prospecting campaign is reliable, and pushing delivery toward a chosen list is not.

Lookalikes are alive, and quietly different

No deprecation notice exists for lookalike audiences, in the help center or in the API reference. Meta still documents the requirement as a source with at least 100 people, recommends a source between 1,000 and 5,000, allows up to 500 lookalikes from one source, and excludes the source members from the result. The familiar 1 to 10 percent range is not on the help pages at all. The API reference documents the ratio as 0.01 to 0.20, so if you want a number Meta actually publishes, that is the one that exists.

Two changes catch people out. You no longer choose a country when you create a lookalike, so new ones show no estimated size in the audience list and take their location from the ad set instead. And inside Advantage+ audience a lookalike arrives as an inclusion, which makes it a suggestion, and inclusions there cannot be excluded. If a lookalike was meant to be a hard boundary rather than a hint, the boundary has to come from a control instead. One mismatch is worth knowing if you build audiences through a tool rather than Ads Manager: the API still asks for a country, and its own reference says that requirement is on the way out too.

Custom audiences: where the data comes from now

This is the layer the cuts did not reach, and the one a small business can still control. Meta's current list holds thirteen sources, split by where the interaction happened.

• –From your own records: website, app activity, catalog, customer list and offline activity. Alongside them sits the exclusion-only list, a variant of the customer list that you are allowed to use only for keeping people out.
• –From engagement on Meta's own apps: video, lead form, Instant Experience, shopping, Instagram account, Facebook Page, events, and on-Facebook listings, which means Marketplace.

Nothing was removed from that list. What changed is how full the browser-based ones get. Meta's own troubleshooting page opens by saying a website custom audience may shrink because it may not include some iOS 14 users, then works through the rest of the gap between traffic and audience: pixel fires are not unique visitors, opted-out accounts never count, and third-party analytics measures sessions rather than people. Its advice when the audience is still too small is to wait until it holds several hundred people, use fewer conditions, or widen the window. That is the closest thing to a size threshold Meta publishes anywhere, and it is advice about one audience type rather than a rule. Retention is a number to check rather than quote: the API reference puts it between 1 and 180 days where it describes the field, then says 365 in its own FAQ further down the same page, and the help pages give no number at all.

The categories Meta closed, and the reason it gave

The 2022 cut was not a performance decision, and reading it as one leads people to look for the replacement interest. Meta framed it as removing options that relate to topics people may perceive as sensitive, and the buckets it named were health, race or ethnicity, political affiliation, religion and sexual orientation. There is no equivalent option waiting somewhere else in the picker, because the objection was to the category existing, not to how well it performed.

A second restriction sits on top of all of this, and it is policy rather than performance. Ads about housing, employment or financial products and services run as special ad category campaigns, and Meta's page is blunt about the cost: age, gender, postal code, exclusion targeting, lookalike audiences and saved audiences are limited or unavailable, some interests vanish from the picker, and a city or pin drop location gets an expanded radius. Advantage+ audience is not offered to these campaigns at all, nor to ads about social issues, elections or politics. Two naming details are worth carrying: the credit category was replaced by financial products and services, and since January 21, 2025 that category is required for advertisers based in the United States or showing such ads to a United States audience. If you lend, hire or rent, this article applies to you with a section removed, and Meta's own advice for what remains is to broaden rather than restrict.

Four mistakes that cost small accounts the most

Narrowing hard because the budget is small

The instinct is to spend thirty dollars a day carefully, so the audience gets cut down to the few thousand people who could plausibly buy. It backfires twice. The delivery system needs conversions to learn from, and a narrow pool produces them slowly, so the ad set sits in the learning phase where results are unstable by design. Then, on the goals listed above, expansion is applied anyway, so the narrowing you agonized over is partly undone by the platform. A small budget is an argument for one ad set, not for a small audience.

Ad sets that quietly compete with each other

Meta calls this auction overlap. When two ads from the same advertiser reach the same auction, only the one with the highest total value is entered and the others are dropped. It does not show up as a bidding war against yourself, which is the confusing part. It shows up as starvation: the losing ad sets underspend, stay learning limited, and read as weak targeting. Running them from separate ad accounts does not help, because Meta says the delivery system may recognize several accounts as one advertiser. Its own remedies are to combine similar ad sets along with their budgets, or switch off the overlapping ones and move the money to the survivor.

Building audiences of a thousand people

A thousand-person audience is not too small because a rule says so. Meta publishes no minimum audience size for an ad set, and the 1,000 figure repeated across the industry could not be found on any Meta page on August 26, 2026. It is too small for a different reason: frequency climbs within days, the ad set cannot gather enough optimization events to stabilize, and you cannot even measure its overlap, since Meta says useful overlap information needs audiences of at least 10,000 accounts. Judge by delivery, not by the estimate under the audience.

Still designing around exclusions

Plenty of accounts are still built on a shape that no longer exists: include interest A, exclude interest B, and split the budget between the halves. That exclusion has been gone since March 31, 2025, and it was pulled out of live campaigns rather than left to run down. If a structure only makes sense with an interest exclusion inside it, it has not been doing what it was drawn to do for over a year. Custom audience exclusions still work, and under Advantage+ audience they are the one input that always binds.

A setup that survives the next round of cuts

The cuts have a direction, and it has held for six years. Everything an advertiser declares about who to reach is drifting toward advisory. Everything derived from data the advertiser owns is staying where it is. A structure built on that asymmetry does not need redrawing every time Meta ships a June release.

1. 1.Put the boundaries you cannot afford to cross into controls rather than suggestions: location, minimum age, language, and the custom audiences you exclude. Everything else in that section is advice to a system that may decline it.
2. 2.Consolidate. One prospecting ad set with a real budget beats four that share an audience and starve each other in the auction, and Meta recommends the same thing on its own overlap page.
3. 3.Keep feeding the sources nobody can deprecate: the customer list you upload, the engagement audiences Meta records on its own apps, and server-side events for the steps where you already hold an email or a phone number.
4. 4.Treat interests as an opening guess and let the system leave them. Before you plan around precision, check whether your optimization goal even permits opting out of expansion.
5. 5.Once a year, open the ad sets nobody has edited since 2024 and look for warning banners. Two of the changes above stopped delivery outright, and a stopped campaign does not send an email.