You have decided to run ads and now the tool wants to know who to show them to. This is a different question from the one you answered when you defined your audience for content, and treating them as the same is where a lot of budget goes to die. Content targeting is about which jobs you serve; paid targeting is about which specific pool of people the platform puts your money in front of, using the mechanisms the ad manager gives you. The pools have names, rules and trade-offs, and picking the wrong one wastes spend before a single creative has had a chance.
This guide covers the four kinds of paid audience, when each fits, the consent and matching caveats that come with the ones built from your own data, and how to sense-check whether an audience is the right size and not quietly overlapping with itself. It is the paid companion to how to define your audience, which is the non-paid, content-first exercise you should do first, because the targeting choices here only make sense once you know who you actually serve.
Core and saved audiences
The default building block is the core audience (sometimes called a saved audience once you store it), which you build by hand from the platform's own attributes: demographics like age and location, plus interests and behaviours the platform infers from what people do.
This is the audience you reach for when you have a clear picture of who you want and no existing data to build from. You describe the person (where they live, roughly how old they are, what they are interested in) and the platform assembles a pool that matches. It is the most manual of the options and the most dependent on your assumptions being right, which is both its weakness and, early on, its point: it is how you reach people who have never heard of you.
Two cautions. First, interest targeting is the platform's inference about a person, not a fact about them, so treat a narrow interest stack as a hypothesis rather than a precision instrument. Second, the more attributes you stack, the smaller and more expensive the pool gets, often without getting better. Over-specifying is a classic way to pay more to reach fewer people for no gain. Meta describes how its targeting attributes work in its own Ads Help Center, and because the available attributes change over time, check the current set there rather than relying on a list.
Custom audiences, and the caveats that come with them
A custom audience is built from people who already have a relationship with you. There are three common sources, and each carries a caveat you cannot skip.
Site and pixel audiences are people who visited your website, built from the tracking pixel or dataset you install. This is one of the most valuable pools you can build, because a site visitor has already shown intent. The caveat is that it depends on the pixel being installed and firing correctly, on consent, and on signal that has been degraded by browser and operating-system privacy changes, so these audiences are smaller and less complete than they were a few years ago.
Engagement audiences are people who interacted with your content on the platform itself: watched a video, opened an interaction, followed you. Because this data lives inside the platform, it does not depend on your own tracking and tends to be more robust than site audiences.
Customer-list audiences are built by uploading contact details you already hold and letting the platform match them to accounts. This is powerful and it is also the one with the sharpest obligations: you must have a lawful basis and the appropriate consent to use those contacts for advertising, and matching is imperfect, so the audience the platform produces is always smaller than the list you uploaded. Never upload a list you do not have permission to use this way. Meta sets out its rules for custom audiences, including customer lists and the consent you are responsible for, in its Business Help Center; read the current terms there before you upload anything, because this is the area where getting it wrong has consequences beyond wasted budget.
Lookalike audiences
A lookalike audience is one the platform builds for you: you give it a seed (a custom audience of your best customers, say) and it finds a much larger pool of people who resemble that seed on the signals it can see. The lookalike audience is how you scale beyond the people who already know you without abandoning targeting entirely.
It fits when you have a good seed and want to grow. The quality of a lookalike is almost entirely the quality of its seed: a lookalike of your highest-value customers is a genuinely useful pool, while a lookalike of everyone who ever visited your site is barely different from broad targeting. The seed also needs to be large enough for the platform to find a pattern in it. Too small a seed and the resemblance is noise. Meta documents the seed requirements and how lookalikes are built in its Ads Help Center, and the thresholds change, so confirm the current minimums there rather than trusting a remembered figure.
Broad and Advantage+ audiences
The newest and, increasingly, the default approach is to define very little and let the delivery system find the audience for you. On Meta this is broad targeting combined with Advantage+ audience, where instead of tightly specifying a pool you give the system light signals (or none) and let its optimisation do the finding.
This fits better than it used to, because the delivery systems have got better at it and the privacy changes that weakened manual targeting affected the system's own optimisation less than they affected your hand-built audiences. It fits especially well when your creative is strong, because a good creative gives the system a clear signal to optimise around, and when your budget is large enough for the system to explore. It fits worse when your budget is small, when you have a genuine reason to exclude part of the population, or when a regulated category limits how you are allowed to target. Meta explains how Advantage+ audience works and how your inputs are treated as signals rather than strict rules in its Ads Help Center; check there for how it currently behaves, because this is one of the fastest-moving parts of the tool.
Sense-checking size and overlap
Two checks will save you from the most common self-inflicted audience problems.
Size. An audience can be too small to deliver efficiently or too large to be meaningfully targeted, and the platform gives you an estimated size as you build. Treat that estimate as a rough band, not a precise count: the platforms are explicit that it is an estimate, and you should not build a plan on the exact number. Too small and the system struggles to spend and to exit its learning phase; too large and you may as well have gone broad. If a hand-built audience is coming out tiny, that is usually a sign you have over-stacked attributes.
Overlap. When you run several audiences at once, they can overlap (the same people sitting in more than one ad set), and then your own ad sets bid against each other in the auction, raising your costs for no benefit. The fix is to keep audiences genuinely distinct, exclude one from another where they would otherwise collide (your retargeting pool excluded from your prospecting pool, for instance), and check overlap before you launch rather than diagnosing it afterwards from strange delivery. Meta provides an audience overlap check and documents how to read it in its Business Help Center.
Match the audience to the funnel stage
The right audience depends on where in the funnel the campaign sits, which is the through-line to the paid-social strategies brands use.
Cold prospecting (reaching people who do not know you) is core, lookalike or broad. Warm and hot retargeting (reaching people who engaged or visited) is custom audiences. Getting this backwards is a frequent and expensive error: retargeting looks wonderfully efficient because you are advertising to people already close to buying, some of whom would have bought anyway, so scaling spend onto it on the strength of that reported efficiency is one of the most common misallocations in paid social. Reading ad results without fooling yourself explains why that efficiency is partly an illusion.
Where Acumin fits
Acumin's role here is upstream of the ad manager. Snapshot reads your public organic numbers, which is how you find the pieces and the segments your real audience actually responds to: the evidence that should shape which seed you build a lookalike from and which interests are worth testing. The Content Strategy cockpit holds the audience definition that all of this serves, so paid targeting is an expression of a decision rather than a guess made inside the ad tool. And Studio is where the creative for each audience gets made, because a prospecting audience and a retargeting audience usually want different creative, not the same ad shown twice.
The honest limit, stated as it is in the product: automatic read-back of Meta ad results is pending Meta's app review. Until then, Acumin helps you prepare and draft while you read results and send campaigns in Meta's own tools. Acumin drafts; you send.
Do this today
Sketch three audiences on one page before you open the ad manager: one cold pool for prospecting (core, lookalike or broad), one warm pool for retargeting (site or engagement), and the seed you would build a lookalike from. Beside each, write the funnel stage it serves and, for anything built from your own data, one line confirming you have consent to use it. Then, when you build them in the tool, run the overlap check before you launch. That single page turns audience selection from a menu you click through into a decision you can defend.