How Meta audience targeting has changed
For years, Meta advertisers built success around narrow interest stacks — combining dozens of interests, behaviors, and demographics to define a precise audience. That approach has steadily lost effectiveness as Meta's algorithm has gotten better at finding converters within broad audiences than advertisers are at defining them manually.
In 2025, the highest-performing accounts use targeting primarily to define audience pools (who should never see this, such as existing customers in an acquisition campaign) and to provide quality signals (Custom Audiences, Lookalikes, Advantage+ inputs) — rather than to micro-target who should see an ad.
This does not mean targeting is dead. It means the skill has moved from "build the perfect interest combination" to "structure your audiences and signals so the algorithm has the best possible inputs to work with."
The three core audience types and what they are for
Saved Audiences (interest and demographic targeting) are now mainly useful for cold prospecting tests and for categories where Meta's signal data is thin — niche B2B products, for example. For most consumer products, broad targeting with strong creative outperforms narrow interest stacks.
Custom Audiences are built from your own data: website visitors (via the Meta Pixel or Conversions API), customer lists (via CRM upload), app activity, or engagement with your Facebook/Instagram content. These are your highest-quality audiences because they reflect real behavior, not inferred interest.
Lookalike Audiences use a Custom Audience as a "seed" and find people who share characteristics with it. A Lookalike built from your last 90 days of purchasers, sized at 1–3%, is typically the strongest acquisition audience available — assuming your seed audience has enough volume (Meta recommends at least 100 people, ideally 1,000+).
- Saved Audiences: cold testing, niche categories, early-stage accounts with little pixel data
- Custom Audiences: retargeting, exclusions, and as seeds for Lookalikes
- Lookalike Audiences: scaling acquisition once you have a quality seed audience of 1,000+ people
Advantage+ audiences: what they are and when to use them
Advantage+ audience (formerly "Advantage Detailed Targeting" and now the default for many campaign types) lets Meta expand beyond your specified targeting when it believes doing so will improve results. Advantage+ Shopping Campaigns (ASC) go further, combining automated audience finding, placement, and creative optimization into a single campaign type built for e-commerce.
Advantage+ works best when you give it strong signals to learn from: a Custom Audience exclusion list (so it does not waste spend on existing customers in an acquisition campaign), a well-tagged conversion event, and creative that performs across a wide range of audiences rather than one narrow segment.
The most common mistake with Advantage+ is treating it as "set and forget" with weak inputs. If your pixel is firing inconsistently, your conversion event is misconfigured, or your creative only resonates with one narrow group, Advantage+ will optimize toward whatever signal it has — including the wrong one.
- Always exclude existing customers from acquisition-focused Advantage+ campaigns using a Custom Audience
- Verify your primary conversion event (Purchase, Lead, etc.) is firing accurately before scaling an Advantage+ campaign
- Feed Advantage+ multiple creative variations — it needs options to find what resonates with each micro-segment it discovers
- Give new Advantage+ campaigns at least 4–7 days and the platform's minimum recommended budget before judging performance
Building a layered audience strategy across the funnel
A layered structure separates budget by funnel stage so each audience type gets evaluated on the right metric. Top-of-funnel (cold/broad or Lookalike) should be judged on cost per landing page view or add-to-cart, not immediate ROAS. Mid-funnel (engagement-based Custom Audiences) bridges interest to intent. Bottom-of-funnel (website Custom Audiences, cart abandoners) should carry the highest ROAS expectations since they are closest to conversion.
A simple, durable structure for most accounts: one broad or Advantage+ campaign for cold acquisition, one Lookalike-based campaign for scaling, and one retargeting campaign built from 7-, 14-, and 30-day website Custom Audiences with messaging tailored to where in the journey the visitor dropped off.
Avoiding audience overlap and fatigue
Audience overlap happens when multiple active ad sets target audiences with significant shared membership, causing them to compete against each other in the same auctions and inflate CPMs for both. Meta's Audience Overlap tool (in Audiences under Assets) shows the percentage overlap between any two saved audiences — anything above 20% is worth investigating.
Overlap is most common between a broad/Advantage+ campaign and a Lookalike campaign, since Lookalikes are themselves drawn from a broad population. If you see CPMs climbing without a clear external cause, check overlap before assuming it is seasonal or competitive pressure.
Creative fatigue compounds this problem: the same audience seeing the same ad repeatedly drives frequency up and relevance down, which Meta's delivery system penalizes with higher costs. Refresh creative every 2–4 weeks for always-on campaigns, and watch frequency — above 3–4 for a retargeting audience within a week is usually a signal to refresh.
Using Claude to monitor audience performance
With MCP Ads connected, you can ask Claude to do the cross-campaign analysis that is tedious to do manually inside Meta Ads Manager. For example: "Compare CPM and frequency across all active ad sets for the last 14 days. Flag any ad set where frequency is above 4 or CPM has increased more than 25% compared to the prior 14 days."
You can also ask Claude to break down performance by audience type: "Group my active campaigns by audience type — broad/Advantage+, Lookalike, and retargeting — and show me cost per result and ROAS for each group over the last 30 days." This gives you the funnel-stage view that Meta's native reporting does not surface directly.
