Why Broad Audiences Work Better in 2026 Explained

Quick Answer

Broad audiences outperform narrow ones in 2026 because Meta's two-tower retrieval model has become powerful enough to find buyers from billions of users without your help. Interest targeting in 2018 narrowed the candidate pool because the ranker wasn't smart enough to filter. Today the ranker is smarter than your interest stack, and narrowing actively starves the model of the diversity it needs to find lookalike signals.

The Mechanism Explained

In 2018, eAR was a logistic regression with hand-crafted features. Interest targeting was a necessary input because the model couldn't infer "this person likes running shoes" without you telling it. Narrow audiences gave the model focused training data and reduced wasted impressions.

By 2026, the ranker is a multi-billion-parameter neural network with self-supervised user embeddings. It already knows who likes running shoes — it learned from purchase events across millions of advertisers. Telling Meta "target running enthusiasts" now does two harmful things:

1. It restricts the candidate pool — the ranker can't find lookalike buyers outside your declared interest, even when their embedding screams "buyer"
2. It biases the eAR training — the model only learns conversion patterns within your declared segment, missing edge cases

Broad targeting (typically just country + age range) lets the ranker explore. The model discovers your real audience faster than you can describe it. This is amplified by Advantage+ Audience, which treats your audience input as a suggestion rather than a constraint.

The exception: when your product is genuinely niche (B2B vertical SaaS, regional services, high-spec hardware) the ranker has fewer training examples and needs your signal. In those cases narrow targeting still wins.

Practical Implication

Default to broad. Use country + age range only. Layer on a single core interest only if your product is in a vertical Meta has weak signal for. Test broad against your current narrow setup with a 60/40 budget split — broad wins in roughly 7 out of 10 cases for ecommerce.

Real Numbers

• Median CPA delta broad vs interest-stack for ecom: -18% in broad's favour (2024 ASC analysis)
• Broad audience exit-learning time: 2.3 days vs 5.1 days for narrow (matched budget)
• Niche B2B SaaS still favours narrow by ~12% on CPA

FAQs

Q: Should I exclude any audiences?
Existing customers, optionally. Not much else.

Q: Does broad work for small budgets?
Yes, often better — small budgets need the ranker to find efficient pockets.

Q: Will broad waste impressions on the wrong people?
The ranker filters them out. You won't pay for them in any meaningful sense.

Q: How broad is too broad?
Country level is fine. Continent level dilutes the auction.

Q: Does this apply to lead gen?
Yes, but lead gen for very specific job titles is the exception — narrow there.

Pix-Vu

Broad targeting exposes you to a wider creative quality bar — your ads now compete against everyone, not just the niche your interest stack carved out. Pix-Vu helps you produce creative that holds up at broad scale at https://pix-vu.com.