Lookalike audiences used to be the closest thing to a cheat code in digital advertising. Upload your customer list to Meta, let the algorithm find people who look like your best buyers, and watch acquisition costs drop 30-50% compared to interest-based targeting. That was 2019. In 2026, the playbook still works, but the rules have changed. Signal loss from Apple's ATT, the deprecation of third-party cookies, and platform shifts have forced every major ad platform to rethink how lookalike targeting operates.
The core concept remains the same: you give a platform a "seed" audience of your existing customers, and the platform's algorithm finds new users who share similar attributes. But how each platform builds that match, and how effective each one is after years of privacy changes, varies significantly.
Lookalike Audiences by Platform
| Feature | Meta Lookalike Audiences | Google Similar Segments | Amazon Lookalike Audiences |
|---|---|---|---|
| Seed source | Customer lists, pixel events, app events, page/profile engagement | Customer Match lists, conversion audiences, GA4 audiences | Brand halo audiences, ASIN remarketing, purchase-based audiences |
| Data signal | Demographics, interests, behaviors, on-platform activity (weakened by ATT for off-platform signals) | Search history, YouTube behavior, Gmail signals, Maps activity, app usage | Purchase history, search queries, browsing patterns (all first-party, unaffected by ATT) |
| Size control | 1-10% of country population (1% = most similar, 10% = broadest) | No manual size control. Google determines optimal reach. | Limited control. Amazon auto-sizes based on seed and campaign goals. |
| Minimum seed size | 100 people (1,000+ recommended for quality) | 1,000 active users in the audience list | Varies by audience type. Generally requires meaningful purchase volume on the ASIN. |
| ATT impact | Significant. iOS signal loss reduced match quality for app-based and pixel-based seeds. Conversions API (CAPI) partially compensates. | Moderate. Google's cross-property data (Search, YouTube, Gmail, Android) provides more signals than Meta post-ATT. | Minimal. Amazon's data is entirely first-party, collected within its own ecosystem. ATT doesn't affect it. |
| Best for | DTC e-commerce, app installs, lead gen. Still the largest social audience for prospecting. | Cross-channel prospecting (Search, YouTube, Display, Gmail). Works well with high-intent signals. | Product-specific targeting on Amazon. Reaching shoppers who buy from competitors or related categories. |
| Typical performance vs. interest targeting | 20-40% lower CPA with quality seed lists (industry case studies). Gap has narrowed post-ATT. | Comparable or better than affinity/in-market segments. Performance varies by vertical. | Strong for Sponsored Display and DSP. Amazon's purchase-based signals are the most conversion-predictive data available. |
How Signal Loss Changed Lookalike Targeting
Before ATT, Meta's lookalike audiences were built on a massive web of cross-app and cross-site behavioral data. The pixel tracked what users did on millions of websites and apps, and Meta used all of that to find patterns among your best customers. When 75% of iOS users opted out of tracking, that data pipeline shrank dramatically.
The practical effect: lookalike audiences on Meta are still effective, but the gap between lookalikes and broad targeting has narrowed. In 2019, a 1% lookalike might outperform broad targeting by 50-60% on CPA. In 2026, the performance gap between iOS and Android campaigns has narrowed as platforms have adapted their modeling, but a meaningful difference persists for many advertisers. Meta's own Advantage+ Audience feature essentially treats your targeting inputs (including lookalikes) as "suggestions" and lets the AI expand beyond them. For some advertisers, Advantage+ broad targeting now matches or beats hand-built lookalike audiences.
Google felt less impact because its signal comes from first-party properties (Search, YouTube, Gmail, Android), which aren't affected by ATT. Amazon felt almost no impact because its entire data ecosystem is first-party. If you browse and buy on Amazon, Amazon sees everything regardless of your iOS privacy settings.
Building Better Lookalike Audiences
The quality of your lookalike depends entirely on the quality of your seed. Here's how to build seeds that produce results:
- Use purchase data, not just pixel events. A seed built from actual buyers will outperform one built from "add to cart" events or page visitors. The more downstream the conversion event, the stronger the signal for the algorithm.
- Segment by customer value. A lookalike built from your top 20% of customers (by LTV or order value) will find higher-quality prospects than one built from all customers. Upload a list of your best 500-1,000 buyers, not your entire database of 50,000.
- Refresh seeds regularly. Customer behavior changes. A seed list from 12 months ago reflects who your customers were, not who they are. Update seeds quarterly at minimum, monthly if you can.
- Test seed sources against each other. Run the same campaign with a lookalike from purchasers vs. one from email subscribers vs. one from high-LTV customers. The performance difference can be 30-50%.
- Go smaller on percentage. On Meta, a 1% lookalike (roughly 2.1 million people in the U.S.) is almost always more efficient than a 5% or 10% lookalike. Start with 1% and only expand when you've saturated that audience.
When to Use Lookalikes vs. Broad Targeting
The shift toward AI-powered broad targeting has created a real debate: are lookalike audiences still worth building?
The answer depends on your situation. Lookalikes still outperform broad targeting when your seed is high-quality and your audience is specific (niche B2B, luxury products, specialized services). Broad targeting performs equally well or better when your product has mass-market appeal, your pixel has extensive conversion data (1,000+ events per month), and the platform's AI has enough signal to optimize without audience constraints.
The pragmatic approach: test both. Run a lookalike campaign and a broad targeting campaign with identical creative and budget. After 2-3 weeks with sufficient conversion volume, compare CPA and ROAS. Let the data decide rather than defaulting to either approach.
Example: Lookalike vs. Broad Targeting Head-to-Head
A DTC skincare brand running $60K/month on Meta tested their 1% purchase-based lookalike audience against Advantage+ broad targeting with identical creative and budget. After three weeks, the results were closer than expected: the lookalike delivered a $38 CPA while broad targeting came in at $41. In 2019, that gap would have been 40-50%. Post-ATT, it was 8%. But when they segmented by product line, lookalikes still won for premium products ($100+ AOV) while broad targeting performed better for the entry-level line ($25 AOV). The takeaway matched a broader trend: Meta's Advantage+ Shopping campaigns generate $4.52 in revenue for every dollar spent (Meta Q3 2024), blending both lookalike and broad targeting signals. The practical answer isn't one or the other — it's using lookalikes as audience suggestions within Advantage+ campaigns and letting the algorithm decide how much weight to give them.
Frequently Asked Questions
Are Meta Lookalike audiences still worth using in 2026?
Yes, with caveats. They're still effective, especially with high-quality purchase-based seed lists and Conversions API (CAPI) data flowing. But the performance advantage over broad targeting has narrowed. Test lookalikes against Advantage+ broad targeting for your specific account. Many brands find the best results come from using lookalikes as "audience suggestions" within Advantage+ campaigns rather than as standalone targeting.
What's the ideal seed list size?
For Meta, 1,000-5,000 high-quality records (actual purchasers with email and phone for matching) is the sweet spot. Larger seeds dilute quality. Smaller seeds don't give the algorithm enough patterns to find. For Google, aim for 1,000+ matched users as the minimum for Similar Segments to activate.
Can I use lookalike audiences for B2B?
On Meta, B2B lookalikes are hit-or-miss because Meta's user data is consumer-oriented. A seed list of "IT directors who bought enterprise software" may produce a lookalike of people who share consumer behaviors with those IT directors, not necessarily other IT directors. LinkedIn's matched audiences tend to perform better for B2B because the professional data is richer. On Google, Customer Match lookalikes can work for B2B if your seed is large enough and your conversion tracking captures the right events.
How do Amazon lookalike audiences compare to Meta?
Amazon's advantage is data quality. The signals are based on actual purchases, not inferred interests. The disadvantage is reach. Amazon's audience is limited to Amazon shoppers, while Meta reaches nearly every internet user. For brands selling on Amazon, Amazon's lookalikes (through DSP) are often the most efficient targeting option because the purchase-intent signal is so strong.