Facebook Lookalike Audiences in 2026: What Still Works After 5 Years of Quality Decay

Lookalikes lost some precision after iOS 14, but they're still the most reliable warm audience. Here's the seed selection and percentage strategy that produces winners now.

Lookalikes are not what they were in 2019. The deterministic pixel signal that used to power them is mostly gone for iOS users. The audiences are vaguer and the lifts smaller. But they still beat broad and beat interests in most accounts when set up right.

Seed selection (the part that matters most)

The lookalike is only as good as its seed. Best seeds in priority order:

1. Top-25% AOV purchasers (last 365 days) — your most valuable customers. LAL of these finds prospects with similar value-buying behavior.
2. All purchasers (last 365 days) — broader signal, more volume.
3. Email subscribers from CRM — uploaded as Custom Audience, used as LAL seed.
4. High-value events (Subscribe, Trial Started) — for SaaS / subscription businesses.
5. Add to Cart events — broader, lower precision, useful for high-volume needs.
6. Video viewers 75%+ — engagement signal when other seeds are too small.

Don't use Page Fans as seeds — they're mostly old organic followers, terrible commercial signal.

Percentage selection

1% LAL: highest precision, smallest reach (about 2.5M people for US). Use for cold prospecting at smaller spend.

2-3% LAL: balanced. The sweet spot for most accounts.

5-10% LAL: maximum reach, lowest precision. Use only at very high spend ($20k+/day) when you've exhausted smaller percentages.

Test 1%, 2%, 3% in parallel as separate ad sets. Each gets its own budget, same creative. Whichever produces the best CPA gets scaled.

Refresh schedule

Custom Audience seeds drift. Refresh on this cadence:

• Lookalikes off rolling Custom Audiences (auto-update): refresh seed monthly
• Lookalikes off uploaded CRM lists: re-upload list quarterly
• After major creative campaign: regenerate engagement-based lookalikes

Common mistakes

• Using Page Fans as seed (mostly junk signal)
• Building LAL off too-small seeds (under 100 events = noise)
• Stacking 4+ different LALs in one ad set (audience too vague)
• Not refreshing seeds for 6+ months (data decay)
• Stopping at 1% only (limits reach unnecessarily)

FAQ

Should I exclude existing customers from LAL ad sets?

Yes — they're already yours. Pay retargeting prices, not cold.

Are value-based lookalikes worth it?

For e-commerce with mixed AOVs, yes — lift of 10-20% over standard LALs. Need clean value data flowing into pixel.

Bottom line

Lookalikes still work. Best seeds are top-AOV purchasers. Test 1%-3% in parallel. Refresh monthly. Skip 10% LALs — at that size you're running broad without admitting it.