Should I use Advantage+ or manual targeting on Facebook Ads?

Test both. Advantage+ works well with strong creatives and enough pixel data (50+ conversions per week). Manual targeting gives you more control during early testing when your pixel is still learning.

What size should my Facebook Ads audience be?

For cold traffic, aim for 2-10 million people. Under 500K starves the algorithm. Over 20M works fine with Advantage+ but may waste budget with manual targeting on small budgets.

Are Facebook Lookalike Audiences still effective in 2026?

Yes, when built from strong sources. Use purchasers or high-value customers as your seed, not page likes or video viewers. Start with 1% for quality, expand to 3-5% for scale.

How many interests should I stack in one ad set?

Three to five related interests per ad set. More than that narrows your audience too much. Facebook's AND logic means stacking interests shrinks your pool fast.

Field Notes Updated May 2026 11 min read

Facebook Ads Targeting in 2026: What Still Works After 5 Years of API Restrictions

Detailed targeting got gutted, lookalikes got vaguer, and broad keeps winning. Here's the targeting toolkit that's actually viable now.

Targeting on Facebook in 2026 isn't what it was in 2019. Apple killed precise pixel signal. Meta gutted detailed targeting categories. Lookalikes work but are less specific. The targeting playbook needs to fit the new reality.

The four buckets that still work

Broad targeting — minimal constraints, let the algorithm find converters
Lookalikes — modeled audiences off your existing data
Custom Audiences — your own data uploaded
Interests + behaviors — what's left of detailed targeting

Broad targeting: usually the right answer at scale

For accounts spending $2k+/day with strong pixel signal, broad targeting (geo + age range only, no interests) often beats narrow targeting. The algorithm has enough conversion data to find converters within a wide audience.

When it works:

You have 100+ conversions/week (algorithm has signal to optimize)
Pixel + CAPI implemented properly
Creative does audience self-selection through specific hooks

When it fails:

Low-volume accounts (algorithm can't find pattern)
Niche offers where most users aren't prospects (e.g. enterprise B2B)
Weak pixel signal (no CAPI, blocked tracking)

Lookalikes: the workhorse

Best lookalike seeds, in priority order:

Past purchasers (last 365 days, top 25% AOV)
Past purchasers (all)
Add to Cart events
Email subscribers (CRM upload)
High-engagement video viewers

Test 1%, 2%, 3%, 5% in parallel. Higher percentages = more reach, lower precision. The sweet spot for most accounts is 1-3%.

Refresh seed data monthly. Lookalikes built off year-old purchase data are weaker than fresh ones.

Same offer tested across 4 targeting types. Each has its place — no single winner.

Custom Audiences: best efficiency, worst scale

Customer email lists, site visitors, video viewers, lead form submitters. Smaller pools but higher conversion intent.

Sweet spots:

Site visitors 30/60/90 days (separate ad sets per window)
Cart abandoners 7d
Video viewers 50%+ on cold creative (warm audience proxy)
Past customers excluded from cold (always)
Past customers included for upsell campaign

Interest targeting: limited but useful

Most interest categories got broader or removed in 2022-2024. What remains:

Brand interests (specific competitor brands)
Major publication interests (industry magazines, blogs)
Behavior categories (purchase behavior, device usage)

Stack 3-5 interests in one ad set rather than running each as a separate ad set. Keeps audience size large enough for the algorithm to optimize.

FAQ

Should I narrow targeting for higher quality?

Counter-intuitively, no — narrow targeting often LOWERS quality because the algorithm can't optimize. Let it find converters in a wider pool.

Are lookalikes still useful?

Yes, especially for accounts under $5k/day. Above that, broad usually wins.

Should I exclude past customers from cold?

Always. Otherwise you pay cold prices for retargeting they'd have done anyway.

Bottom line

Broad + LAL + CA covers 90% of needs. Detailed interests as a fourth layer only when justified. Stop trying to micro-target — the algorithm finds patterns better than you do at any meaningful scale.