How lookalike seeds find twins

Seeds decide everything#

Seed quality is the ceiling: purchasers beat visitors, qualified leads beat raw fills, high-value cohorts beat averages. A lookalike of a weak seed is weakness at scale.

Build seeds from one value tier, never blended: payer twins differ from browser twins. Meta own lookalike guide lists the source options. Start at the Meta Ads hub.

Exclude the seed from prospecting where overlap would corrupt reads. Seed construction starts the work.

Rebuild seeds after range overhauls: new products attract new buyer types, and old seeds anchor models to yesterday's customer. Rebuilds cost little and prevent months of drift. Catalogue shifts always trigger a seed review.

The 1,000 to 5,000 band#

Too few starves, too many dilutes: under 1,000 the model guesses, over 5,000 distinctiveness washes out. The 100-person floor builds an audience, not a good one.

Prefer 2,000-plus purchasers where available; supplement with qualified leads only when buyers run short. Quality floor matters more than quantity ceiling. B2B seeding runs leaner by necessity.

Consolidate thin seeds rather than running five tiny lookalikes: merged value tiers often beat fragmented precision. Consolidation feeds learning faster. Consolidation logic applies to seeds too.

Document seed composition with dates: future audits need to know what the model learned from. Ask our team for a seed audit.

Percentage bands that work#

One per cent for precision, broader for scale: start tight, expand only with evidence. Each wider band trades similarity for reach, and the trade is rarely linear.

Test bands in separate ad sets with identical creative: blended bands hide which similarity level pays. Graduate widening on CPA, never on curiosity. Band tests read cleanly.

Pair band tests with creative holds: identical creative isolates the similarity variable, so winners reflect audience quality rather than accidental creative skew. Hold creative constant until bands settle. Creative holds keep band reads honest.

Record band economics beside CPA: CPM and CTR by band reveal whether wider reach cheapens enough to offset lower similarity. Economics decide expansion, not curiosity. Ecommerce economics frame the tradeoff clearly.

Location now follows the ad set, so set geography deliberately per band. Location controls bind the expansion.

Lookalikes versus broad#

Lookalikes guide, broad roams: seeds help thin-signal accounts, while mature signal often lets broad outperform. Test lookalike stacks against broad quarterly, not once.

Feed winning lookalikes back as Advantage+ suggestions rather than permanent structures. Advantage+ audience absorbs proven seeds.

Retire lookalikes that trail broad twice: sentimentality about clever seeds is expensive. Scaling rules reallocate to winners.

Refresh before decay#

Seeds rot as businesses change: refresh quarterly with recent buyers, and rebuild entirely after pivots, repricing or range overhauls. Last year's buyers mislead this year's model.

Version seeds with dates and keep prior versions paused for rollback. Version discipline prevents midnight confusion.

Align seed refreshes with prospecting calendars: new seeds deserve dedicated prospecting budgets, not leftover retargeting change. Fresh models need room to roam before judgement. Prospecting structure gives new seeds a fair trial.

Log seed lineage in campaign notes: parent list, date range, value filter and band tested. Lineage turns mysterious winners into repeatable systems. Setup notes carry the lineage forward.

Watch delivery narrow over time: shrinking reach on fixed budgets signals seed exhaustion. Exhaustion reads distinguish seed decay from creative fatigue.

Special-category limits#

No lookalikes for special ad categories: housing, employment, financial products and social issues exclude them. Plan broad-with-controls instead, and verify current rules per account.

Document the compliance path before structuring: category declarations shape every later choice. Category rules list the constraints.

When in doubt, declare: under-declaration risks disapproval at scale. Meta special-category guide is the authority.

Frequently asked questions#