What Is Seed Audience?
A seed audience is the source list of high-value users an ad platform uses to build a lookalike or similar-audience expansion. Common seeds: past purchasers, high-LTV customers, qualified leads, high-intent site actions, and email subscribers. The quality of the lookalike is bounded by the quality of the seed; a large but weak seed produces a large but weak lookalike.
Worked example
A Bangladesh SaaS brand builds three Meta lookalikes: (1) all 12,000 signups from the last 180 days, (2) 4,500 paying customers, (3) 900 customers whose LTV exceeds their CAC by 3x or more. All three cross the Meta minimum, but the 1 percent lookalike from seed 3 converts at 2.4x the CPA of the lookalike from seed 1. Same audience size, very different results, because the seed was tuned for value, not for volume.
Benchmarks
- Meta lookalike minimum seed size: 100 users, but 1,000 to 50,000 is where quality becomes reliable.
- LinkedIn lookalike minimum seed size: 300 members in the source list.
- Google Similar Audiences (customer match): now retired for most use cases; use Customer Match + Smart Bidding signals instead.
- Typical lift from a value-scored seed vs a volume seed: 30 to 100 percent lower CPA on the same 1 percent lookalike, based on standard case studies.
Meta's own advertiser guidance is explicit that lookalike quality depends more on the value of the seed than the size, recommending seeds of 1,000 to 50,000 of the highest-value customers rather than all customers indiscriminately.
Why it matters
Prospecting scale in 2026 depends on giving the algorithm a clean signal. A well-scored seed is the highest-leverage input a media buyer controls; it tells Meta or LinkedIn what a 'good' customer looks like so the delivery system can find more of them. Everything downstream, targeting, bid strategy, creative testing, sits on top of that seed.
Common mistakes
- 1.Using every signup or every site visitor as the seed; volume wins on size but loses on quality.
- 2.Rebuilding the same lookalike weekly; Meta refreshes automatically, manual rebuilds reset the learning phase without new information.
- 3.Not layering seeds by geography or product line; a global seed produces a global lookalike that ignores local buying patterns.
- 4.Feeding a seed built entirely on discount-code buyers; the resulting lookalike converts on discount and never at full price.
Put Seed Audience to work
Free calculators
Related services
FAQs about Seed Audience
How big should my seed audience be?
Meta officially accepts seeds from 100 users, but reliable results start around 1,000 and lookalike quality plateaus somewhere between 10,000 and 50,000. Prefer 1,000 to 5,000 of your highest-value customers over 100,000 mixed ones.
What makes a 'good' seed?
Value density, high LTV, high purchase frequency, or high qualified-lead-to-close rate. If you can score customers, seed on the top 20 to 30 percent by value, not on the full list.
Can I use website visitors as a seed?
Yes, but filter aggressively, high-intent actions only (pricing page, add-to-cart, demo-request), and de-duplicate against existing customers so you are seeding new-user acquisition, not remarketing.
How often should I refresh the seed?
Every 30 to 90 days for most accounts. Meta rebuilds the lookalike automatically off the underlying custom audience; the manual refresh is really about updating the source rules (LTV threshold, date window).
Does seed quality matter more than lookalike percentage?
Yes. A 1 percent lookalike off a value-scored seed usually outperforms a 3 percent or 5 percent lookalike off an unscored seed. Fix the seed first, expand the percentage later.
Related terms
New audience Meta/LinkedIn builds from a high-value seed list.
Meta audience built from your customer, pixel, or engagement data.
Users who already know your brand, site visitors, email list, engagers.
Total gross profit a customer generates across their relationship.
Google audience built from your hashed first-party email/phone list.
Meta's AI-driven campaign setup that automates targeting and placements.