A lookalike audience is the platform finding people who resemble a seed list you provide. What you put in the seed determines what comes out.
The seed is the whole game
The platform learns the shared characteristics of your seed and finds people like them. Seed it with "people who click ads" and that is what you get back — which can be a different population from people who spend.
Good seed: users who purchased repeatedly or rank in top LTV
Risky seed: users who only installed or only signed up
This is a frequent explanation for cheap CPI alongside below-average LTV.
Size matters too
Too small and no pattern emerges; too large and your users' distinctiveness washes out. If narrowing to top-LTV users underperforms, suspect seed volume before seed quality.
Users who purchased repeatedly or sit in your top [LTV](/glossary/ltv) band. Seeding from installs or signups alone produces resemblance to 'people likely to click an ad', which is a common cause of cheap-looking CPI with below-average LTV.
What happens if the seed is too small?
The platform cannot find a pattern and expansion stalls. Too large and what makes your users distinctive gets diluted, so underperformance from a top-LTV seed may be a volume problem rather than a quality one.