What is a lookalike audience? It is a group of new people an ad platform finds because they statistically resemble a "seed" list of your existing customers. You hand Meta (or Google, TikTok, LinkedIn) a list of buyers, leads, or engaged users. The platform studies what they have in common, then goes and finds strangers who match the pattern. Same playbook, fresh faces. The quality of your seed decides everything.
What is a lookalike audience, in plain English?
Retargeting goes after people who already know you. Lookalike audiences do the opposite: they go after people who have never heard of you but behave a lot like the people who already buy from you.
Here is the mechanic. You upload a "seed" audience, usually built from first-party data: your customer list, your purchasers, your highest-value leads, or a chunk of people who engaged with you somewhere. The ad platform reverse-engineers what those people have in common. Not just age and zip code. Device type, browsing patterns, purchase history, what they watch, how long they linger. Then it scans its entire user base and pulls out the strangers who light up the same signals.
That output is your lookalike audience. It is the engine behind cold prospecting on paid social, because it lets you target "people like my customers" without manually guessing at interests and demographics you will probably get wrong.
The thing nobody tells you up front: a lookalike is only as smart as the seed you feed it. Model a list of 500 actual buyers and you get a sharp, high-intent audience. Model 10,000 random newsletter signups and you get a blurry crowd of people who like free things. Garbage in, garbage out, at scale.
How lookalike audiences are built
Three inputs decide whether a lookalike works or flops.
1. The seed (this is the whole ballgame)
The seed is your source list. Best practice, in order of quality:
- Purchasers and high-LTV customers. Model the people who paid you, ideally your most valuable ones.
- Qualified leads or trial signups.Strong second choice for B2B, where the purchase is slow.
- High-intent engagement. Add-to-cart, video completions, time-on-site. Useful when your buyer list is thin.
- Newsletter or top-of-funnel lists. Weakest seed. These people opted into content, not your product. Use only as a last resort.
Meta's technical floor is roughly 100 people from a single country. Ignore the floor. The practical recommendation is 1,000 to 5,000 quality records before the model has enough pattern to work with, and the signal keeps sharpening up to around 25,000 before diminishing returns set in. A clean list of 1,000 buyers beats a sloppy list of 50,000 contacts every time.
Two details that quietly decide quality. First, how the seed gets in. A list upload is a static snapshot that ages the moment you hit save; a server-side feed through a conversion API keeps the seed fresh and matches better in a privacy-restricted, post-cookie world. Second, freshness itself. A seed of buyers from the last 90 to 180 days models a sharper picture than a three-year-old export of everyone who ever bought, because customer behavior drifts. Rebuild the seed on a schedule instead of treating it as set-and-forget.
2. The similarity tier (how close a match)
On Meta you pick a percentage from 1% to 10% of a target country's users.
- 1% = the closest, most concentrated match. The "twins." Smaller reach, highest intent. Start here.
- 5-10% = a wider net. More reach, looser resemblance. The "distant cousins." Use to scale only after 1% proves out.
Bigger is not better. Going straight to 10% is how budgets get torched on people who barely resemble your buyers.
One move most people skip: layer an exclusion. Tell the platform to remove your existing customers and recent converters from the lookalike, so prospecting budget finds new people instead of re-buying traffic you already own. The lookalike finds the look; the exclusion keeps it pointed at strangers.
3. The platform's modeling engine
Each platform runs its own algorithm. Meta calls the output a Lookalike Audience. Google calls its equivalent "similar segments" inside its broader audience system. TikTok and LinkedIn ship their own versions. The math differs, the principle is identical: find me more of these. What changes platform to platform is the depth and type of behavioral data feeding the model, which is exactly why the same seed performs differently across channels and why you test rather than copy-paste one audience everywhere.
Lookalike vs. custom audience: not the same thing
People mix these up constantly. They are opposites.
| Custom Audience | Lookalike Audience | |
|---|---|---|
| Who it targets | People who already know you | Strangers who resemble your customers |
| Built from | Your data directly (list, pixel, engagement) | A custom audience used as a "seed" |
| Funnel stage | Retargeting / re-engagement | Cold prospecting / acquisition |
| Goal | Convert warm traffic | Find new warm-shaped traffic |
A custom audience is the people on your list. A lookalike audience is the model built from that list to find new people. You almost always create the custom audience first, then spin a lookalike off it. If you want to re-engage people who already touched your brand, you want retargeting, not a lookalike.
Where lookalikes earn their keep
Lookalikes live on paid social. That is their home turf, because Meta and TikTok sit on enormous behavioral datasets that make the modeling genuinely good.
They shine when:
- You have a clean, sizable seed of real buyers (not just leads). Ecommerce and DTC brands usually have this in spades, which is why lookalikes are a staple of their prospecting.
- Your product has a repeatable customer profile worth cloning.
- You are trying to scale past the ceiling of manual interest targeting.
They struggle when:
- Your seed is tiny, stale, or full of low-intent contacts. This is the trap for B2B, where the real buyer list is small and slow to grow, so seeding off qualified leads or trial signups often beats waiting for enough closed deals.
- Your audience is so niche that "people like them" barely exist.
- You are in the EU, where privacy rules and Meta's targeting changes have narrowed how lookalikes can be built and used. Validate the current rules for your market before you bank on them.
One honest 2026 note: on accounts with strong conversion volume, Meta's automated Advantage+ audiences now often match or beat manually built lookalikes. The smart move is increasingly to feed your customer list in as a signal and let the system expand it, rather than hand-building a 1% lookalike and walking away. Lookalikes are not dead. They are just no longer the only tool, and on high-volume accounts they are sometimes not the best one. A good operator tests, does not assume.
Whichever way you go, judge it on the right number. A lookalike that wins on cheap clicks but loses on cost per acquisition is not winning, and attribution settings change the verdict, so set your attribution window deliberately before you call a tier a success or a flop. If prospecting can't rely on a list at all, the alternative is contextual targeting: reaching people by the content they are consuming rather than who they resemble.
Stop guessing at interests. Clone your best customers instead.
Most agencies treat lookalikes as a checkbox: upload a list, pick 1%, walk away. That is how you end up paying to reach people who like free newsletters. We build the seed right, set the exclusions, test the tiers, and know when a manual lookalike beats Advantage+ and when it doesn't. No guessing, no jargon, no junior running your spend.
See how we run paid social, check what it costs on our pricing page, or browse the rest of the glossary. Want a straight read on your prospecting setup? Get in touch. No pressure and no jargon, just the real picture.
