Meta Ads

Lookalike Audiences: How to Build One That Works

The quality of a lookalike comes from the list you feed it, not from the algorithm. A bad source scales a bad audience very fast.

A lookalike audience is a new audience built from users who behave like a source list you provide. The platform finds the shared patterns in your source and lets you target people who match those patterns but do not know you yet. The critical point is this: quality comes from your list, not from the algorithm. A weak source scales a weak audience very quickly.

How does a lookalike audience work?

The process has three stages. First you provide a source audience: a customer list, purchasing site visitors or app users. The platform extracts the shared signals of that group. Then it ranks users in your chosen country by similarity to that pattern and hands you the top slice as an audience. The percentage is precisely the size of that slice.

What is the difference between 1% and 10%?

The percentage indicates how large a slice of the country's population enters the audience. 1% is the most similar and narrowest slice: high accuracy, limited reach. 5% and above produce a much larger audience where similarity thins out and conversion rate falls. The right approach is to start narrow and widen only as needed.

Narrow versus broad lookalike1% lookalike5% lookalike100420Reach100138Cost per acq.10062Conversion rateIllustrative index
A narrow lookalike reaches fewer people but converts better; a broad one buys scale at the cost of efficiency.

How do you choose the source audience?

This is the most important decision in the whole process. All site visitors is an easy source but contains many people with no purchase intent. The best results usually come from a high-value customer list: repeat buyers, high basket values and no refunds.

Source audience quality by type100 endeksHigh valuebuyers88 endeksAllpurchasers71 endeksAdd to cart54 endeksSitevisitors38 endeksVideoviewersIllustrative index
Source quality by type: the more selective the list, the more accurate the lookalike.
  1. Pick the source: where possible, use a list of customers with high lifetime value.
  2. Clean the list: remove refunders, one-off discount buyers and test orders.
  3. Choose the percentage: start at 1% and widen in steps if volume is insufficient.
  4. Test it: compare the new audience against your current one with the same creative and offer, one variable at a time.
The order for building a lookalike1Pick thesourceYour mostvaluable customerlist2Clean thelistRemove refundsand one-offs3Choose thepercentageStart narrow,widen if needed4Test itCompare onevariable at atime
The build order: source, cleanup, percentage and a controlled test.

Which audience actually returns profit?

Ads Sensor analyzes campaigns across four platforms with AI and shows audience and creative driven performance gaps with reasoning.

Join the beta →

When do lookalikes stop working?

Three situations disappoint. First, if the source list is too small no pattern can be extracted; a few hundred records is usually the floor. Second, if the source is heterogeneous, similarity loses meaning: mixing wholesale and retail customers in one list blurs the audience. Third, if your product sits in a very narrow niche, the number of similar people available is limited to begin with.

1%
the narrowest and most accurate similarity slice
Source list
the factor that actually determines quality
A few hundred
minimum records for a usable source list

Do automated targeting options make lookalikes obsolete?

No, but they change the role. Meta's broad automated targeting tries to find the audience itself; a lookalike tells the system where to start. That starting signal still matters, especially in new accounts and narrow niches. We covered when automated targeting is the right call in our Advantage+ guide; for the wider audience picture see audience targeting.

A lookalike is the advertising-side expression of your first-party data: as data quality improves, audience quality rises with it. For how to collect that data, see first-party data strategy, and to see campaign results in one dashboard, join the Ads Sensor beta.

Frequently asked questions

How many people does the source list need?
The floor is usually a few hundred records, but a few thousand is preferable for a stable pattern. The smaller the list, the noisier the similarity.
Should I use 1% or 5%?
Start narrow. 1% gives a higher conversion rate; widen to 3% or 5% when reach runs short. Testing both at once quickly shows which works for your product.
Which country should I build the lookalike in?
The audience is drawn from users in the country you select. If you advertise in several countries, building a separate audience per country usually works better than one multinational audience.
How often should I refresh the source list?
Every few months is enough if your customer profile and product range are stable. For seasonal businesses or after a change in price positioning, refresh at the start of each season.
Should I exclude remarketing audiences from a lookalike campaign?
Yes. Exclude existing customers and recent site visitors; otherwise you spend new-audience budget reaching people who already know you, and the results look misleadingly good.

Make audience decisions with data

Ads Sensor unifies Google, Meta, TikTok and Criteo and shows which audience actually produces profit.

Join the beta →