For years the standard Meta setup was a stack of lookalike audiences at different percentages. That has changed. Then broad targeting with strong creative started beating them in account after account, and the industry swung hard the other way.
The truth is situational. Here is when lookalikes still win.
When lookalike audiences still work
1. Small budgets with a narrow product
Broad targeting needs volume to learn. If you are spending ₹30,000 a month on a specialised product, a lookalike gives the algorithm a head start it cannot earn on its own at that spend.
2. When your source list is genuinely good
A lookalike is only as good as its seed. A list of 2,000 repeat customers with high order values produces a useful audience. A list of 40,000 people who downloaded a free PDF produces an audience of people who like free PDFs.
3. When you need to exclude, not just include
Often the real value is the source list itself as an exclusion — stop paying prospecting rates to reach existing customers. That works regardless of whether the lookalike is any good.
When broad beats them
- You have conversion volume. Above roughly 50 conversions a week, the system finds buyers better than your seed list describes them.
- You have creative variety. With ten genuinely different angles, the creative does the targeting — each ad finds its own audience.
- Your product has wide appeal. Narrowing an audience that did not need narrowing just raises your cost per thousand.
A lookalike of a weak list is a confident way to find more of the wrong people.
Source list quality, in order
- Repeat purchasers. Best seed available, even if small.
- High-value customers. Value-based lists, where supported, beat flat ones.
- All purchasers. Fine, and the usual default.
- Qualified leads from your CRM. Good for lead-gen, if you actually mark qualification.
- All form fills. Weak. You are describing form-fillers.
- Page engagers, video viewers. Weakest. Engagement is not intent.
A 500-person list of repeat buyers beats a 50,000-person list of newsletter signups. Size is not the quality signal people assume it is.
Practical rules
- Start at 1%. Widen only if delivery is constrained. Wider percentages converge towards broad anyway.
- Refresh the seed quarterly. Last year's customers describe last year's market.
- Do not stack five percentages in one campaign. You are splitting learning across audiences that overlap heavily.
- Always exclude existing customers from prospecting, whatever your targeting approach.
How to actually decide
Run both. One campaign broad, one on a 1% lookalike from your best list, same creative, same budget, three weeks. Compare on cost per qualified lead.
The answer differs by account, which is why the confident universal advice on this topic is usually wrong. Test it on yours rather than adopting whichever position is currently fashionable.
If the account behind this needs someone in it every day rather than every Friday, that is what we do on Meta.
Questions people ask
Do lookalike audiences still work?
In some accounts, yes — particularly where the source list is small, high quality and genuinely distinctive. In many others, broad targeting with strong creative now beats them, because the delivery system finds the same people without the constraint.
What makes a good source list?
Recency and purity before size. Five hundred recent high-value customers outperform fifty thousand mixed site visitors, because the model copies whatever you give it, including the noise.
What percentage should I use?
One percent when the source is strong and the market is large; wider when delivery is constrained. The percentage matters far less than the list behind it, which is where most of the difference comes from.
Should I test broad against lookalikes?
Yes, and it is a cheap test. Run them as separate ad sets with the same creative and judge on cost per result after both have left the learning phase. Account-level answers beat general advice here.
Want help setting that test up? Send us your account details.