There’s a ton of bad advice out there about Facebook ads, especially when it comes to scaling with lookalike audiences. Too many marketers are still using old playbooks, burning through their budget and wondering why they aren’t seeing real growth. If you want your digital ads to actually work, you have to understand how to properly expand your audiences with lookalikes.
Key Takeaways
- You’ll get much better results building lookalikes from your absolute best customers (like the top 5% of spenders or repeat buyers) instead of just using your entire customer list.
- Always test different lookalike percentages, like 1%, 3%, and 5%, against each other so you have real data on what percentage gives you the best balance of audience size and conversion quality for your goal.
- Your lookalikes go stale. Refresh the source audience data you use to build them at least once a month to keep performance from dropping off as customer behavior changes.
- Don’t layer interest or demographic targeting on top of your lookalikes. It’s almost always a bad idea that just restricts the algorithm and drives up your costs. Let the lookalike do its job.
- Lookalikes built from video viewers or post engagers are great for scaling top-of-funnel campaigns, especially if you have a solid retargeting strategy to follow up.
Myth 1: Larger Lookalike Percentages Always Mean More Reach
People think a 10% lookalike will always beat a 1% lookalike because it’s bigger, but that’s a classic misconception. Sure, a 10% audience is numerically larger, representing a bigger slice of Facebook users similar to your source. But that doesn’t mean it’s more *effective* reach. The entire point of a lookalike is to find people who are freakishly similar to your best customers. As you climb from 1% to 5% or 10%, you’re watering down that similarity. A 1% lookalike gives the algorithm a super-focused target, finding the most precise matches possible. In our experience, these smaller lookalike percentages, usually in the 1% to 3% range, almost always deliver a better cost per acquisition (CPA) or return on ad spend (ROAS). The algorithm just has a much clearer signal to work with. For example, if you sell high-end artisanal coffee, a 1% lookalike finds you the people who own three different kinds of grinders. A 10% lookalike includes people who buy a bag of Folgers once every two months. I’ve seen campaigns where a simple 1% lookalike brought in a 20% lower CPA than a 5% lookalike, even though the potential reach was way smaller. Your goal isn’t just to reach a ton of people. It’s to reach a ton of the *right* people who will actually buy something.
Myth 2: You Only Need One Lookalike Audience
I see this all the time: an advertiser makes one lookalike from their entire customer list or all website visitors and just uses it for everything. This approach leaves so much of the power of Facebook ads on the table. The real magic happens when you segment your lookalikes based on specific goals and where someone is in the funnel. A customer who has spent $500 with you is worlds apart from someone who just clicked on a blog post. If you treat them the same when building audiences, you’re ignoring the most valuable signals you have. You should be building a whole portfolio of lookalikes based on different source audiences. For example, you should have:
- A 1% lookalike of your top 10% highest-spending customers. This is your whale-hunting audience, built to find more people likely to make big purchases.
- A 2% lookalike of users who initiated checkout. These are people who showed intense buying intent, and you want more just like them.
- A 3% lookalike of people who watched 75% or more of your main product video. This is perfect for building awareness with a highly engaged group.
- A 1% lookalike of your email subscribers. These people already opted into your brand, making them a strong signal for finding others who are open to it.
Every one of these sources gives Meta’s algorithm a different set of signals to work with. By feeding it these distinct profiles, you’re telling it to find new users who are similar to very specific *types* of valuable people. It’s no surprise that industry analysis, like a 2024 report from eMarketer (https://www.emarketer.com/content/meta-ad-spending-forecast-2024), keeps pointing toward the need for more granular audience strategies as the platforms get smarter. This kind of diversification is what separates campaigns that work from campaigns that work spectacularly well.
Myth 3: Layering Interests and Demographics on Lookalikes Always Improves Targeting
The logic seems reasonable enough: “I have a lookalike of my customers, so if I ALSO target only women aged 25-45 who are interested in ‘sustainable fashion’, I’ll get a super-audience, right?” Wrong. In almost every test we’ve run, this is a bad move that hurts your campaign. When you start throwing extra targeting layers on top of a lookalike, you’re basically putting handcuffs on Meta’s machine learning. The algorithms, especially where they are today and where they’re heading by 2026, are incredibly sophisticated. When you give it a good seed audience (like your top 5% of customers), it’s identifying thousands of data points that those people have in common, many of which are far more predictive than any broad interest you could manually pick. Adding your own filters forces the algorithm to ignore all those subtle signals. This shrinks your potential audience, which usually just drives up your CPM because you’re now in a more competitive bidding war for a tiny, artificially constrained group of people. It also prevents the algorithm from finding surprising pockets of customers you would’ve never thought to target yourself. We’ve run A/B tests where simply removing the interest layers from a lookalike ad set dropped the CPA by 15-30% and massively increased conversions. You have to trust the algorithm. Reports from groups like the IAB (https://www.iab.com/insights/iab-digital-ad-revenue-report-full-year-2023/) consistently back up the idea that giving a smart algorithm broad-but-qualified data to work with beats overly narrow manual targeting.
Myth 4: Lookalikes Are Only for Prospecting New Customers
Thinking that lookalikes are only for top-of-funnel audience expansion is a shortsighted view that misses some really clever applications. They can be incredibly useful at every stage of the funnel, even for reactivation. Let’s say you have a list of customers who bought from you two years ago but haven’t been back. You can build a 1% lookalike of those “lapsed” customers. This tells Meta to go find *new people* who share the characteristics and behaviors of customers who *used to be* loyal to you. It’s a surprisingly effective way to find new prospects who are already predisposed to what you sell. You can also use lookalikes to work your existing customer base. For instance, if you run a loyalty program, create a lookalike of your most engaged, highest-point-earning members. Then, target that audience while excluding the loyalty members themselves. This lets you serve ads encouraging sign-ups to the segment of your *current customers* who are most likely to become power users. This isn’t about acquisition. It’s about deepening engagement. Lookalikes are a flexible tool for your entire marketing plan, not just a prospecting gadget.
Myth 5: You Can Set and Forget Your Lookalike Audiences
Your lookalike audiences have a shelf life. The digital ad space changes constantly, and so does customer behavior. Believing that a lookalike you built six months ago is still going to be a top performer today is a huge mistake. The source audience that you built the lookalike from is a living thing, new customers are coming in, old ones are changing their habits, and market trends shift. If you don’t refresh your data, your lookalike will get stale and your performance will slowly degrade. We make it a rule to refresh the source audiences for our lookalikes at least once a month, and even more often if it’s a high-spend account. That means uploading a fresh customer list every few weeks or ensuring your Meta Pixel is constantly feeding the most current data into your website-based audiences. This constant refresh process guarantees that your lookalike is always being built from the most relevant, up-to-date behavioral signals. As reports like Nielsen’s annual marketing report (https://www.nielsen.com/insights/2023/nielsen-annual-marketing-report-2023/) show, fresh data is non-negotiable for accurate targeting. That same logic is critical for maintaining healthy lookalike audiences. Not doing this maintenance is like telling the algorithm to find people who look like your customers from last Christmas. It’s a losing strategy. To really make Facebook lookalike audiences work, you need to ditch the common myths and get more strategic. By prioritizing quality source audiences, segmenting them properly, trusting the algorithm, using lookalikes for more than just prospecting, and keeping your data fresh, you’ll see a real lift in your Facebook ads performance. This approach to audience expansion is what builds efficient, long-term growth.
What’s the best source audience size for a lookalike?
Meta says you need at least 1,000 people, up to 50,000. But the quality of that list is way more important than the size. A clean list of 1,000 of your best customers will always outperform a messy, unqualified list of 50,000 people.
Should I exclude my existing customers from my lookalike campaigns?
Yes, and you absolutely should. When you’re running a campaign to get new customers, excluding your current customer list is basic campaign hygiene. There’s no point in paying to acquire people you already have.
How many different lookalike audiences should I have?
It really depends on your business, but having 5 to 10 distinct lookalike audiences is a great place to start. You should aim to have a separate lookalike for each of the high-value actions you track, like top spenders, people who watch 75% of a video, newsletter signups, or add-to-carts.
What’s the difference between a website visitor lookalike and a customer list lookalike?
A website visitor lookalike is dynamic and built from your Meta Pixel, which tracks user behavior on your site in real time. A customer list lookalike is static and built from a data file you upload (like emails or phone numbers) that Meta then matches to user profiles. Both are useful but are based on totally different data inputs.
Are value-based lookalikes worth it?
100%. If you can give Meta customer data that includes a lifetime value or purchase value column, you need to be using value-based lookalike audiences. This tells the algorithm to find new people who don’t just act like any of your customers, but specifically like your *most profitable* customers. It’s one of the best ways to improve your ROAS.
