Is It Time to Abandon Lookalike Audiences?

Alright, welcome back to the Pubcast. I'm Jon Loomer, and today let's talk about lookalike audiences.

Now, this is a feature—it’s 2014, I know, because I wrote about it back in March of 2014. And back then, it made a lot of sense.

You may recall that the targeting inputs we provided were super important. We defined who would see our ads. We provided interests and maybe custom audiences. In 2014, there weren’t a ton of options. It was really hard to know who the right people were to include—especially with interests.

But with lookalike audiences, we could tell Meta:

"These are our ideal people—our website visitors, our email list, our paying customers, our Facebook followers. Now go find people like them."

This allowed us to scale, expand our audience, and find people beyond those already engaged with us. We assumed they were similar to our existing audience.

There was a lot of voodoo involved—we didn’t really know how it worked. We just took it for granted that it was a good way to create a broader audience and reach new people.

But targeting works differently now.

First, it’s much smarter. Second, there’s far less control.

Whether you use broad targeting or not, your audience is automatically expanded in many cases. If you use Advantage+ Audience, whatever you provide is just a suggestion.

If you use original audiences and optimize for a purchase or conversion, your lookalike audience is expanded. Your detailed targeting is expanded.

It’s all just us giving the algorithm ideas—potentially. But we really don’t know how much it matters.

I’ve run tests suggesting that Meta pretty much ignores anything we put in there now if it can be expanded.

Now, targeting is built around the algorithm searching for people likely to perform our desired action. If you want purchases, it’s going to search out people who are likely to purchase from you.

And especially with purchases, the algorithm bakes in remarketing.

I’ve tested this with audience segments and sales campaigns. I’ve found that anywhere from 25 to 50% of my budget is automatically spent on remarketing when I go broad—and that’s a big deal.

So, in a way, the algorithm is already functioning as one big lookalike audience while also remarketing at the same time.

While lookalike audiences made sense in theory—and may have been required 10 years ago because targeting was so confusing—they're not as necessary now.

Back then, we had too many options for interests and behaviors, and we didn’t know which ones were valid or effective. There was so much testing involved.

Now, the algorithm does all of that for us.

And even if you argue that algorithmic targeting has flaws (which it does), you can’t really avoid it anymore.

When you use a lookalike audience, it’s usually just a suggestion. In many cases, it’s going to expand beyond that anyway.

Ten years ago, the algorithm needed lookalike audiences. It needed us to define a source audience so it could find similar people.

But now? The concept is outdated.

Meta already knows who your most valuable audience is—and it prioritizes those people. That’s proven with audience segments.

Now, I won’t ever say something never works or that you should never use lookalike audiences. But I’ve got to admit—I’m pretty close to saying that here.

Providing lookalike audiences, in my opinion, is unnecessary busyness.

You think it’s necessary. You think what you’re doing is productive and impactful. But I don’t believe it’s doing much of anything.

Now, you can prove this. Don’t just take my word for it—run tests.

And I don’t mean just one test at $10 a day. This is something you need to continually test and evaluate.

Run tests using lookalike audiences like you normally do.

And first, understand:

• If you use lookalike audiences with Advantage+ Audience, they’re pretty much ignored anyway.
• If you use lookalike audiences with original audiences and they’re expanded, remember—you’re still relying on the same algorithm to build that lookalike audience. That same algorithm is running algorithmic targeting for you, too.

So, test it.

Try original audiences with lookalike audiences. Then try Advantage+ Audience with no suggestions at all.

It sounds scary, but do it—especially when optimizing for purchases.

The bottom line? Find what works for you.

If you’ve proven to yourself—with large sample sizes and real budgets—that lookalike audiences still work, I’m not going to tell you you’re crazy. Keep doing what works for you.

But in most cases, you’re just creating more work for yourself than necessary—especially if you’re creating multiple ad sets with different targeting (one for lookalikes, one for something else, and so on). That’s inefficient.

Keep questioning your assumptions.

What I believed 10 years ago is no longer true for me.

It wasn’t easy to make that change, but the system changed. The features changed. The algorithm changed.

And as a result, your strategy should change too.

Test it.