Lookalike Audiences vs Interest Targeting: When Each Still Works in 2026

Most agency playbooks treat lookalike audiences as the default for any DTC account on Meta at scale. Stack 4 to 7 of them. Layer them across the account. Report on the cleaner-looking conversion attribution. The honest answer is that lookalikes are over-used at $30K+ monthly spend, and the brands paying for stacked-lookalike maze structures are usually getting outperformed by simpler accounts with 2 to 3 clean lookalikes plus broad targeting paired with strong creative. The patterns repeat across $150M+ in managed ad spend.

This post is the framework. When lookalike audiences (LAL, audiences Meta builds from a seed of your existing customers) still earn their slot. When interest targeting beats them. And the seed-quality audit to run on your account this week.

What lookalike audiences and interest targeting each actually do

Both tools live in the audience tab on Meta. Both rely on Meta's behavioral data underneath. They do different jobs.

Lookalike audiences (LAL). You give Meta a seed audience. Past customers, site visitors, video viewers, email subscribers. Meta builds an audience of users who share behavioral signals with that seed. Audience sizes run from 1% (closest match, around 2 million users in the US) to 10% (broadest, around 20 million users). The operator defines the quality of the input.

Interest targeting. You pick from Meta's catalog of interest categories. "Interested in fitness." "Interested in supplements." "Interested in small business software." Meta serves the ad to users who match those interests based on its behavioral signals. The operator defines the category, not the quality.

The structural difference: lookalikes require you to define quality. Interest targeting requires you to define category. Both then hand the work back to Meta's algorithm.

What changed in 2024 and 2025: Meta's broad targeting got materially better at finding buyers when paired with strong creative. Meta's own data shows the shift. In a study of 15 A/B tests, Advantage+ Shopping campaigns drove 12% lower cost per purchase conversion than advertisers' standard manually-targeted ads, and the lift is largest in accounts where the manual targeting was lookalike-heavy. That is the contrarian point of this post. The accounts most stacked with lookalikes are the ones Advantage+ beats by the widest margin.

The takeaway: lookalikes and interest targeting still work. The lever that does most of the audience-building work in 2026 is the creative. The targeting decisions still matter at the edges. They no longer carry the whole account. This is the same diagnostic logic Step 8 of the Meta ads audit method applies to signal hygiene: clean signals beat clever stacking.

The seed audience problem (why most lookalikes are broken)

Most operators build lookalikes from convenient seeds. Whatever is one click away in the audience tab. They never audit what is inside the seed. The result: the lookalike Meta builds reflects the noise in the seed, not the signal.

Three common seed mistakes show up across $150M+ in managed spend.

1. The "all customers" seed. The operator uploads the full customer list. The list includes one-time low-AOV (average order value) buyers, returners, refunded orders, and brand-fanatic repeat buyers. The lookalike Meta builds reflects this mixed quality. The algorithm finds users who look like the average of all those buyers. The average of a high-LTV (lifetime value) buyer and a refunded order is statistical noise.

2. The "30-day site visitors" seed. The operator uses website visitors as the seed. Most site visitors did not buy. The lookalike Meta builds finds users who behave like visitors, not like buyers. The downstream conversion rate reflects this. The CPM (cost per thousand impressions) is cheap. The cost per purchase is high.

3. The under-1,000-conversion seed. Meta's own documentation on lookalike audiences recommends a source audience of 1,000 to 5,000 people. Below that, the audience does not have enough behavioral signal to produce a useful match. Most DTC accounts under $50K monthly spend do not have 1,000 high-LTV purchasers in the last 180 days. Their lookalikes are extrapolating from noise.

The fix is not to abandon lookalikes. The fix is to audit the seed. The right seed for a DTC supplements brand looks like this: customers who placed 2 or more orders in the last 90 days, filtered to AOV above the brand median. That seed produces a useful lookalike. The "all customers ever" seed produces a useful-looking but useless one.

Enter your seed size and lookalike percentage to see the audience reach and a readiness score against Meta's recommended minimum:

When interest targeting beats lookalikes in 2026

Lookalikes are not always the right tool. Five scenarios show up where interest targeting structurally outperforms.

1. New product launch with no conversion data. No seed audience exists yet. There is no list of past customers to feed Meta. Interest targeting is the only structured option until conversion data accumulates. Trying to build a lookalike off 50 early purchasers does more harm than starting broad with interest layered in.

2. Category expansion. A skincare brand expanding into supplements has rich skincare conversion data and zero supplement conversion data. Lookalikes built from skincare purchasers will not find supplement buyers efficiently. The behavioral overlap is too thin. Interest targeting on supplement-adjacent categories is the better starting point.

3. Geographic expansion. A US brand launching in Australia has zero Australian customer data. Lookalikes built from US customers do not transfer cleanly to the Australian market. Cultural, seasonal, and platform-usage patterns differ. Interest targeting on the new market is more efficient until local conversion data builds up.

4. B2B (business-to-business) and professional audience targeting. Lookalikes struggle with professional audience identification because the seed (your customer list of business buyers) is usually small. A B2B SaaS brand with 800 customers cannot build a useful lookalike. Interest targeting on job titles, industries, and business interests often outperforms.

5. Niche product categories. A brand selling specialized gear (mountaineering equipment, surgical tools, audiophile headphones) often has a small total addressable audience. Lookalikes built from a small seed extrapolate weakly. Interest targeting on the niche category, paired with creative that names the buyer specifically, is more efficient.

The pattern: the limit is not lookalikes as a tool. The limit is lookalikes as the default. Use them when the seed is high-quality and the conversion volume is sufficient. Use interest targeting when those conditions do not hold or when discovery is the goal.

The audit checklist for your current targeting setup

Five questions to run on your account this week. Each one takes 10 to 15 minutes. The whole audit is under an hour.

1. For each lookalike audience running, what is the seed? Open the audience itself. Look at the source. Is it "all customers" or 30-day site visitors? Or is it filtered (recent buyers, AOV above median, repeat purchasers)? Generic seeds produce generic audiences. Note which lookalikes have generic seeds. Those are the cut candidates.

2. How many high-quality events does each seed contain? Meta recommends 1,000 to 5,000 people in the source. Below 1,000 high-quality events, the lookalike is noise. Above 1,000, it can work. Most accounts under $50K monthly spend will find that 3 or 4 of their lookalikes fail this check.

3. How many overlapping lookalike audiences are running in parallel? Stack 5 lookalikes on a $50K account and the audiences overlap by 30 to 60% on Meta's overlap tool. Meta's auction algorithm then bids against your own audiences. CPM rises. The right count for most accounts at this spend tier is 2 to 3 clean lookalikes, not 5 to 7 stacked ones.

4. Is broad or Advantage+ Shopping running alongside the lookalikes? If yes, check the lift. Many accounts find that the lookalikes are claiming credit for conversions broad would have driven on its own. Meta's own Advantage+ data points to this dynamic. The lookalike campaigns look like they are working. They may just be tracking what broad would have done.

5. What does the creative look like on each campaign? Per creative is the new targeting in 2026, the creative is doing most of the audience-building work. If the lookalike campaigns use generic creative, the targeting is not compensating. If the interest campaigns use sharp hooks, they are qualifying audiences more effectively than the lookalikes.

The closing position: most operators at $30K+ monthly spend should cut their lookalike count to 2 or 3 (clean seeds, 1,000+ events), introduce 1 or 2 interest targeting campaigns for discovery, and let creative do the rest. The simpler structure outperforms the stacked-lookalike maze. The right number of lookalikes also depends on which account stage you are in. The full framework is in the 4 stages of a Meta account. For the full diagnostic order this fits inside, see the audit method's full 10-stage diagnostic order. And before judging any targeting type by reported ROAS, read the number against your margin first. The gross-margin math behind a good ROAS is the part most operators skip.

The hot take: 4-7 stacked lookalikes is the wrong default

Most DTC accounts at $30K+ monthly spend have 4 to 7 lookalike audiences running in parallel. The reason is not that lookalikes are uniquely effective at that spend tier. The reason is that they are the easiest audience to build, the easiest to report on, and the easiest for an agency to justify on a monthly retainer call.

"Here is your 2% LAL of all-purchasers. Your 5% LAL of high-LTV. Your 3% LAL of 30-day site visitors. Your 1% LAL of newsletter subscribers." It sounds sophisticated. It looks like work. It bills.

The reality at scale: those audiences overlap heavily with each other. They overlap with Meta's broad-targeting algorithm. The seeds they are built from often have fewer than 1,000 high-quality conversions, which Meta's own documentation states is below the recommended threshold for a useful match. Brands running 7 stacked lookalikes built on small or contaminated seeds are paying Meta to find audiences the algorithm could find faster on its own.

This is the same structural failure mode Leak 5 on signal layering in stacked ad sets describes from a slightly different angle. Too many signals stacked on one ad set confuses the algorithm. Too many lookalikes stacked across the account does the same thing one layer up.

The cleanup is simple and unpopular. Cut the lookalike count to 2 or 3. Audit the seeds. Run broad or Advantage+ Shopping alongside. Put the creative effort into hooks that qualify the audience in the first 3 seconds. The account gets simpler. The CPMs come down. The cost per purchase comes down. The "audience strategy" deliverable on the agency invoice shrinks. That is why most agencies will not name this position. The simpler structure does not bill as well as the maze.