Why the biggest lookalike audience in Meta ads is usually the wrong one

32 min read | Updated March 2026

Lookalike audiences have been a foundational targeting tool in Meta Ads since 2013, but their mechanics, strategic value, and best practices have shifted dramatically in the post-ATT era. In our experience working across hundreds of app campaigns, we can say definitively: lookalikes in 2026 are not the same product they were even two years ago. They remain a powerful lever for scaling app installs and subscriptions, but only if you understand how Meta’s algorithms now calculate similarity, how signal loss has changed seed quality requirements, and when to use lookalikes versus broad targeting. This guide covers everything a mobile marketer needs to know about lookalike audiences in Meta Ads mobile app targeting as they function today, with specific benchmarks, structural recommendations, and real campaign data.

What exactly is a lookalike audience in Meta Ads and how does it work in 2026?

A lookalike audience is a targeting audience in Meta Ads built by algorithmically identifying users who share behavioral and demographic patterns with a source (seed) audience you provide. In 2026, Meta uses a combination of on-platform engagement signals, Conversions API (CAPI) data, and probabilistic modeling to build these audiences, scoring every user in a given country on a similarity index from 0% to 10%. A 1% lookalike in the United States represents approximately 2.6 million users who most closely resemble your seed.

The underlying mechanics work like this: you upload or define a source audience (for example, all users who completed a subscription purchase in the last 90 days), and Meta's machine learning models analyze hundreds of signals across those users.

These signals include content interaction patterns, ad engagement history, purchase behavior on and off Facebook, device and connectivity data, and app usage patterns. The system then scores every eligible user in your target geography and returns a ranked audience at the percentage size you specify.

A 1% lookalike is the top 1% most similar users; a 5% lookalike expands to the top 5%, adding more reach but diluting similarity.

According to Meta's Business Help Center documentation on lookalike audiences, the algorithm requires a minimum seed of 100 users from a single source country, though Meta recommends 1,000 to 50,000 for optimal performance.

In practice, industry experience consistently points to substantially better results with seeds of at least 2,000 to 5,000 high-quality events, well above the documented minimum threshold. What changed materially in 2025-2026 is the weight Meta places on on-platform signals versus off-platform data. With Apple’s App Tracking Transparency framework now firmly entrenched and ATT enforcement and tracking opt-in rates, Meta’s lookalike algorithms have adapted to rely more heavily on modeled conversions.

With Apple’s App Tracking Transparency framework now firmly entrenched, global opt-in rates sitting around 25-30% on iOS as of early 2026 per AppsFlyer’s Performance Index (where dominant self-attributing networks in 2025 in 2025/2026), Meta relies more heavily on modeled conversions and aggregated event measurement.

This means your seed audience quality and your CAPI implementation directly determine how effectively Meta can build a useful lookalike.

• Seed audience: The source list of users (custom audience) Meta uses as a template. Can be based on app events, website visitors, customer lists, or engagement.
• Similarity score: Meta scores every user in a country from 0-10% similarity. A 1% lookalike in the US is approximately 2.6M users.
• Signal inputs: On-platform behavior (likes, video views, ad clicks), CAPI server-side events, aggregated off-platform purchase data, device/demographic signals.
• Minimum seed size: 100 users required according to Meta's Business Help Center documentation, but 2,000-5,000 high-value users recommended for best algorithm performance based on industry experience.
• Post-ATT change: Meta now leans more on modeled conversions and on-platform signals, making CAPI integration and event setup more critical than ever.

How is a lookalike audience different from a custom audience?

A custom audience is a defined set of known users: people who installed your app, visited your website, or appear on your CRM list. A lookalike audience is a net-new prospecting audience that Meta creates by finding unknown users who resemble your custom audience. Think of the custom audience as the input and the lookalike as the output. In our experience running campaigns for subscription apps at RocketShip HQ, the distinction matters operationally because custom audiences are used for retargeting (re-engaging known users) while lookalikes are used for acquisition (finding new users). A common mistake is conflating the two or not excluding existing users from lookalike campaigns, which wastes spend on people who already know your app.

What signals does Meta actually use to calculate lookalike similarity?

Meta has never published the full list of signals, but based on patent filings and observed behavior across thousands of campaigns, the primary inputs include: ad interaction history (clicks, video views, conversions), page and group engagement, content consumption patterns (what types of posts users linger on), purchase behavior tracked via Meta's commerce tools, demographic attributes (age, location, device), and server-side conversion events sent through Meta's Conversions API. Since ATT, modeled conversions (Meta's probabilistic estimates of who converted even without tracking consent) feed into lookalike construction. This is why campaigns optimized toward events with higher conversion volume tend to produce better lookalikes: more data gives Meta more confidence in its models.

Why do lookalike audiences still matter for mobile app growth in 2026?

Lookalike audiences matter because they give Meta's algorithm a directional signal about the type of user you want, which is especially valuable during the learning phase of new campaigns. In our experience, campaigns using value-based lookalike seeds (top 10% spenders) as initial targeting have consistently delivered meaningfully lower cost-per-trial-start compared to completely cold broad campaigns in the first 7 days.

The argument for lookalikes in 2026 is nuanced. On one hand, broad targeting vs interest stacks for campaigns spending above $500/day because Meta's Advantage+ algorithms have gotten better at finding high-value users without explicit audience inputs.

On the other hand, lookalikes serve a specific strategic purpose: they accelerate learning phase exits and provide a stronger initial signal for campaigns optimizing toward deeper funnel events like subscriptions or in-app purchases.

When you launch a brand new campaign optimized for purchase events, Meta needs roughly 50 conversion events per week (per ad set) to exit learning phase, according to Meta's Business Help Center documentation.

A well-constructed lookalike audience effectively narrows the search space, helping Meta find those initial 50 conversions faster by starting with a more relevant pool.

We have observed this pattern consistently: lookalike-seeded campaigns tend to exit the learning phase faster than broad campaigns, and that head start often translates into structurally lower CPAs that persist even after both campaign types have stabilized. Additionally, lookalikes remain useful for geographic expansion.

When a fitness app wants to expand from the US to the UK, a value-based lookalike built from US subscribers gives Meta a strong starting point in a market where it has zero conversion history for that app.

In our experience with geographic expansion campaigns, this approach has meaningfully reduced early-phase CPA versus broad targeting during launch phases in new markets.

• Learning phase acceleration: Lookalikes help Meta find initial conversions faster, which in our experience can meaningfully reduce time spent in learning phase compared to cold broad campaigns.
• Deeper funnel optimization: For purchase or subscription events with low daily volume, narrowing the audience with a high-quality seed improves signal density.
• Geographic expansion: Lookalikes from mature markets give Meta useful signal when entering new countries with zero historical data.
• Cost efficiency in early phases: In our experience, value-based lookalike seeds can deliver meaningfully lower CPAs in the first 7 days compared to broad targeting for subscription-optimized campaigns.

What types of seed audiences produce the best lookalikes for app campaigns?

In our experience, value-based seed audiences (users ranked by LTV or purchase amount) consistently outperform event-based seeds (all purchasers treated equally) on ROAS metrics. The single best seed we have found for subscription apps is a renewed subscription users by revenue, filtered to the top segment by total revenue, because second-purchase events typically produce seeds that outperform first-purchase.

Seed quality is the single most important variable in lookalike performance. Meta lets you create lookalikes from several source types: app event audiences (e.g., everyone who triggered a 'purchase' event), customer list uploads (email/phone matched to Meta profiles), website custom audiences, and engagement audiences (video viewers, page engagers).

For mobile apps, the hierarchy of seed quality based on our testing is clear. Customer lists of high-LTV users perform best because the data is deterministic and represents actual business value. App event audiences rank second, with purchase events outperforming install events.

Engagement audiences (like video viewers or ad engagers) rank last because the correlation between watching a video on Facebook and actually paying for a subscription is weak.

A critical nuance: you want your seed to be high-quality but not too small. Meta's documentation says 100 users minimum, but seeds below 1,000 users produce unreliable lookalikes because there isn't enough data for the algorithm to identify meaningful patterns.

Conversely, seeds above 50,000 users become too diverse, and the resulting lookalike essentially converges toward broad targeting anyway. In our experience, a sweet spot of roughly 2,000-10,000 users tends to produce the strongest lookalike performance.

For value-based lookalikes (where Meta weights similarity by user value, not just behavior), you need to include a value column in your customer list upload. At RocketShip HQ, we have tested this approach for subscription fitness apps and found that switching from a standard ‘all purchasers’ seed to a value-based seed of the top LTV segment meaningfully reduced cost-per-trial-start and improved Day 30 retention.

We have observed that value-based lookalikes consistently outperform standard lookalikes on Day 30 ROAS for subscription apps, with directional improvements seen across gaming and fitness verticals as well.

Running Meta Ads for subscription apps specifically benefits from using renewal or retention-based seeds rather than trial-start seeds, because Meta then optimizes for users who actually stick around, not just those who click 'Start Free Trial.'

What is a value-based lookalike audience and when should you use it?

A value-based lookalike is created from a customer list that includes a monetary value column (lifetime revenue, predicted LTV, or subscription value). Instead of treating all seed users equally, Meta weights the similarity model toward users who look like your highest-value customers. You should use value-based lookalikes whenever you have at least 2,000 users with reliable revenue data and your goal is ROAS or LTV optimization rather than pure install volume. In our experience, value-based lookalikes consistently reduce cost-per-subscriber compared to standard event-based lookalikes for subscription apps across categories including meditation and language learning. The trade-off is slightly lower reach and higher CPMs, but the quality improvement more than compensates.

How often should you refresh your lookalike seed audiences?

Refresh seeds at least every 30-60 days for active campaigns. Stale seeds degrade performance because the users in your seed may no longer represent your ideal customer profile, especially if your product, pricing, or market has evolved. We recommend automating seed refreshes through your MMP (like AppsFlyer or Adjust) by syncing audiences to Meta on a rolling basis. We learned this lesson firsthand with a meditation app client at RocketShip HQ whose lookalike campaigns had been running on the same seed for four months. After refreshing with a rolling 60-day window of subscribers, CPA dropped meaningfully within the first week. In our experience, gaming clients have also seen notable improvements in CPI simply by switching from a static, months-old seed to a rolling purchaser audience.

What lookalike audience percentage should you use for app install campaigns?

For most app install campaigns in the US, start with a 1-3% lookalike. In our experience, 1% lookalikes tend to deliver the lowest CPAs but limited scale, while 3% lookalikes offer a strong balance of efficiency and volume. Beyond 5%, performance typically degrades to near-broad-targeting levels.

The percentage you choose represents a direct trade-off between precision and scale. Here is what that trade-off looks like across subscription and freemium app campaigns, based on commonly observed industry patterns, normalized to the 1% lookalike as the baseline:

Notice that 10% lookalikes are actually slightly worse than broad targeting on CPA metrics.

This is counterintuitive but consistent: at 10%, the lookalike is so diluted that it adds noise rather than signal, while broad targeting lets Meta's algorithm use its full optimization capabilities without artificial constraints.

The practical recommendation for most app marketers spending $500–$5,000 per day is to run a 1-2% lookalike alongside a 3-5% lookalike and a broad campaign, then allocate budget toward whichever delivers the best marginal CPA.

For smaller budgets under $500/day, concentrate spend on a single campaign type rather than splitting across multiple lookalike percentages, because each ad set needs sufficient conversion volume to optimize properly.

How do lookalike audiences compare to broad targeting and Advantage+ campaigns in 2026?

In mature campaigns with strong conversion data, Advantage+ app campaigns can match or beat manual lookalike campaigns on CPA, and often scale well. However, in our experience, lookalikes tend to outperform broad during the learning phase and for campaigns with fewer than 100 weekly conversions per ad set.

This is the most debated question in mobile UA right now, and the answer depends heavily on your campaign maturity and conversion volume. Meta's Advantage+ shopping campaigns (and the equivalent Advantage+ app campaigns) essentially automate audience selection, and Meta has been aggressively pushing advertisers toward these formats.

Understanding how Meta's ad auction works helps explain why: the auction algorithm already optimizes for the users most likely to convert within any audience you define, so the value of explicit audience targeting is really about narrowing the starting search space.

Here is a practical framework based on our experience at RocketShip HQ:

The key insight from post-ATT testing across 15+ accounts is that the value of explicit targeting decreases as your conversion volume increases.

If you are getting 300+ conversions per day, Meta's algorithm has learned enough about your ideal user that a lookalike constraint may actually hurt performance by excluding potential converters who do not fit the lookalike profile but would have converted anyway.

Conversely, if you are optimizing for a rare event (like subscription renewal) with only 20-30 weekly occurrences, a value-based lookalike gives the algorithm a much-needed boost.

• Below 50 weekly conversions per ad set: Lookalikes tend to significantly outperform broad in our experience, as the algorithm benefits from the tighter audience signal to guide early optimization.
• 50-200 weekly conversions: Lookalikes and broad perform similarly; test both and allocate dynamically.
• 200+ weekly conversions: Broad or Advantage+ often matches or beats lookalikes; the algorithm has learned your user profile.
• Advantage+ campaigns can include lookalike audiences as 'audience suggestions' rather than hard constraints, giving Meta flexibility.

Should you use Advantage+ audience expansion with lookalikes?

Yes, in most cases. Advantage+ audience expansion (formerly 'lookalike expansion') allows Meta to serve ads beyond your defined lookalike when it predicts a conversion is likely. In our experience, enabling expansion on a 3% lookalike can meaningfully increase install volume with only a modest increase in CPA, a worthwhile trade-off for apps prioritizing scale. The one exception is when you are running a tightly controlled test (e.g., comparing seed audiences) and need clean audience separation; in that case, disable expansion to keep your test valid.

How should you structure Meta campaigns that use lookalike audiences for app installs?

Need help scaling your mobile app growth? Talk to RocketShip HQ about how we apply these strategies for apps spending $50K+/month on UA.

The optimal structure for lookalike-based app campaigns in 2026 is a consolidated approach: one campaign with 2-4 ad sets, each targeting a different lookalike percentage or seed type, with 3-6 creatives per ad set. Avoid fragmenting into more than 4-5 ad sets, as this dilutes conversion volume and keeps ad sets stuck in learning phase — a pattern we consistently observe across app accounts we work with.

Campaign structure is where many app marketers go wrong with lookalikes. The temptation is to create separate ad sets for 1%, 2%, 3%, 5%, and 10% lookalikes, plus broad, plus interest targeting. This results in 6-7+ ad sets competing against each other in Meta's auction, each starved for conversions.

The right campaign structure for Meta app ads prioritizes consolidation. Here is the structure we recommend at RocketShip HQ for a subscription app spending $2,000–$5,000 per day on Meta:
Campaign 1 (Prospecting, AEO or VO): Ad Set 1 targets 1-2% Lookalike from value-based subscriber seed with 4-6 creatives.

Ad Set 2 targets 3-5% Lookalike from the same seed with 4-6 creatives (can overlap with Ad Set 1 creatives).

Ad Set 3 targets Broad (no audience defined) with 4-6 creatives. Campaign-level budget optimization (CBO) distributes spend toward the best-performing ad set automatically. Campaign 2 (Advantage+ App Campaign): No manual audience selection; Meta handles targeting. Use this as a scale campaign once you have enough conversion history.

The reason consolidation matters is math: if you have a $3,000 daily budget split across 6 ad sets, each gets $500. At a $15 CPA, that is only 33 conversions per ad set per day, below the 50 per week threshold for reliable optimization but marginal for rapid learning.

Consolidating to 3 ad sets gives each $1,000 and 67 daily conversions, a much healthier signal.

right number of creatives per ad set is equally important: too many creatives fragment impressions, while too few limit Meta's ability to find winning combinations.

• Use Campaign Budget Optimization (CBO) to let Meta allocate between lookalike and broad ad sets based on performance.
• Exclude existing users (installers, purchasers) from all prospecting ad sets to avoid wasted spend.
• Avoid overlapping lookalike percentages in the same campaign (e.g., don't run 1% and 2% separately; combine as 1-2%).
• Refresh creatives every 2-3 weeks to combat creative fatigue, which degrades lookalike performance indirectly by lowering relevance scores.

How does ATT and iOS signal loss affect lookalike audience quality on Meta in 2026?

ATT has materially degraded iOS lookalike quality by reducing the conversion data Meta can use for seed construction and similarity modeling. In our experience, iOS lookalike CPAs run meaningfully higher than Android lookalike CPAs for equivalent events, and the gap tends to be wider for deeper funnel events like subscriptions (where iOS opt-in rates and measurement gaps to deterministic attribution as analyzed by MobileDevMemo).

The impact of Apple's App Tracking Transparency framework on lookalike audiences is indirect but significant. ATT does not prevent you from creating lookalikes; rather, it reduces the quality of the inputs.

When only 25-30% of iOS users opt in to tracking according to AppsFlyer’s Performance Index, Meta receives complete conversion data from a fraction of your user base. The rest are either modeled (probabilistic) or missing entirely if you have not implemented server-side Conversions API.

This creates two problems for lookalike seeds. First, your iOS custom audiences are smaller than reality because Meta cannot match all converters.

If you had 5,000 iOS subscribers but only 1,500 were matched to Meta profiles with tracking consent, your seed represents only 30% of your actual subscriber base, and potentially a biased 30% (users who consent to tracking may have different demographic and behavioral profiles than those who do not).

Second, Meta's similarity model has fewer off-platform signals to work with, so iOS lookalikes increasingly resemble on-platform engagement profiles rather than purchase-behavior profiles.

Practical mitigations we employ at RocketShip HQ include: server-side Conversions API to recover lost events, using cross-platform seeds (combining Android and iOS converters into one seed audience, since Android provides richer signal), and prioritizing on-platform seed types for iOS campaigns (engagement audiences or lead form submitters).

Additionally, consolidating AEO campaigns to achieve 128+ daily installs helps Meta's modeled conversions become more accurate, which indirectly improves lookalike quality. Using Apple Search Ads alongside Meta also helps by providing deterministic iOS conversion data that can inform overall strategy.

• ATT opt-in rates: Approximately 25-30% globally as of early 2026 per AppsFlyer’s Performance Index, meaning 70-75% of iOS users are not fully tracked.
• CAPI recovery: Implementing server-side events can recover a meaningful share of otherwise lost conversion data, improving seed quality and optimization signal.
• Cross-platform seeds: Combining iOS and Android converters into one seed leverages richer Android signal data.
• iOS vs Android CPA gap: In our experience, iOS lookalike CPAs tend to run higher than Android equivalents for subscription events, reflecting the signal loss caused by ATT.

What are the biggest mistakes marketers make with lookalike audiences in Meta Ads?

In our experience, the three most common mistakes are: using install-based seeds instead of value-based seeds (which inflates CPA meaningfully), creating too many overlapping lookalike ad sets that fragment budget, and never refreshing seeds (leading to gradual performance decay over time).

Mistake number one is seed quality negligence. Many app marketers default to creating a lookalike from 'all app installers in the last 180 days.' This is a mediocre seed because it includes every user, from high-LTV subscribers to people who installed and never opened the app.

The resulting lookalike optimizes for the average, not the ideal. In our experience, switching from an all-installer seed to a top-revenue-percentile seed can substantially reduce cost-per-subscriber — a pattern we have observed across productivity and subscription apps. Mistake number two is over-segmentation.

We regularly audit accounts that run 8-10 lookalike ad sets at different percentages, each with a $50-100 daily budget.

As discussed in our analysis of why early-stage apps should not diversify ad spend, each ad set needs meaningful conversion volume for the algorithm to learn. Ten ad sets at $100 each means none of them reliably exit learning phase for events with CPAs above $10.

Mistake number three is seed staleness. A seed audience created 6 months ago reflects your user base 6 months ago, not today. If your product, pricing, or market positioning has evolved, your old seed is training Meta to find the wrong type of user.

Mistake number four (bonus) is not excluding existing converters. Without exclusions, Meta will happily show your prospecting ads to existing subscribers who match the lookalike profile.

This inflates reported conversion numbers while wasting budget on people who already pay you. Always exclude custom audiences of installers and purchasers from lookalike ad sets.

How can you tell if your lookalike audience is underperforming?

Compare your lookalike ad set's CPA and ROAS against a broad targeting ad set running in the same campaign with the same creatives and optimization event. If the lookalike CPA is within 5% of broad, the lookalike is adding minimal targeting value and you should either improve the seed or switch to broad. If the lookalike CPA is more than 30% higher than broad, the seed is likely poor quality or stale. At RocketShip HQ, we flag any lookalike that underperforms broad by more than 10% for immediate seed review. The key diagnostic metrics are: CPA (primary), CTR (if CTR is low, the audience is not resonating with your creative), and frequency (if frequency exceeds 3.0 in under 2 weeks, the audience is too small).

How do you create a lookalike audience in Meta Ads Manager step by step?

Creating a lookalike takes about 5 minutes in Ads Manager: go to Audiences, click Create Audience, select Lookalike Audience, choose your source (custom audience), select target country, set percentage (1-10%), and click Create. The audience typically populates within 1-6 hours and refreshes automatically every 3-7 days according to Meta’s Business Help Center documentation.

Here is the detailed walkthrough for 2026, accounting for the latest Ads Manager UI changes:
Step 1: Build your source custom audience first. Navigate to Audiences in Meta Ads Manager. Click Create Audience and select Custom Audience.

For app-based seeds, choose 'App Activity' and select your app, then define the event (e.g., Purchase) and time window (recommend 30-180 days depending on volume). For customer list seeds, choose 'Customer List' and upload a CSV with email, phone, and optionally a value column.

Meta typically matches 50-70% of uploaded records according to Meta’s Business Help Center documentation. Step 2: Create the lookalike. Return to Audiences, click Create Audience, and select Lookalike Audience. Choose your custom audience as the source.

Select the target location (one or more countries). Set the audience size percentage. You can create up to 6 lookalikes from one source simultaneously (e.g., 1%, 2%, 3%, 5% in one batch). Step 3: Apply to campaigns.

When building your ad set, under Audience, click the custom audiences field and search for your lookalike by name. Add demographic, geographic, or placement constraints as needed (though we generally recommend minimal layering on top of lookalikes, as additional constraints reduce Meta's optimization flexibility). Step 4: Pair with strong creatives.

The audience is only half the equation. Structuring creatives for different placements ensures your ads render well across Feed, Stories, Reels, and Audience Network, maximizing the effective reach within your lookalike.

One important note: Meta automatically refreshes lookalike audiences every few days to account for new users entering the platform and shifts in user behavior. However, the underlying seed does not auto-refresh unless you built it from a dynamic source (like an app event audience with a rolling time window). Static customer list uploads require manual re-uploading to refresh the seed.

Can you use lookalike audiences with Meta's Advantage+ app campaigns?

Yes, but with an important distinction. In Advantage+ app campaigns (and Advantage+ shopping campaigns), lookalike audiences function as 'audience suggestions' rather than hard targeting constraints. Meta will prioritize showing ads to users within your suggested lookalike but will expand beyond it when the algorithm identifies likely converters outside the audience. In our experience, this hybrid approach tends to deliver more volume than a hard-constrained lookalike at comparable CPAs, because the automated expansion captures incremental converters the strict lookalike would have missed.

Advantage+ campaigns represent Meta's push toward fully automated targeting, and they handle lookalikes differently than manual campaigns.

When you add a lookalike audience to an Advantage+ app campaign, you are essentially telling Meta's algorithm: 'Start here, but feel free to go wider.' Meta uses the lookalike as an initial signal and then expands as it gathers conversion data.

In practice, we have found this works well for apps with moderate conversion volume (50-150 daily conversions). The lookalike suggestion gives Advantage+ a head start that pure broad targeting does not, while the automated expansion captures incremental converters the lookalike would have missed.

We have observed that adding a value-based lookalike suggestion to an Advantage+ campaign can meaningfully reduce first-week CPA compared to running Advantage+ with no audience suggestion, particularly for apps still building conversion history.

However, for apps with 300+ daily conversions, the audience suggestion adds negligible value because Advantage+ already has enough data to optimize effectively. At that scale, the suggestion may even slightly constrain the algorithm. Choosing the right bidding strategy matters more than audience definition at high volume.

The strategic play for most growing apps in 2026 is to run both: a manual campaign with hard-constrained lookalike targeting (for consistent, predictable CPA) and an Advantage+ campaign with the same lookalike as a suggestion (for scale).

Allocate 60-70% of budget to whichever delivers the better marginal CPA, and use Meta's automated rules to shift budget dynamically based on daily performance.

What does a lookalike audience cost, and does it affect CPMs or CPIs?

Lookalike audiences themselves are free to create and use. However, tighter lookalikes (1-2%) typically carry higher CPMs than broad targeting because you are competing in a narrower auction with other advertisers targeting similar high-value users. Despite higher CPMs, tighter lookalikes usually deliver lower CPIs and CPAs because of better conversion rates.

Understanding the cost dynamics of lookalike audiences requires separating auction costs from business outcomes.

CPM (cost per thousand impressions) is a function of auction competition: when you target a 1% lookalike in the US, you are competing for a pool of approximately 2.6 million users who, by definition, look like high-value consumers. Other sophisticated advertisers are targeting similar profiles, driving up auction prices.

Here is what the cost breakdown typically looks like across targeting types for subscription apps in the US (figures are illustrative of industry patterns and should be validated against your own account data):

The pattern is clear: 1% lookalikes have the highest CPMs but the best conversion rates, resulting in the lowest downstream CPAs. These benchmarks align with broader Meta cost data on iOS vs Android.

This is why evaluating lookalike performance on CPM alone is misleading. The real question is always: what is your cost per meaningful business event (trial start, subscription, or purchase)? Setting the right testing budget ensures you generate enough data to evaluate these metrics reliably before making targeting decisions.

How do you combine lookalike audiences with custom product pages on the App Store?

In our experience, pairing specific lookalike audiences with tailored custom product pages (CPPs) on the App Store can meaningfully improve install conversion rates. The strategy is to match the messaging angle of your Meta ad creative to a CPP that reinforces the same value proposition, ensuring a cohesive user journey from ad impression to app install.

Custom product pages for Meta ads in 2026. Here is how the integration works: when a user taps your Meta ad and arrives at the App Store, they see your default product page.

But with CPPs, you can create up to 35 alternative product pages, each with unique screenshots, app preview videos, and promotional text. Meta allows you to link specific ads or ad sets to specific CPP URLs. The strategic opportunity with lookalikes is segmented messaging.

Say you are a fitness app with two core user segments: weight loss users and muscle building users.

You could create a value-based lookalike from your top weight loss subscribers and pair it with weight-loss-themed creatives pointing to a weight-loss CPP. Simultaneously, a separate lookalike from top muscle-building subscribers uses different creatives and a different CPP.

We have observed that matching CPPs to lookalike audience segments can substantially improve tap-to-install conversion versus sending all lookalike traffic to the generic product page, effectively reducing CPI without any change to bidding, budget, or creative.

The implementation requires coordination between your UA team and your ASO team. Each CPP needs to be created in App Store Connect, and the unique URL must be used as the destination in Meta's ad setup.

Track performance at the CPP level using App Store Connect analytics to understand which page-audience combinations deliver the best conversion rates.

• Create CPPs that match the messaging theme of your Meta ad creatives for each lookalike audience segment.
• Use unique CPP URLs as ad destinations in Meta Ads Manager.
• Monitor tap-to-install conversion rates in App Store Connect to validate CPP performance.
• Industry patterns suggest meaningful improvement in install conversion rate versus the generic product page when CPP messaging is well-matched to the audience segment.

What are the most important benchmarks and KPIs for measuring lookalike audience performance?

The primary KPI is cost per key event (CPA for your target conversion event, whether that is trial start, subscription, or in-app purchase), not CPI or CPM. Secondary KPIs include ROAS (Day 7 and Day 30), learning phase exit rate, and audience overlap percentage with other active ad sets. In our experience, a healthy lookalike campaign should exit learning phase within 7 days and deliver a CPA reasonably close to your target.

Measuring lookalike performance correctly requires looking beyond vanity metrics.

Here is the KPI framework we use at RocketShip HQ:
Primary metrics: Cost per key event (the event you are optimizing for in Meta, e.g., purchase or subscribe), ROAS at Day 7 and Day 30, and incremental volume (are you getting conversions you would not have gotten with broad targeting alone?).

Secondary metrics: Learning phase exit rate (if more than 30% of your lookalike ad sets fail to exit learning phase, your structure is too fragmented), audience overlap (check this in Ads Manager under Audiences; if two active ad sets have more than 30% overlap, consolidate them), and frequency (rising frequency with declining CTR signals audience saturation).

Diagnostic metrics: CTR (mobile app install ads on Feed tend to outperform those on Reels, so benchmark CTR expectations by placement), install-to-trial rate (a function of your onboarding, not just Meta targeting, but a useful diagnostic), and thumbstop ratio for video ads (commonly benchmarked as 3-second video views divided by impressions).

A common trap is comparing lookalike performance to broad targeting on CPI alone. A lookalike might have a 10% higher CPI but a 25% lower cost-per-subscriber because the users it attracts are more likely to convert downstream. Always evaluate on the deepest funnel metric you have reliable data for.

If you are using RevenueCat or a similar subscription analytics platform, pipe revenue data back to Meta through CAPI to enable value-based optimization and more accurate ROAS measurement.

Lookalike audiences in Meta Ads remain one of the most effective prospecting tools for mobile app growth in 2026, but they require more thoughtful execution than ever before. The post-ATT landscape has raised the bar for seed quality, CAPI implementation, and campaign structure. The practical takeaway is this: start with a value-based seed of 2,000-10,000 high-LTV users, test 1-3% lookalikes against broad targeting in a consolidated CBO campaign, and evaluate performance on cost per key event rather than CPM or CPI. If you need help building and testing lookalike strategies at scale, RocketShip HQ works with app teams to optimize every layer of Meta targeting, creative, and measurement. Your next step should be auditing your current seed audiences and replacing any install-based seeds with purchase or subscription-based alternatives, a change that consistently ranks among the highest-impact improvements we see in account audits.

Frequently Asked Questions

How do lookalike audiences work on TikTok and Google compared to Meta?

Meta offers the most sophisticated lookalike implementation with 1-10% granular targeting and mature algorithmic modeling built over years of social data. TikTok uses “Custom Audiences” with similar expansion capabilities, but their lookalike modeling relies heavily on in-app behavior and video engagement signals rather than broader web activity.

Google’s approach centers on “Optimized Targeting” which replaced their similar segments feature, focusing more on search intent and contextual signals across their ecosystem. Each platform accesses different signal sets: Meta leverages social connections and cross-app activity, TikTok emphasizes video consumption patterns, while Google prioritizes search behavior and browsing intent.

How has Apple’s App Tracking Transparency changed lookalike audience strategy in 2025-2026?

Successful advertisers have pivoted to server-side event tracking via Conversions API (CAPI) to maintain signal quality for lookalike construction, reducing reliance on pixel-based tracking. Seed audience construction now prioritizes first-party data sources like email lists, phone numbers, and CRM data over web pixel events.

Many brands leverage SKAdNetwork postbacks as seed sources, using iOS install and revenue data directly from Apple’s framework for more accurate modeling. Meta’s modeled conversions help fill data gaps, allowing advertisers to build lookalikes from statistically modeled events rather than only confirmed conversions.

When should you NOT use lookalike audiences?

Avoid lookalikes when your account generates 500+ weekly conversions, as broad targeting often outperforms with sufficient conversion volume for Meta’s algorithm. Skip lookalikes for very niche markets with total addressable audiences under 100,000 users, where broad targeting may actually reach more relevant prospects.

Brand awareness campaigns focused on maximum reach rather than conversion precision typically benefit more from broad demographic targeting or interest-based audiences. Finally, accounts with seed audiences smaller than 1,000 users lack sufficient data for meaningful lookalike modeling and should focus on building larger custom audiences first.

Can you combine lookalike audiences with retargeting in a full-funnel strategy?

A complete funnel strategy uses lookalikes for cold prospecting at the top, retargeting engaged users in the middle, and highly specific retargeting for bottom-funnel conversions. Structure your ad sets with proper exclusions: exclude middle and bottom-funnel audiences from your lookalike prospecting campaigns to avoid overlap and inflated costs.

A common starting framework allocates the majority of budget to top-funnel lookalikes, a meaningful share to middle-funnel engagement retargeting, and a smaller portion to high-intent bottom-funnel audiences like cart abandoners or app users who haven’t purchased. This approach maximizes reach while ensuring each funnel stage targets the most relevant audience without competition between your own ad sets.

What is the future of lookalike audiences as third-party data declines?

First-party data collection becomes critical as advertisers must own their customer relationships and data rather than relying on third-party cookies and tracking. Server-side measurement through APIs like Meta’s CAPI will dominate, allowing direct data sharing between advertiser systems and ad platforms.

Privacy-preserving technologies like differential privacy and federated learning will enable lookalike modeling while protecting individual user privacy. Meta’s on-platform signals (in-app activity, social interactions, content engagement) will gain importance as external web tracking diminishes, potentially shifting lookalike construction toward platform-native behaviors rather than cross-web activity patterns.