On this page
Most guides to lookalike audiences explain Meta's setup process in detail and leave Google as a footnote. In practice, paid media managers running both platforms are using the same seed list on Meta and Google Performance Max and getting meaningfully different results, sometimes by a wide margin. That gap isn't random. Meta and Google read your seed audience through fundamentally different algorithms, so the strategy that lifts CPA on one channel can actively hurt the other. If you're spending on both, this article closes that gap.
Quick Answer
- Meta builds lookalike audiences from behavioral and social signals; Google Performance Max uses conversion history and Search intent data, so the same seed list produces materially different audiences on each platform.
- Seed quality matters more than seed size: a 500-person list of high-LTV buyers outperforms a 10,000-person list of all-time purchasers on both platforms, but the quality tiers look different for each channel.
- Running both platforms from a single audience strategy creates budget waste. An intent-stage allocation model, with Meta for upper-funnel and Google for lower-funnel, reduces CAC without reducing reach.
What Lookalike Audiences Are and Why the Platform Changes Everything
A lookalike audience is a targeting tool that takes a seed (your best customers, recent converters, or high-value users) and asks the platform to find new people who resemble them. That definition sounds platform-agnostic, but the matching algorithm behind it is not.
Meta and Google do not measure "resemblance" the same way. Upload the same customer list to both platforms and you get two audiences with different sizes, different behavioral profiles, and different conversion tendencies. Understanding why requires a brief look at what each platform actually reads.
How Meta Builds Its Lookalike Model
Meta's algorithm scores its user base against your seed on hundreds of behavioral and social signals: content engagement patterns, purchase behavior on Meta-tracked sites, app activity, demographic overlaps, and the social graph. When you create a 1% lookalike, you're getting the top percentile of users in your target country ranked by similarity score, per Meta's Lookalike Audiences documentation. That 1% threshold is the tightest similarity band Meta offers. The model is probabilistic and recalculates as your seed updates.
Since iOS 14.5 reduced pixel-based conversion data, Meta has leaned harder on first-party signals uploaded through the Conversions API and on its Advantage+ system, which treats your seed as a suggestion rather than a hard filter. A lookalike seed uploaded today tells Meta's AI where to start, not where to stop.
How Google Performance Max Reads Your Seed
Performance Max does not use a similarity score in the same sense. When you upload a Customer Match list, Google uses it to identify intent signals: search history, YouTube watch patterns, Gmail topics, and conversion behavior within Google's ecosystem. Then Google's system extends to reach people showing similar intent signals, even if their demographic profile looks different from your seed.
The practical implication is this: a seed built from past purchasers produces a Meta lookalike that resembles your customers demographically and behaviorally. The same seed in PMax produces a Google audience that looks like people actively researching to buy. That audience may not resemble your customers in a traditional sense, but they're further along in the purchase decision.
That's not a design flaw. It's why the two platforms complement each other when you use them correctly.
Why the Same Seed Produces Different Results on Meta vs Google
Running the same first-party customer list as a seed on Meta and Google Performance Max is the most common lookalike audience mistake in 2026. Meta ranks its 3.2 billion daily active users by behavioral and social similarity to your seed. Google PMax indexes intent signals such as search queries, YouTube engagement, and conversion paths, so users who resemble your seed's behaviors may never surface, while users who are actively researching your category do. BeFreed saw a 38% reduction in cost per install after Hellyeah's AIMA engine separated Meta and Google audience seeds by signal type rather than using one shared customer list across both channels.
In practice, the divergence shows up in two places: audience size and signal weighting.
On audience size: a 1% Meta lookalike from 1,000 purchasers in the US gives you roughly 2 million users. The same list in Google PMax produces a Customer Match audience that Google then extends, and how far it extends depends on how many of your customers Google has cross-device data on, typically 30-60% of uploaded email addresses. Your effective starting pool on Google is smaller, and the extension logic is opaque.
On signal weighting: Meta prioritizes recency and engagement. Your best seed for Meta is buyers from the last 30-90 days who interacted with your content. Google weights intent. Your best seed for Google is buyers who found you through Search or who showed active category research behavior, even if they bought months ago.
These two realities call for different seed segmentation rules on each platform, which is the most commonly skipped step in cross-platform audience strategy.
How to Build a Seed Audience That Performs on Both Platforms
The starting point is your CRM, not your ad account's pixel data. Pixel data reflects who Meta could track; your CRM reflects who actually bought. That distinction matters more post-iOS 14.5 than it ever did before. Upload a hashed customer list through Meta's and Google's respective data upload tools, rather than relying on pixel-generated custom audiences.
Segment before you upload, not after.
Tier Your Customer List by Value, Not Just by Conversion
For Meta, the highest-performing seed is your top 10-20% of customers by lifetime value, filtered to the last 90 days. Smaller is better when the signal is strong. A 500-person list of your highest-LTV buyers will outperform a 5,000-person list of all-time purchasers because Meta's algorithm weights the seed's internal coherence: users who look alike within your seed produce a tighter external match. Upload your full customer base and you're asking Meta to model everyone from your occasional buyers to your whales, which produces a noisy average.
For Google, the effective seed is broader and older. Google needs 1,000 or more matched users to activate Customer Match, and its intent-signal model benefits from seeing a wider range of purchase paths. A seed of 2,000-5,000 customers from the last 12 months works better on Google than the narrow 90-day LTV tier that Meta prefers. The key segmentation rule for Google is to filter out customers acquired through Meta-only touchpoints, since they may have no Google-side intent signals and typically produce lower Customer Match rates in Google Ads.
The Refresh Schedule That Prevents Performance Decay
Both platforms degrade when seeds go stale, but the degradation looks different.
On Meta, a seed that doesn't refresh regularly starts including churned customers and users whose behavior has shifted. Most practitioners refresh Meta seeds every 60-90 days at minimum. The lookalike audience keeps targeting based on a behavioral fingerprint that no longer matches your active buyers. You'll see frequency rising and CTR falling as the audience becomes less relevant.
On Google, the Customer Match list depreciates more slowly but the signal problem is subtler. Google de-weights old conversion events in its intent model, so a list built from Q1 purchasers will have less influence on PMax audience expansion by Q3. Refresh your Google Customer Match list monthly with your most recent converters to keep the intent signal current.
In a live campaign with BeFreed, AIMA handled seed refreshes automatically on both channels, pushing updated customer lists every 30 days and monitoring match rate changes to catch signal degradation before it showed up in CPA.
Author's Comment
"The seed refresh cadence is the piece most teams skip because it feels like maintenance, not strategy. But stale seeds are the single biggest reason lookalike audiences plateau at month two. AIMA was built to treat audience data as a live input, not a one-time upload, and that change alone is why we see continued CPA improvement where manual teams see decline."
Jay Ma, Co-Founder, Hellyeah AI
Setting Up Meta Lookalike Audiences for Awareness and Leads Objectives
Hellyeah's platform runs Meta Ads across two objectives: Awareness (optimized for impressions and reach) and Leads (using Meta's Instant Forms, which collect name, email, and phone inside Facebook or Instagram without a website click). The audience strategy for each objective differs, and lookalike audiences behave differently in each context.
For Awareness campaigns, a 3-5% lookalike of your video viewers or engaged audiences works well. The goal is reach, not conversion, so a looser similarity threshold increases the addressable pool without meaningfully reducing relevance. The seed for this tier should be users who watched at least 75% of your content. They've demonstrated intent to engage, which Meta weights heavily in its similarity model.
For Leads campaigns with Instant Forms, tighten the seed and the similarity percentage. A 1-2% lookalike of your best-fit leads (not all leads, but the ones who converted downstream) produces Instant Form responses your sales team can actually qualify. Many teams make the mistake of seeding Lead campaigns from all past lead form submitters. That seed includes low-intent respondents and poisons the lookalike toward users who click forms without converting.
When Advantage+ Outperforms a Manual Lookalike
Meta's Advantage+ Audience system treats your uploaded custom or lookalike audience as a suggestion it can expand beyond, rather than a hard targeting constraint. Meta positions this as a performance improvement, and for accounts with 50 or more weekly optimization events, the data generally supports that claim. In published testing across multiple advertiser accounts, Advantage+ Audience consistently produces lower CPA than manual lookalikes once the pixel has sufficient conversion history.
Below 50 weekly conversion events, manual lookalikes tend to outperform. Without enough conversion signal, Advantage+ has too little data to make its expansion decisions reliably, and the tighter manual lookalike keeps spend from drifting toward low-intent users.
The practical decision rule: if your account hits 50 or more purchase or lead events per week consistently, test Advantage+ Audience with your highest-LTV seed and compare against your best-performing manual lookalike over 14 days. If your account is newer or lower-volume, stay with manual 1-2% lookalikes built from your strongest seed segment.
Three Seed Sources That Consistently Outperform Interest Targeting
In a live Meta Ads campaign across consumer app clients, the seed sources with the widest performance gap over interest targeting are: high-LTV purchasers from the last 90 days; users who initiated checkout but didn't complete (a strong purchase intent signal); and users who watched 75% or more of video content and then visited the product page. Each signals a different stage of intent, and building separate lookalikes from each gives you a set of audiences you can test against each other without overlap.
Audience Size and Saturation Signals on Meta
The frequency metric is your saturation early warning system. When average frequency for a prospecting audience crosses 2.0-2.5 within a 7-day window, your audience is exhausting. The standard fix is to expand the similarity percentage (from 1% to 2-3%) or add a second seed segment to the lookalike creation.
Do not layer lookalike percentages in a single campaign. Running a 1%, 3%, and 5% lookalike as separate ad sets within the same campaign causes auction overlap, where your ad sets bid against each other for the same users. Each similarity tier should run in its own campaign with its own budget.
Setting Up Google Performance Max Audiences With Similar Segments
Google Performance Max does not offer lookalike audiences as a standalone feature. What PMax uses instead is Customer Match combined with similar segments, Google's equivalent of lookalike expansion. When you upload a Customer Match list, Google automatically generates a similar segment from it, targeting users with comparable intent signals across Search, YouTube, Gmail, Display, and Discover. The minimum list size for Customer Match activation is 1,000 matched users, which requires uploading a raw list of at least 2,000-3,000 emails given typical match rates of 30-60%.
The setup differs from Meta, and the interface is less explicit about what's happening. In Google Ads, navigate to Audience Manager under Tools and Settings, create a Customer Match audience from your CRM list, then attach it to your PMax campaign as an "audience signal." PMax treats it as a signal to optimize toward, not a boundary to stay within.
How PMax Uses Customer Match to Power Similar Segments
When you upload a Customer Match list to a Google Ads PMax campaign as an audience signal, two things happen. Google identifies your uploaded users in its ecosystem and measures their intent characteristics: what they search for, what YouTube content they watch, which email newsletters they engage with. Then it builds a similar segment automatically, targeting users not on your list who show the same intent pattern.
The quality of the similar segment depends on how many of your uploaded users Google can match and how consistent their intent signals are. A customer list made up of buyers who came through organic Search and Gmail channels will produce a richer similar segment than a list of buyers acquired through Meta ads, because the former group has deeper Google-side behavioral data.
When Google's AI Expands Beyond Your Seed
Performance Max will expand beyond the similar segment if it finds better-performing users elsewhere; the system moves when it believes conversion probability is higher in a different audience cluster. You can constrain this by tightening your Performance Max bidding (Target CPA or Target ROAS) and ensuring your conversion tracking is configured correctly with the Hellyeah tracker installed.
Without clean conversion data feeding PMax, the AI expands aggressively toward broad audiences with no performance anchor. This is the most common reason PMax campaigns with good audience signals still deliver poor CPA: the conversion signal is missing or fragmented.
The Saturation Signal in Google PMax
On Meta, saturation shows up as rising frequency and falling CTR. On Google PMax, it's more indirect. Watch for rising cost-per-conversion alongside falling new-customer share. When PMax starts recycling conversions from your existing retargeting pool rather than finding net-new prospects, you've hit the effective ceiling of your current audience signal. Fix it with a fresh Customer Match list using different segmentation logic, not a budget increase.
Hellyeah's X-Ray analytics surface this signal before it shows up in aggregate CPA, by separating new-user and returning-user conversion paths in the same reporting view.
Cross-Platform Budget Allocation When Running Both Channels
Running Meta lookalike campaigns and Google PMax with similar segments at the same time requires an allocation framework that accounts for the intent stage each platform reaches. Treating them as interchangeable awareness channels leads to paying for the same users twice at different funnel stages, which inflates effective CAC without adding incremental reach.
An Intent-Stage Model for Meta vs Google Spend
Meta lookalike audiences reach users before they have active search intent. Google PMax similar segments reach users while they're actively searching or showing high-intent behavioral signals. This difference suggests a natural allocation structure.
Upper-funnel budget (awareness, video, reach): 60-70% toward Meta. Your Meta Awareness campaigns with 3-5% lookalikes of video viewers work here. The goal is building brand recognition in the audience pool before they enter a search-active phase.
Lower-funnel budget (conversions, leads): 60-70% toward Google. PMax with a Customer Match signal of your highest-converting users, optimizing toward your core conversion event, catches the users who saw your Meta content and are now ready to convert through Search.
This allocation should shift based on where each platform is finding net-new users efficiently. When Meta's frequency rises and Google's new-user conversion rate is strong, shift budget toward Google. When Search volume for your category is flat and Meta frequency is low, lean into Meta.
How Hellyeah Handles Budget Reallocation Between Platforms
The Hellyeah platform, powered by AIMA, monitors both channels simultaneously and reallocates budget automatically when one platform's efficiency drops. In a live campaign across multiple app growth clients, AIMA detected Meta audience saturation (rising CPMs, falling CTR) and shifted spend to PMax within the same billing day, without a human approval step. The net effect was a 15-20% reduction in effective CAC during saturation events that would otherwise have required a manual campaign pause and budget adjustment.
This kind of cross-channel speed is the operational gap between managing Meta and Google separately versus running them as a unified paid growth system from the Hellyeah platform.
Measuring Whether Your Lookalike Audiences Are Adding Incremental Value
Lookalike audiences appear to work in standard attribution because they target users who resemble buyers and some of those users convert. What standard attribution doesn't show is whether your lookalike campaigns are adding net-new conversions or simply claiming credit for users who would have found you anyway through organic or direct channels.
For a proper read on lookalike lift, you need at least one of two measurement approaches.
Meta's Conversion Lift tool runs a holdout test: a portion of your lookalike audience is held back from seeing your ads, and Meta measures the difference in conversion rate between the test and holdout groups. This is the cleanest measure of incremental lift, and running it even once on your primary lookalike campaign gives you a baseline to compare against.
On Google PMax, incrementality testing is harder. Google's Conversion Lift tool for Performance Max uses geo-based holdouts (different regions see different ad exposure) which requires enough volume per region to reach statistical significance. For accounts under $10K/month on PMax, the holdout group may be too small to produce reliable results.
The metrics that tell you a lookalike is working without a full lift test: new-customer conversion rate (not total conversion rate), CAC for new users specifically, and the match rate between your seed and the converting audience. If 80% of your conversions from a lookalike campaign share the same behavioral cluster as your seed, the audience is doing its job. If the cluster distribution is random, your lookalike has drifted into broad targeting.
The Related blog posts section below covers attribution model comparisons and single-dashboard management for both channels.
Related blog posts
Frequently asked questions
What is a lookalike audience on Meta?
A Meta lookalike audience finds new users who share behavioral and social patterns with a seed audience such as your best customers. Meta identifies the top percentile of users in your target country ranked by similarity score to your seed, based on hundreds of signals including content engagement, purchase behavior, and the social graph.
Does Google Performance Max have lookalike audiences?
Google Performance Max does not use the term lookalike audiences. Instead, it uses Customer Match combined with similar segments. You upload a customer list as an audience signal, and Google automatically generates a similar segment targeting users with comparable intent signals across Search, YouTube, Gmail, Display, and Discover.
Can you use the same seed list for Meta and Google lookalike audiences?
Technically yes, but it underperforms because the two platforms read seed audiences differently. Meta prioritizes behavioral and social similarity; Google PMax prioritizes intent signals from Search and YouTube. A narrow 90-day high-LTV list works best for Meta, while a broader 12-month customer list works better for Google Customer Match.
How often should you refresh a lookalike seed audience?
Meta seeds should be refreshed every 60-90 days to prevent performance decay from stale behavioral data. Google Customer Match lists should be updated monthly to keep intent signals current for PMax audience expansion.
How do you know when a lookalike audience is saturated?
On Meta, watch for average frequency rising above 2.0-2.5 within a 7-day window alongside falling CTR. On Google PMax, look for rising cost-per-conversion with falling new-customer share, which signals the system is recycling from your retargeting pool rather than finding net-new prospects.

Co-founder
Co-founder of Hellyeah. Writes about building durable growth loops that compound over time.




