What Is a Similar Audience?
A similar audience (also called a lookalike audience) is a targeting segment built by an ad platform’s machine learning model to find new users who share behavioral, demographic, and interest patterns with an existing high-value customer group. Advertisers supply a seed list, such as purchasers or email subscribers, and the platform identifies millions of comparable profiles across its user base.
How Similar Audiences Work
The process runs in three stages.
- Seed definition. The advertiser uploads a source audience, typically a CRM list, pixel-based custom audience, or app event segment. Quality matters more than size; a list of 1,000 verified buyers generally produces a stronger model than 10,000 casual site visitors.
- Signal extraction. The platform profiles the seed against its first-party data, mapping shared attributes such as content consumption patterns, purchase intent signals, device usage, location patterns, and social graph connections.
- Match expansion. The model scores the full user base and surfaces users above a similarity threshold. Most platforms let advertisers control expansion breadth, trading precision for scale.
Meta’s Lookalike Audiences, Google’s Similar Segments, TikTok’s Lookalike Audience, and LinkedIn’s Lookalike Audiences each follow this core architecture, differing primarily in the signals available within their respective data ecosystems.
Similarity Score and Expansion Range
Meta expresses the expansion as a percentage of a country’s population, from 1% (highest similarity) to 10% (broadest reach). A 1% lookalike in the United States represents roughly 2.2 million people. A 10% lookalike covers approximately 22 million, with each step out diluting the average match quality.
Here is how that tradeoff plays out in practice:
| Expansion Level | Estimated US Reach | Typical Use Case |
|---|---|---|
| 1% | ~2.2M | Prospecting with tight ROAS targets |
| 2–3% | ~4.4–6.6M | Scaling proven creative sets |
| 5–10% | ~11–22M | Brand awareness, high-funnel reach |
Google’s Similar Segments do not expose a numeric slider; instead, the system auto-generates the segment and advertisers control reach through bid adjustments and campaign budget allocation.
Seed Audience Quality
The predictive power of a similar audience depends on the quality of its seed. A seed built from all website visitors carries noise from bounced sessions and accidental clicks. A seed built from customers who have made at least two purchases in the past 90 days carries a much cleaner signal of genuine buying intent.
Common high-signal seed sources include:
- Purchasers above a minimum order value threshold
- Subscription or loyalty program members
- High-lifetime-value customer cohorts from CRM exports
- Users who completed a key app event, such as checkout or level completion
- Email list segments with strong open and click history
Platforms generally recommend a minimum seed size of 100 to 1,000 matched users, with 1,000 to 50,000 as the practical sweet spot. Seeds larger than 50,000 can dilute the model’s specificity.
Real-World Performance Benchmarks
Dollar Shave Club, the subscription grooming brand acquired by Unilever in 2016 for $1 billion, used Meta Lookalike Audiences built from its subscriber seed during its growth phase. The approach reduced cost-per-acquisition by roughly 30% compared to interest-based targeting alone.
Mobile gaming publisher Jam City reported that TikTok lookalike segments drove a 50% lower cost-per-install versus broad demographic targeting in a 2022 campaign for its Harry Potter: Hogwarts Mystery title.
These figures vary by vertical, seed quality, and creative. Treat them as directional data points, not targets to replicate.
Value Similarity Index (Estimated)
Some advertisers estimate a rough value similarity index to compare seed and lookalike performance:
Value Similarity Index = (Lookalike Conversion Rate / Seed Conversion Rate) × 100
A score of 60 or above generally indicates the model has captured meaningful behavioral overlap. Scores below 40 often point to an undifferentiated seed or excessive expansion breadth.
Similar Audiences vs. Interest Targeting
Interest-based targeting groups users by declared or inferred affinities, such as “fitness enthusiasts” or “small business owners.” Similar audiences instead use the behavioral fingerprint of actual customers, which often surfaces users who would not appear in standard interest categories but share purchase patterns with existing buyers. The two approaches work best in combination: interest targeting for initial prospecting and hypothesis testing, similar audiences for scaling segments where conversion data already exists.
For related targeting concepts, see custom audience and behavioral targeting.
Privacy Constraints and Platform Changes
Apple’s App Tracking Transparency framework, introduced in iOS 14.5 in April 2021, significantly reduced the volume of mobile signals available to Meta’s lookalike models. Meta reported a $10 billion revenue impact in fiscal year 2022 attributed in part to signal loss. Platforms have responded by building modeled conversions and privacy-preserving APIs such as Meta’s Conversions API and Google’s Enhanced Conversions to partially restore signal quality without relying on device-level tracking.
LinkedIn deprecated its lookalike feature in early 2023, directing advertisers toward its Audience Expansion and Predictive Audiences tools instead. This reflects a wider industry trend of platforms building proprietary expansion logic rather than exposing explicit lookalike controls.
Advertisers who rely heavily on similar audiences should maintain parallel first-party data strategies, including email capture programs and CRM enrichment. This keeps seed lists robust as third-party signal availability continues to shift.
When to Use Similar Audiences
Similar audiences tend to perform best under the following conditions:
- The seed audience contains at least 500 to 1,000 high-intent users
- Conversion tracking is properly configured to validate downstream performance
- Creative is tailored to a prospecting mindset rather than a retargeting one
- The campaign objective is prospecting or upper-funnel volume, not remarketing
For advertisers in early-stage growth with limited conversion data, interest targeting or contextual advertising may be more appropriate until the conversion seed reaches a usable size. Once sufficient data exists, similar audiences offer one of the more efficient paths to scaling acquisition without proportional increases in cost per result.
For a deeper look at how audience data feeds into the broader media strategy, see audience segmentation.
Frequently Asked Questions About Similar Audiences
What is the difference between a similar audience and a lookalike audience?
Similar audience and lookalike audience are the same thing. “Lookalike audience” is Meta’s branded term; “similar segments” is Google’s label for the same concept. The underlying method is identical: a machine learning model finds new users who match the behavioral profile of an existing seed list.
How large does a seed audience need to be?
Most platforms require a minimum of 100 matched users to generate a similar audience, but 1,000 to 50,000 is the practical sweet spot. Seeds larger than 50,000 can dilute the model’s specificity. Quality matters more than size: 1,000 verified buyers outperform 10,000 casual site visitors as a seed.
Do similar audiences still work after iOS 14?
Yes, but with reduced precision on mobile inventory. Apple’s App Tracking Transparency framework cut the mobile signal volume available to Meta’s models, contributing to a reported $10 billion revenue impact in 2022. Platforms have partially offset this through modeled conversions and tools like Meta’s Conversions API, but mobile-heavy campaigns may see lookalike performance decline compared to pre-2021 results.
What makes the best seed audience for a lookalike?
High-intent, consistent customer data produces the strongest models. Purchasers above a minimum order value, loyalty program members, and high-lifetime-value CRM segments all outperform broad website visitor lists. The seed should represent the outcome you want to replicate, not just the users you have the most data on.
Did LinkedIn remove lookalike audiences?
LinkedIn deprecated its lookalike audience feature in early 2023. Advertisers were directed to Audience Expansion and Predictive Audiences as replacements. The move reflected a wider platform trend toward building proprietary expansion logic rather than exposing explicit lookalike controls to advertisers.