Every paid social platform offers lookalike audiences. Not every platform delivers them equally. Run a Meta lookalike off a 200-person email list and you'll get noise. Run a LinkedIn lookalike off a 500-account CRM export and you might find your next enterprise customer. The mechanism, seed data requirements, match quality, and strategic purpose differ enough across platforms that treating them as interchangeable is one of the more expensive mistakes in paid social.
This post is an opinionated breakdown of how lookalike audiences work on Meta, LinkedIn, TikTok, and Reddit - and how to build a cross-platform strategy that accounts for those differences.
What Lookalike Audiences Actually Do and Why Seed Data Quality Is Everything
Lookalike audience algorithms work by identifying common attributes among people in your seed list, then finding users on the platform who share those attributes. On Meta, it's behavioral signals and interest graphs; on LinkedIn, it's professional identity data; on TikTok, it's engagement and content consumption patterns.
The seed data determines everything. A lookalike built from your best 500 customers will outperform one built from all 5,000 email subscribers - because the algorithm finds people who look like your actual buyers, not people who signed up for a lead magnet three years ago.
Seed data types by quality tier:
- Purchasers or closed-won customers (highest quality)
- High-value website visitors (checkout abandoners, pricing page visitors)
- Engaged leads with explicit product intent
- General email subscribers or social followers (lowest quality)
Meta Lookalike Audiences: How They Work in a Post-iOS-14 World
Meta's lookalike audiences are the most battle-tested in the industry. The algorithm has access to behavioral signals across Facebook, Instagram, Messenger, and the Audience Network - including off-platform conversion signals via the Meta Pixel and Conversions API.
How they work: You upload a seed audience, set a percentage range (1% - 10% of the target country's population), and Meta identifies users who statistically match the seed's behavioral and interest attributes. A 1% lookalike in the US is roughly 2.2 million people.
Post-iOS-14 mitigations: Use the Conversions API (CAPI) alongside the Meta Pixel for server-side conversion events. Upload customer lists directly rather than relying solely on pixel-based audiences. Use aggregated event measurement.
What Meta lookalikes are best for: Volume-driven prospecting for B2C and high-volume B2B. The algorithm's optimization depth is unmatched. Where they struggle: Precision targeting for niche B2B categories like VP of Engineering at mid-market logistics companies.
LinkedIn Lookalike Audiences: When Higher Cpcs Are Justified by Match Quality
LinkedIn lookalike audiences are built on professional identity data - company size, industry, job title, seniority, skills - rather than behavioral patterns. The minimum seed size is 300 members or 300 companies.
The B2B case: If your seed list is 500 closed-won customers that are all 100 - 500 employee SaaS companies in financial services, LinkedIn's lookalike finds companies matching those firmographic attributes. Meta cannot do this reliably. The trade-off is CPCs of $8 - $15 vs. $1 - $3 on Meta.
Best for: Enterprise SaaS with long sales cycles, recruiting-adjacent products, financial services and compliance products.
TikTok and Reddit Lookalike Audiences
TikTok
TikTok's lookalike is built on content consumption and engagement signals. It skews toward a younger demographic (18 - 34) and entertainment-forward content. Best for: Consumer brands with visual products, entertainment, gaming, fashion. Minimum seed: 100 users, but 1,000+ performs significantly better.
Reddit's audience similarity is based on community interest graphs - the subreddits users engage with. Best for: Niche technical audiences, developer tools, cybersecurity, open-source software, gaming. Reddit is not a primary channel - it's a precision supplement for technical niches.
Building a Cross-Platform Lookalike Strategy
Seed list architecture: Build three distinct seed lists: (1) Closed-won customer list for highest-quality seeding. (2) Engaged pipeline - CRM contacts who reached demo or proposal stage. (3) High-intent website visitors - pricing/demo page visitors from the last 90 days.
Percentage range guidance: Meta: Start at 1 - 2% for quality, expand to 3 - 5% if performance holds. LinkedIn: Standard lookalike, monitor performance. TikTok: 1 - 5% testing range.
For B2C: Meta first, TikTok second. For B2B: LinkedIn for precision, Meta for scale. For how much budget to allocate to lookalike prospecting vs. retargeting, prospecting lookalikes fill the top of funnel while retargeting converts the warm pool.
Refresh Cadence for Seed Audiences
Seed lists age. Customers who converted six months ago may no longer represent your best buyers, and a stale seed produces a lookalike that mimics yesterday's demand. Set a refresh interval that matches your sales cycle: fast consumer products can refresh quarterly, long enterprise deals annually or after a major campaign.
Refresh does not mean discard. Compare the new lookalike's performance against the old one before cutting budget, because a small shift in seed quality can move cost per result more than a new creative ever will. Treat the seed as a living input, not a set-and-forget setting.
Sizing Lookalikes Correctly
Bigger is not better by default. A 1 percent lookalike is a tight mirror of your seed and usually the highest intent; a 10 percent lookalike spreads into weaker matches that behave more like a broad audience. The right size depends on how much volume you need against how much precision you can afford.
Run adjacent sizes as separate ad sets so you can read the tradeoff directly. Many accounts find the sweet spot is a small core lookalike for efficiency plus a larger one for scale, with budget shifted toward whichever is hitting target cost. Sizing is a lever, not a fixed choice.
Privacy Changes and Their Long-Term Effect
iOS and browser privacy shifts reduced the signal platforms use to build lookalikes, which is why post-2021 match quality varies so much by channel. Platforms compensate with modeled audiences, but modeled reach is not the same as verified similarity, and performance can drift without warning.
Plan for a future with less deterministic signal by diversifying seed sources: first-party CRM data, site engagement, and value-based events beat a single email upload. The accounts that treat lookalikes as one input among several weather the privacy changes better than those that depend on one fragile seed.
Frequently Asked Questions
What Are Lookalike Audiences in Paid Social?
Lookalike audiences are prospecting audiences built by algorithms that identify users who share attributes with your seed audience. Each platform uses different underlying signals: Meta uses behavioral patterns, LinkedIn uses professional identity data, TikTok uses content engagement, and Reddit uses community interest graphs.
How Big Does Your Seed Audience Need to Be for Lookalike Audiences?
Meta requires a minimum of 100 people but performs significantly better with 1,000+. LinkedIn requires 300 contacts or 300 companies. TikTok works with 100 users but improves substantially with larger seeds. Seed quality matters more than size.
Do Lookalike Audiences Still Work After iOS 14?
Meta lookalike audiences have been affected by iOS-14's tracking restrictions. The mitigation is using the Conversions API (CAPI) for server-side event tracking and uploading customer lists directly. Lookalikes still perform - the impact is more visible on low-seed-size campaigns.
Key Takeaways
- Lookalike audience quality is determined first by seed data quality - closed-won customers outperform general email lists by a significant margin as a seed source.
- Meta lookalikes are volume-optimized and best for B2C; LinkedIn lookalikes are professionally filtered and best for precision B2B targeting despite higher CPCs.
- TikTok lookalikes skew toward younger demographics; well-suited for consumer brands with visual products and less reliable for B2B.
- Reddit lookalike audiences are community interest-based and most effective for niche technical audiences.
- Cross-platform lookalike strategy requires treating each platform's lookalike as serving a distinct role - not running the same seed list everywhere.
- Start at 1 - 2% on Meta for quality prospecting; expand only after confirming performance holds.