Skip to content
All posts

Lookalike Audiences for CTV: Why Seed Quality Beats Scale

November 22, 2026 · 5 min read

Connected TV advertising has borrowed a lot of playbook pages from social, and lookalike audiences are one of the biggest imports. But CTV is not social. The targeting mechanics, the measurement lag, and the way platforms build lookalikes all behave differently once you move from a feed based environment to a living room screen. If you buy or plan media across CTV and social, understanding how lookalike audiences for CTV actually work, and where they break down, will save you a lot of wasted spend.

Why CTV Lookalikes Behave Differently Than Social Lookalikes

On social platforms, lookalike models have years of behavioral signal to draw from: likes, shares, dwell time, click patterns. CTV environments are thinner. Most CTV lookalike models are built from a mix of household level identifiers, limited ad interaction data, and third party data layered on top. That means the quality of your seed list matters even more on CTV, because the platform has less of its own behavioral data to fill in the gaps.

A few practical differences worth knowing:

  • CTV lookalikes are usually modeled at the household level, not the individual level, so you lose some precision compared to social.
  • Measurement lag is longer. You might not see a true signal on performance for two to three weeks, versus a few days on social.
  • Inventory is more fragmented across apps and devices, so even a well built lookalike audience can get spread thin if your budget is modest.

None of this means CTV lookalikes do not work. It means you have to feed them better and expect a different rhythm.

Start With a Seed Audience That Actually Reflects Conversion

The single biggest driver of lookalike performance, on any channel, is the seed audience. If your seed list is built from people who merely clicked an ad, your lookalike will find more clickers, not more buyers. For insurance, home services, solar, Medicare, and finance verticals, the seed should be built from your best available signal of a real outcome:

  • Closed sales or funded policies, not just submitted leads
  • High intent actions like quote completions or scheduled appointments, when sales data volume is too thin
  • Recent conversions, ideally within the last 60 to 90 days, since consumer behavior and eligibility change quickly in these verticals

If you only have a small batch of true conversions, resist the urge to pad the seed with lower quality leads just to hit a volume threshold. A smaller, cleaner seed almost always outperforms a larger, noisier one on CTV, because the household modeling has less to work with and needs every signal to be meaningful.

Build Separate Lookalikes for CTV and Social, Even From the Same Seed

A common mistake is building one lookalike audience and pushing it everywhere. CTV and social lookalike engines model differently, and treating them as interchangeable wastes the advantage of each platform's strengths.

Instead:

  • Upload the same seed data to each platform separately and let each platform's own modeling do the work
  • Set different lookalike percentage thresholds. CTV often performs better with a slightly broader match percentage because inventory is more limited
  • Review performance independently. A lookalike that crushes it on social may underperform on CTV simply because of device and household targeting differences

Running them separately also gives you a cleaner read on which channel is actually driving quality leads or sales, instead of muddying attribution with a single blended audience.

Refresh Your Seed Data on a Schedule, Not an Afterthought

Lookalike models decay. As your seed audience ages, it starts to reflect who converted months ago, not who is likely to convert now. This is especially true in Medicare and insurance, where enrollment periods and rate changes shift who is actively shopping.

A practical refresh cadence:

  • Weekly or biweekly refreshes for high volume campaigns with steady conversion flow
  • Monthly refreshes at minimum for lower volume campaigns
  • Immediate refresh after any major shift, like a new product launch, a state expansion, or a seasonal enrollment window

Set a calendar reminder if your team does not have an automated pipeline. Stale seed data is one of the most common and most fixable reasons lookalike campaigns quietly underperform over time.

Sync Lookalike Targeting Across CTV and Social for a Cleaner Funnel

One underused tactic is coordinating lookalike targeting across channels so CTV and social reinforce each other instead of operating in silos. CTV is strong for reach and brand presence, social is strong for direct response and retargeting. When both are pulling from aligned seed data, you get a more consistent full funnel experience.

Ways to do this:

  • Use the same conversion based seed logic across both channels, even if the lookalike percentages differ
  • Layer social retargeting against households that were likely exposed to your CTV lookalike campaign, using whatever frequency or geographic overlap data is available
  • Watch for audience overlap and adjust bids or budgets so you are not bidding against yourself across platforms

This kind of coordination is more work upfront, but it tends to lower blended cost per acquisition because the two channels stop competing for the exact same households independently.

Set Realistic Expectations for Testing and Optimization

CTV lookalike testing requires patience that social advertisers are not always used to. Because of the measurement lag and the household level targeting, you need a longer test window and a larger budget commitment before drawing conclusions.

Practical guardrails:

  • Give any new CTV lookalike at least three to four weeks before making a kill or scale decision
  • Track downstream metrics, not just impressions or completion rates. Tie performance back to actual leads, appointments, or sales where possible
  • Avoid changing the lookalike percentage and the seed data at the same time, so you can isolate which change actually moved performance

Closing Thoughts

Lookalike audiences for CTV can genuinely extend the reach of your best performing customer profiles, but only if you respect how differently CTV models behave compared to social. Clean, conversion based seed data, separate audience builds per channel, regular refresh cycles, and patient testing windows are what separate CTV lookalike campaigns that scale from the ones that quietly burn budget.

If you want help building cleaner seed audiences or syncing targeting across CTV and social, Book a demo.