Lookalike audiences have been a core Meta Ads tool for years. The logic is elegant: give the platform a list of your best customers, and it finds more people who resemble them. For a long time, it worked well enough that most paid media teams used it on autopilot. Then third-party data deprecation, iOS tracking changes, and Meta’s own algorithmic shifts reshuffled the deck. At Choco Media, we’ve been auditing how lookalike audiences actually perform in 2026 across client accounts, and the picture is more nuanced than “they’re dead” or “nothing changed.”
The short answer: lookalike audiences still work. But whether they work for your account now depends almost entirely on the quality of your seed data — something most advertisers haven’t updated since they set up their pixel years ago. This post walks through what we’ve changed, what we test, and the specific decisions that determine whether lookalikes pull their weight or quietly drain budget.
This is written for paid media managers and performance marketers who already run Meta Ads and want a tactical refresh, not a beginner introduction to the format.
Why Third-Party Data Deprecation Changed Lookalike Audiences
The original lookalike model relied heavily on behavioral data collected across the web — what people searched, what sites they visited, what they bought elsewhere. Meta’s pixel, combined with its vast off-platform data network, let the algorithm triangulate identity and interest with high confidence. That model weakened after Apple’s App Tracking Transparency (ATT) rollout in 2021 and has continued eroding as browsers restrict cross-site tracking.
The practical result: a pixel-based lookalike seed (people who visited your site or added to cart) now represents a smaller, less representative slice of your customer base than it did three years ago. The signal loss is real, and it shows up as broader, less precise lookalike expansion.
- Event match quality scores dropped for many accounts after ATT — even high-traffic sites see significant gaps between actual conversions and reported pixel events.
- Cookie-based audiences are smaller than they appear, because a meaningful portion of users browse with tracking disabled or across devices that don’t match.
- Modeled conversions (which Meta introduced to fill the gap) help with campaign optimisation but don’t improve the underlying seed data quality for lookalike generation.
None of this means lookalikes are broken. It means the quality of your first-party seed matters more than it ever did, because you can no longer lean on Meta’s third-party graph to compensate for a thin or stale seed list.
What Makes a Good Seed List in 2026
The most consistent performance we see across accounts comes from seeds built on actual customer behaviour rather than pixel events. The hierarchy, roughly in order of reliability:
- Purchasers (lifetime value segmented) — your top 20–30% of customers by LTV, not all purchasers. A broad “all buyers” list dilutes signal with low-value, one-time transactions.
- Email list — converted subscribers — people who opted in and then purchased, booked, or qualified. Not the full list.
- CRM export — active accounts — for B2B, decision-makers at accounts that closed or renewed in the last 12 months.
- High-engagement video viewers — 75%+ video view audiences can work as seed for TOFU lookalikes, but expect lower purchase intent downstream.
- Pixel events (purchase/lead) — still viable when event match quality is high (7+) and volume is sufficient, but verify your EMQ score before relying on it.
What doesn’t work well as a seed anymore: all website visitors, add-to-cart without purchase, broad email lists including non-buyers, and any pixel event where your EMQ is below 6.
The Size Floor
Meta’s official minimum is 100 matched profiles. In practice, seeds under 500 produce erratic lookalikes — the algorithm doesn’t have enough signal to define the audience reliably. The floor we work with is 1,000 matched profiles, and for most accounts we aim for 1,000–5,000. Above 10,000, additional seed size tends not to improve performance meaningfully unless you’re segmenting further.
A seed list of 800 high-LTV customers almost always outperforms a seed list of 50,000 one-time buyers. The algorithm is looking for consistent patterns, and one-time buyers are a noisy signal.
Lookalike Size: The Percentage Decision
Meta generates lookalikes as a percentage of the country population — 1% is the most similar, 10% is the broadest. The standard advice has always been to start at 1% and expand if volume is limited. That advice is still broadly right, but the nuance has changed.
- 1% lookalikes remain the default for prospecting when budget is limited and seed quality is high. They produce the smallest but most targeted audience.
- 2–3% lookalikes are often the practical sweet spot for accounts spending €3,000–€10,000/month on prospecting — enough volume to exit the learning phase without sacrificing too much precision.
- 5–10% lookalikes start to overlap significantly with interest-based or broad targeting. At this level, test directly against broad before assuming lookalike adds value.
One pattern we observe repeatedly: stacking lookalike percentages (using 1–3% and 3–6% as separate ad sets) often wastes budget on audience overlap rather than genuinely expanding reach. Use Audience Overlap in Meta’s Audience section to check before you stack.
When Broad Targeting Now Beats Lookalikes
This is the part of the conversation that matters most in 2026. Meta’s algorithm has become significantly better at finding buyers without explicit audience input. For many accounts — particularly those with consistent purchase history and good pixel data — running creative against a broad (no detailed targeting, no lookalike) audience now matches or beats lookalike performance.
This happens because Meta’s conversion optimisation is effectively building its own dynamic lookalike in real time based on who responds to your creative. When you provide a lookalike constraint, you’re sometimes limiting the algorithm’s search space to a predefined snapshot rather than letting it optimise dynamically.
- Broad targeting tends to outperform lookalikes when your account has 50+ purchase events per week — enough data for Meta to optimise without explicit audience guidance.
- Lookalikes tend to outperform broad when you’re launching in a new market, targeting a very specific buyer profile (e.g. B2B decision-makers), or when purchase volume is too low for the algorithm to learn quickly.
- The safest approach is to run both in an A/B test before allocating budget — not to assume one wins based on prior experience.
If you want a framework for deciding where to put your first budget as you scale, our post on Meta vs Google Ads in 2026 covers the channel decision that comes before the audience question.
Refreshing Your Seed: How Often and How
Stale seed data is one of the most common underperformance causes we find in paid media audits. An advertiser sets up a lookalike audience based on purchasers from 2022, it performs reasonably well, and they never touch it again. Meanwhile, their customer base has shifted — different channels, different products, different average order values — and the lookalike is quietly optimising toward a customer profile that no longer exists.
Our current practice:
- Quarterly seed refresh as a minimum. For fast-moving e-commerce or SaaS with short sales cycles, monthly is better.
- Segment by recency — a separate seed of purchasers from the last 90 days often outperforms a rolling 2-year list, because recent buyers reflect current positioning and creative.
- Export directly from your CRM or e-commerce platform, not from Meta’s pixel events. Platform exports include email, phone, and name fields that improve hash matching rates significantly.
- Check match rates after upload — anything below 40% suggests data quality issues (formatting, non-professional emails, outdated contacts) worth fixing before running.
For accounts running our paid media management, we build seed refresh into the quarterly review cycle rather than treating it as a one-time setup task.
Lookalike Audiences on Google: Customer Match and Similar Segments
Meta isn’t the only platform where this matters. Google’s equivalent — Customer Match combined with Similar Segments — fills much of the role that third-party audience targeting played before cookie deprecation in Chrome.
The mechanics differ from Meta, but the principles are the same:
- Customer Match lets you upload a CSV of customer emails or phone numbers, which Google matches against signed-in Google accounts. Match rates are typically 40–60% depending on list quality.
- Similar Segments (formerly Similar Audiences) are Google’s lookalike equivalent — available for Search, Display, YouTube, and Discovery campaigns. They were briefly removed and then reintroduced in modified form for non-personalised advertising contexts.
- For Performance Max, providing a customer list as an audience signal — rather than relying solely on pixel events — consistently improves early campaign learning speed in our experience.
One practical note: Google’s Customer Match requires an account in good standing with a sufficient spend history (the threshold varies, but accounts under €500 total lifetime spend often can’t access it). If you’re running both platforms, prioritise Customer Match setup on Google early — it unlocks before many other advanced targeting options.
The Lookalike Audit: Four Questions to Ask Your Account
Before your next campaign refresh, run through these four checks:
- When was this seed list last updated? If the answer is “when we set it up,” refresh it now before drawing any conclusions about performance.
- What is the current event match quality score for the conversion event I’m using as seed? Check in Events Manager. If it’s below 7, prioritise an upload-based seed over pixel events.
- Have I tested this lookalike against broad targeting in the last 90 days? If not, you don’t actually know which is winning — you’re running on assumption.
- What does my audience overlap look like between active ad sets? If lookalike ad sets are overlapping significantly with each other or with retargeting, you’re creating internal auction competition.
These four questions catch the majority of the lookalike performance issues we find in paid media account reviews. None of them require new creative or budget reallocation — they’re structural fixes that improve efficiency without spending more.
A Practical Testing Framework
If you’re unsure where to start, this is the sequence we’d recommend for a mid-market account spending €5k–€15k/month on prospecting:
- Export your top 20% of customers by LTV from the last 12 months. Aim for at least 1,000 records with email, phone, and name fields.
- Upload to Meta and Google. Note match rates.
- Create a 1% and a 2–3% lookalike from the Meta list. Create a similar segment from the Google Customer Match list.
- Set up a structured A/B test: Lookalike 1% vs. Broad on Meta, running the same creative, same budget, minimum 2 weeks.
- After 2 weeks (and at least 50 purchase events per variant), compare cost per result and ROAS. The winner gets 70% of prospecting budget; the other runs at 30% as a continuous test.
- Refresh the seed list every quarter. Re-run the A/B test if more than 90 days have passed since the last one.
This isn’t complicated, but it requires treating lookalike audiences as something you maintain rather than something you set up once. The accounts that do this consistently get more reliable performance data and spend less time debugging mystery underperformance.
What We’d Prioritise in 2026
If we had to distil everything above into a single priority order for a performance marketer in 2026:
- First: Clean up your first-party data. A good CRM export beats a sophisticated lookalike strategy built on a stale pixel.
- Second: Test broad vs. lookalike properly. Don’t assume either wins without data from your specific account.
- Third: Build a quarterly seed refresh into your workflow. Put it on the calendar now.
- Fourth: Check audience overlap. Stacked lookalikes eating each other is a common, silent budget drain.
- Fifth: Set up Customer Match on Google if you haven’t. It’s the highest-leverage move for accounts that only use Meta lookalikes today.
Lookalike audiences haven’t stopped working — they’ve just stopped being something you can set up once and forget. The shift to first-party data isn’t a temporary workaround; it’s the new baseline for paid media targeting. Accounts that invest in data quality now will compound that advantage as tracking restrictions tighten further.
If you’d like a review of how your current paid media setup handles audience targeting and seed data quality, get in touch — it’s usually the first thing we look at in any new account.