Why lookalikes stop working and how to refresh

Quick Answer

Lookalikes stop working for four reasons: (1) the seed has gone stale, (2) the audience is saturated, (3) creative fatigue is being mistaken for audience fatigue, or (4) Meta's algorithm has changed how it weights signals. The fix is a four-step refresh: rebuild the seed with fresh data, rotate creative, expand the lookalike size, and verify the issue is the audience and not the funnel.

Why lookalikes fade over time

Lookalikes are not static. Meta refreshes them every few days to keep them current with platform behaviour. But the seed you built them from gets older every day. If your seed is "customers from 2024" and it is now late 2026, the seed reflects an audience that may no longer match your current ICP.

More subtly, if you have been running the same lookalike for 60+ days at significant volume, you have shown your ad to most of the people in it. They are no longer fresh prospects — they are tired ones.

The four reasons performance drops

1. Stale seed. Your custom audience source is too old. Customers in the seed have moved on, behaviour has shifted, or the cohort no longer represents your buyer.

2. Audience saturation. You have served ads to most of the lookalike. Frequency is climbing, CPA is rising, and new conversions are slowing.

3. Creative fatigue. Your audience is fine, but you have run the same creative for too long. Click rate drops, then CPA rises. People mistake this for audience fatigue and rebuild the lookalike — then the new lookalike fails too because the creative is the actual problem.

4. Algorithm shift. Meta updates the signals it weights. Your old lookalike was optimised for last year's signal mix. The platform has moved on.

Step-by-step refresh playbook

1. Diagnose first. Pull the last 14 days of data. Look at: frequency, CTR, CPC, CPA. If frequency is climbing AND CPA is rising, audience fatigue. If frequency is steady but CTR is dropping, creative fatigue.
2. Refresh the seed. Rebuild the custom audience that feeds the lookalike using only the last 90 days of data. Old customers go out, new customers come in. Save with a new version number.
3. Rebuild the lookalike. Audiences > Lookalike > pick your refreshed seed. Save with a clear name like LAL_1%_Purchasers_v3_2026Q4.
4. Rotate creative. Even if the audience is the issue, fresh creative helps the relaunch. Three or four new ads minimum.
5. Launch as a parallel ad set. Do not pause the old one immediately. Run them side-by-side for 7 days. The new one should outperform.
6. Pause the old one once the new lookalike has clearly won.

Real examples

DTC food brand. Lookalike CPA went from £18 to £29 over 8 weeks. Frequency was 4.2 (saturation). Refreshed seed with last 90 days of purchasers, rebuilt 1% lookalike, and added new creative. CPA dropped back to £19 within a week.

B2B SaaS. Lookalike from a 2023 customer list was still being used in 2026. CPA had drifted from $42 to $78. Rebuilt seed using only 2025-onwards customers (a much smaller cohort). New lookalike CPA: $48. The seed was the problem all along.

Coaching business. Lookalike CPA tripled in two weeks. Frequency was steady. CTR had crashed from 1.4% to 0.6%. The lookalike was fine — the creative had fatigued. Three new ad variants and CPA returned to baseline.

How often to refresh

A loose schedule for healthy accounts:

• Seed refresh: Every 60 to 90 days


• Lookalike rebuild: Every 90 days, OR when CPA drifts more than 25%


• Creative refresh: Every 14 to 21 days (separate from audience)


• Audit: Monthly review of frequency, CPA trend, and overlap

FAQs

Will rebuilding the lookalike lose my learning data?
Not exactly. Meta keeps your event history. But the new ad set will go through a fresh learning phase (50 conversions). Make sure you have the budget to spend through it.

Should I delete the old lookalike or just pause the ad set?
Keep it for 30 days as a fallback. If the new one underperforms, you can revert.

Does Meta auto-refresh lookalikes?
Meta refreshes the people inside the lookalike automatically every 3 to 7 days. But the seed source is fixed when you create it. You have to rebuild from a refreshed seed to get genuinely new signal.

Can I avoid refreshing by using Advantage+ Audience?
Advantage+ reduces the need for manual refresh because Meta explores beyond your seed. But you still need fresh seed data for it to work from.

Is it worth keeping multiple lookalike versions running?
Yes, briefly. Run the new and old in parallel for 7 days, then pause the loser.

Refresh creative as fast as you refresh audiences

Half the time "audience fatigue" is actually creative fatigue. Pix-Vu tests fresh creative against your audiences before you launch them so you can refresh both at once and avoid the slow CPA decay. Try Pix-Vu free at https://pix-vu.com.