Why most retargeting underperforms
The most common retargeting mistake is running the same ad to every past visitor regardless of how they interacted with the brand. Someone who watched 50% of your brand video needs a different message than someone who started a checkout and abandoned. Treating both as one "retargeting audience" produces mediocre results for both.
The second most common mistake is a retention window that is too wide. A retargeting campaign targeting everyone who visited in the last 180 days includes people whose interest is long cold. Shorter windows — 7 to 30 days — produce higher conversion rates because recency correlates strongly with intent.
The three retargeting layers every funnel needs
Layer your retargeting by warmth. The hottest layer converts the most but reaches the fewest people. The coldest layer reaches more people but needs more creative work to re-engage.
- Hot layer (7–14 days): Cart abandoners, checkout starters, lead form openers who did not submit, high-intent page visitors (/pricing, /contact)
- Warm layer (14–30 days): All website visitors, product page viewers, video viewers 50%+, Instagram and Facebook profile engagers
- Cool layer (30–60 days): Video viewers 25–50%, people who clicked but did not visit a key page, email list custom audiences
Writing retargeting copy that works
Retargeting copy should acknowledge that the user has seen your brand before. References to "what you were looking at," "before you go," and "you were this close" outperform generic product ads because they are contextually honest about where the user is in the funnel.
For the hottest audiences, the message should be urgency and proof: limited availability, a testimonial from someone like them, or a specific objection addressed directly. For warmer audiences, the message should be a different angle on the value proposition — assuming the original angle did not convert them.
- Cart abandoner: "You left something behind. Here is why it is worth coming back."
- High-intent page visitor: "Seen our pricing page? Here is what teams like yours say about ROI."
- Video viewer: "You watched our overview. Here is what happens after you sign up."
- Cool retargeting: "You found us a while back. Here is what has changed since then."
Audience overlap and frequency management
Retargeting audiences overlap more than cold audiences because a single user can qualify for multiple layers. Use audience exclusions to keep the layers clean: exclude your hot layer from your warm layer campaign, and exclude your warm layer from your cool layer campaign. Without exclusions, the same person sees the wrong message at the wrong stage.
Frequency caps matter more in retargeting than cold traffic because the audience is smaller. A hot retargeting audience of 2,000 people exposed to a high-frequency campaign fatigues in days. Set frequency caps at 2–3 per week for the hottest audiences.
Using MCP Ads to monitor retargeting performance
Once your retargeting campaigns are running, ask Claude to compare CPL across your funnel layers and flag any audience with declining performance: "Show me my retargeting ad sets from the last 14 days, sorted by CPL, with frequency and CTR for each. Flag any ad set with frequency above 4 or CTR down more than 20% from the previous period."
That prompt runs in under 30 seconds with MCP Ads connected and gives you the full retargeting health check without manually checking each ad set.
Doing this with Claude connected to the account
A three-layer retargeting funnel is easy to describe and tedious to build: three custom audiences, three ad sets with the right exclusions, and a frequency check every week to see whether any layer is burning out. With Meta Ads connected through MCP Ads, each of those is a prompt.
Prompt 1: "List every custom audience in the account with its type, size, and the ad sets it is used in. Tell me which of the three retargeting layers — engagement, site visitors, cart abandoners — I am missing." Claude calls meta_ads_audiences, which returns the audience list with approximate sizes and source type, and cross-references it against meta_ads_list_adsets. What comes back is a gap list rather than a raw table. As an illustrative example: an engagement audience of 40,000 exists but is not used anywhere, a 30-day site visitor audience is live, and there is no cart abandoner audience at all.
Prompt 2: "Create a website custom audience of people who fired AddToCart but not Purchase in the last 14 days, and an engagement audience of anyone who interacted with the Instagram account in the last 90 days." Claude uses meta_ads_create_website_audience and meta_ads_create_engagement_audience. Both are writes: Claude shows the pixel, the event rules, and the retention window before creating anything, and waits for approval. If the pixel is not firing AddToCart at all, meta_ads_pixel_stats will show that first, which saves you from building an audience that never fills.
Prompt 3: "Show frequency and CPM for each retargeting ad set over the last 7 days versus the 7 before, and flag anything above frequency 4." This is meta_ads_frequency combined with meta_ads_adset_performance. The response is a short list of ad sets that need a creative swap or a cap, with the numbers to justify it. Building the ad sets themselves goes through meta_ads_create_adset, and anything Claude creates starts paused so you can check exclusions before a single impression is served.
The Meta Ads skills include a funnel analysis and a creative fatigue detector that run this as a scheduled review.
Where the manual way still wins
Audience sizes from the API are estimates, and Meta will not return a size at all for small audiences, so Claude cannot tell you exactly how many people are in a 7-day cart abandoner list or whether it will clear the delivery minimum this week. Overlap between two custom audiences is also not exposed through the API in a way Claude can query, so the audience overlap tool in Ads Manager still has a place.
The creative for each layer is where a human should stay. Claude can tell you the cart abandoner ad set is at frequency 6 and needs new creative; it should not be the one deciding what the offer in that ad is. Those choices depend on margin and on what you are willing to give away to close a hesitant buyer.
