Skip to main content

Documentation Index

Fetch the complete documentation index at: https://adadvisor.ai/docs/llms.txt

Use this file to discover all available pages before exploring further.

The AdAdvisor targeting skill encodes the post-iOS 14 reality of Meta (Facebook) ads: broad has decisively won for prospecting, and creative does the targeting. Lebesgue’s 2025 data and Meta’s own published case studies converge. Broad targeting plus strong creative outperforms detailed interest stacking at virtually every budget level.

​
When should I use the targeting skill?

Triggers on phrases like:
  • “Audience”
  • “Lookalike”, “LAL”
  • “Retargeting”
  • “Interests”
  • “Exclude”
  • “Custom audience”
  • “Broad targeting”
  • “Audience size”
  • “Who should I target”
  • “Estimate reach”
  • “Build a lookalike”

​
What is the right Meta audience hierarchy in 2026?

For prospecting:
PriorityApproachWhen
1 (default)Broad + Advantage+ Audience ONMost ecom, most lead-gen, almost all DTC
2Lookalike (1% LAL of high-value seed)When you have a seed with 1,000+ high-value customers
3Lookalike (3% to 5% LAL)For scale, lower precision but more reach
4Interest targetingOnly when broad and LAL are exhausted, OR for very specific niches
5Lookalike (10%)Essentially equivalent to broad. Use broad instead.
6Detailed-interest stackingDead since June 2025 (Meta consolidated). Skip.
For retargeting (layered funnel):
LayerAudienceRetention
Warm-1All website visitors30 days
Warm-2View Content / Add to Cart30, 60, 90 days
Warm-3Email subscribers (customer list)n/a
EngagementPage / IG profile engagers180 days
Video viewers25%+ or 50%+ completion180 days

​
How do I build a Meta lookalike audience that actually works?

The skill’s seed-quality hierarchy, best to worst:
  1. LTV-tier customer list (top 25% by lifetime value)
  2. Repeat purchasers (2+ orders)
  3. All purchasers (180-day)
  4. High-engagement video viewers (75%+ completion)
  5. All-page-view (worst seed. Resembles broad.)
Then start at 1% LAL for precision and expand to 3% and 5% in separate ad sets as you scale. Value-based LALs (seeded with revenue data per customer) outperform basic LALs per Salesforce 2026 data, but require sending purchase value through pixel and CAPI. Meta requires seeds of 100+ people. The skill flags any seed below 1,000 as statistically thin.

​
Which exclusions should every prospecting ad set use?

Cannibalization is the silent killer. The skill enforces exclusions:
Otherwise you bill yourself to serve ads to people who already bought. Use excluded_custom_audiences on every prospecting ad set.
Same logic. Customer-list audiences from CRM go in the excluded list.
Otherwise ASC and manual retargeting bid against each other.
Avoid conflicting messaging from two different retargeting flows hitting the same person.

​
What audience sizes work for prospecting, retargeting, and lookalikes?

Use caseHealthy rangeBelow threshold
Prospecting1 million to 50 millionUnder 500K means slow learning. Over 50 million means Meta over-delivers poorly.
Retargeting1,000+Below 1,000 means delivery is unstable.
Lookalike seed100+ (Meta requirement)Under 1,000 is statistically thin.
Custom audience for active delivery1,000+Below means Meta pauses delivery.
The skill uses adadvisor:estimate_audience_size to validate any new targeting spec before committing.

​
What is the AdAdvisor targeting MCP workflow?

adadvisor:list_custom_audiences         → existing audiences with subtype and delivery_status
adadvisor:search_targeting(...)         → interest, geo, behavior IDs (never invent these)
adadvisor:estimate_audience_size(...)   → reach sanity check
adadvisor:create_website_audience(...)  → build retargeting from pixel events
adadvisor:create_lookalike_audience(...)→ build LAL from a seed (1 to 6 hours to populate)
adadvisor:update_adset_targeting(...)   → apply with auto-fix for deprecated interest IDs
The MCP server runs Meta’s /reachestimate validation and auto-fixes deprecated interest IDs (up to 3 retry attempts) before applying the change. If it cannot fix, the skill surfaces the exact validation error.

​
What anti-patterns does the targeting skill prevent?

Dead since June 2025. The skill defaults to broad with Advantage+ Audience on.
Retargeting is tactical, not scalable. The skill keeps prospecting on broad and lookalike.
Always exclude existing customers. The skill enforces this on every prospecting create_adset call.
Interest IDs are account-scoped per Meta. The skill always re-queries via search_targeting in the current account.
Advantage+ Audience finds buyers outside your spec. That is the point. The skill leaves it on unless there is a regulatory reason.
Meta requires 100+. The skill flags any seed below 1,000 as statistically thin.

​
How do I install the targeting skill?

Claude Code

Ships with the full plugin install.

Claude.ai & Desktop

Download adadvisor-targeting.zip and upload via Settings, Capabilities, Skills.

​
Frequently asked questions

Broad targeting with Advantage+ Audience ON, for almost every account. Lebesgue’s 2025 data and Meta’s own case studies converge: broad plus strong creative outperforms interest stacking at virtually every budget level. Detailed-interest stacking has been dead since Meta’s June 2025 detailed-targeting consolidation. The targeting skill defaults to broad.
LTV-tier customer list (top 25% by lifetime value) gives the best lookalikes. After that: repeat purchasers (2+ orders), then all purchasers within 180 days, then 75%+ video completers. All-page-view is the worst seed since it resembles broad. Meta requires 100+ people in the seed. Under 1,000 is statistically thin.
The percentage is the share of the country’s population that resembles your seed audience. 1% LAL is the most similar but smallest (about 2 million people in the US). 3% and 5% trade similarity for reach. Use 1% for precision when you have proven creative, 3% to 5% for scale. 10% LAL essentially equals broad targeting and is usually skippable.
Otherwise you are paying Meta to serve ads to people who already bought. Cannibalization inflates CPA and tanks measured ROAS. The targeting skill enforces excluded_custom_audiences=[purchasers] on every prospecting create_adset call. Use a 30 to 90 day purchaser audience as the exclusion.
Advantage+ Audience lets Meta find buyers outside your targeting spec when it sees high-value signal. It is on by default on most objectives in 2026 and the targeting skill leaves it on unless there is a regulatory reason to disable it. Disabling it deliberately limits Meta’s ability to find buyers and almost always hurts performance.
1 million to 50 million people. Below 500K means slow learning (Meta cannot collect 50 conversions per week to exit the learning phase at a healthy budget). Above 50 million, Meta over-delivers poorly because the signal-to-noise ratio drops. The skill uses adadvisor:estimate_audience_size to validate before committing.

Launch

Pairs with launch when targeting is part of a new campaign.

Scale

Horizontal scaling = audience expansion via new LAL tiers or geos.

Audit

Audit surfaces audience overlap and missing exclusions. This skill fixes them.
Last modified on May 13, 2026