Meta Ads Optimization Recommendations#

When to Use#

Loaded when user wants to know what to do about their campaign performance — after seeing analysis results, or alongside analysis when they ask for both.

---

Prerequisites#

Before giving recommendations, you need performance data. Either:

1. Run an analysis skill first (meta-ads-analysis, meta-ads-lead-gen-analysis, meta-ads-audience-analysis, etc.)
2. Or use the analysis data already in the conversation

If no analysis data exists: "I need to look at your campaign data before I can give recommendations. What time period should I analyze?"

---

Session Context Keys This Skill Reads#

This skill is the single exit point for all analysis outputs. Read the relevant keys from session context before producing recommendations. See the full table at the bottom of this file.

---

Step 1: Understand the Full Picture#

Review the analysis conclusions and extract:

• Funnel weak points
• Trend signals
• Anomalies
• Data quality

Factor in brand or campaign context:

• Early-stage brand → higher CPL is normal initially
• Testing campaign → focus on learning, not scaling
• Scaling campaign → focus on efficiency and volume capacity

---

Step 2: Map Problems to Actions#

Problem You SeeLikely CauseRecommended ActionCTR less than 1% or decliningCreative fatigue, weak hook, audience mismatchTest new creative angles; narrow audience if broadLPV rate less than 70% or decliningAd doesn't match landing page; page loads slowlyCheck landing page relevance and speed; align messagingConversion rate decliningLanding page or offer issueReview page UX, form friction, offer claritycost_per_result risingSaturation, competition, fatigueCheck frequency; if > 3, refresh creative or widen audienceFrequency > 3Audience seeing ads too oftenExpand audience or rotate in new creativeSpend less than 50% of budgetAudience too narrow, bid too low, ad quality issuesWiden targeting; check if ad is stuck in learningStuck in Learning phaseNot enough results to exitIncrease budget or consolidate to fewer adsetsOne segment vastly outperformingSystem finding winners correctlyShift more budget to winner; pause underperformersAll healthy but low volumeBudget ceilingIncrease budget gradually — max 20% at a time

---

Step 3: Prioritize — Pick 1 to 3 Actions#

Prioritization rules:

1. Severity — How much is this hurting results right now?
2. Actionability — Can it be fixed with tools/data available?
3. Confidence — Is the data clear enough to be sure?

Optional additions (0–2, only if genuinely useful):

• Things worth monitoring but not urgent
• Early warning signs to watch

If nothing needs fixing: say so and give 1–2 things to keep an eye on.

---

Step 4: Landing Page Diagnostic#

When funnel data points to the landing page (LPV rate or CVR is the weak point), run this diagnostic before recommending creative or audience changes.

Step 4a: LPV Rate Check#

First — determine campaign type:

• Lead gen / Conversions objective → use Lead gen benchmarks below
• E-commerce / Purchase objective → use E-commerce benchmarks below

Campaign TypeLPV RateIndicatesE-commerce< 70%Landing page issue likelyE-commerce≥ 70%Ad-to-page alignment is healthyLead gen< 50%Investigate form or pageLead gen≥ 50%LPV is not the bottleneck

E-commerce only: If LPV is healthy but CVR is low → problem is deeper in the funnel (offer, pricing, trust signals).

Lead gen only: If LPV is healthy but CPL is still high → check form friction (Step 4c) and CAPI status (Step 4b) before concluding the landing page is fine.

Step 4b: CAPI Connection Check#

Before diagnosing an LPV or CVR problem:

1. Verify Meta pixel is firing on the landing page (check Events Manager → Test Events)
2. Verify Conversions API is connected and sending events back
3. If CAPI is not connected, CPL shown will be artificially high — this is a data problem, not an ad problem

Step 4c: Form Friction Check (Lead Gen)#

• Number of form fields: > 4 fields causes severe drop-off
• Form submit rate benchmark: > 20% is healthy
• If below 20%: recommend reducing fields to essential only
• Mobile usability: can users easily complete the form on phone?

Step 4d: Disconnect Diagnosis#

SignalLikely CauseRecommended ActionLPV < 70%, CVR OKAd-to-page messaging mismatchReview headline and CTA alignmentLPV OK, CVR < 50%Offer or landing page UX issueInvestigate page content and trust signalsBoth LPV and CVR lowFunnel-wide problemFix landing page first before changing adsLPV and CVR OK but CPL highAudience too broad or CAPI not connectedCheck targeting or verify CAPI is sending offline data

---

Step 5: Format Each Recommendation#

Every recommendation needs:

1. What — Specific action (which ad, which adset, what change, exact numbers)
2. Why — The data point that led to this conclusion

Optional: 3. Expected outcome — What should improve and by how much (only if it adds real clarity)

Good example:

Pause ad "Blue Banner v2" — Why: Spent $320 but only 2 leads (CPL $160) vs. account average CPL $45. CTR 0.6% vs. 1.8% average. Drags down overall efficiency.

Bad example:

"Consider optimizing your creative." (no specific action, no data)

---

Step 6: Constraints#

Budget change exceptions — when to break the 20% rule#

ConditionMax adjustment per changecost_per_result is 15%+ below target (winner signal)+50%cost_per_result is above target20% (standard)Newly launched, spending < 50% of budgetCheck delivery first, then adjust

General rules#

• Budget changes: Max 20% per adjustment — larger jumps disrupt learning
• Pausing ads: Pause only when CPL is 2x+ higher than average AND meaningful spend. Don't pause during learning phase unless clearly failing.
• Audience changes: Major targeting changes reset learning — prefer creating a new adset.
• Creative recommendations: Be specific about direction: "Test a hook around [specific pain point]", not just "new creative". Suggest 2–3 new variations.
• Landing page issues: Acknowledge we can't see the page directly. Infer from LPV rate and conversion rate data.

Auction Overlap#

When multiple ad sets share overlapping audiences, Meta excludes the lower-value ad from competing.

Diagnosis:

1. Check Opportunity score in Account Overview
2. Look for multiple ad sets stuck in Learning Limited simultaneously

Solutions:

1. Combine similar ad sets — consolidates learning
2. Turn off overlapping ad sets — move budget to the active ad set

---

Step 7: Output Format#

Brief context sentence, then:

Recommendations:

1. [Action title] — Why: [data-backed reason] — How: [specific steps in Ads Manager]
2. [Action title] — ...
3. [Action title] — ...

Other observations: Things worth watching, not urgent.

---

Multi-turn Refinement#

If user pushes back ("too aggressive", "I don't want to pause that"):

• Adjust only that specific point
• Don't repeat the full list unless asked

---

Tone#

Advisory — professional but warm. Specific and data-driven. "Execution needs to happen in Meta Ads Manager — I can walk you through the steps." "Once you've made these changes, I can re-analyze in a few days to see the impact."

---

Restrictions#

• Every recommendation must cite specific data — never vague
• Never recommend without data support
• Maximum 1–3 recommendations — quality over quantity

---

Session Context — What This Skill Reads#

This skill is the single exit point for all analysis outputs. Read the relevant keys from session context before producing recommendations:

KeyWritten ByDescriptionfunnel_weak_pointsmeta-ads-analysisWhere the biggest funnel drop-off occurstrend_signalsmeta-ads-analysisDirection of key metricsanomaliesmeta-ads-analysisUnusual findingsdata_qualitymeta-ads-analysisWhether data is sufficient to act onlp_diagnosismeta-ads-lead-gen-analysis (primary)Ad side vs. landing page side — lead-gen-specific diagnosislp_diagnosis_generalmeta-ads-analysis (fallback)Ad side vs. landing page side — general diagnosiscapi_statusmeta-ads-lead-gen-analysisCAPI connection statuscpl_breakdownmeta-ads-lead-gen-analysisWhich funnel stage is the CPL bottleneckrecommended_fix_prioritymeta-ads-lead-gen-analysisRanked fix order for lead genbudget_reallocation_planmeta-ads-audience-analysisSpecific audience budget shiftsaudience_issuesmeta-ads-audience-analysisOverlap and misallocation findingsrotation_pipelinemeta-ads-creative-fatigueCreative inventory by statusfatigue_levelmeta-ads-creative-fatiguePer-creative fatigue classificationdays_until_deathmeta-ads-creative-fatigueEstimated creative lifespanprimary_root_causemeta-ads-drop-diagnosisRoot cause of sudden performance droprecovery_planmeta-ads-drop-diagnosisStructured recovery steps

If no analysis context keys are present, ask the user: "I need to analyze your campaign data first. What time period should I look at?"