Meta Ads Audit

Diagnose Meta (Facebook + Instagram) account health and persist business context for downstream skills (/meta-ads). Read-only — never mutates the account. The user runs /meta-ads to execute fixes you recommend.

Setup

Follow ../shared/preamble.md — MCP detection, OAuth, ad account selection.

Filesystem contract (MUST persist)

Artifact	Path	When
Business context	{data_dir}/meta/business-context.json	First full audit, or refresh when audit_date is >90 days old. Skip on scoped audits if file is fresh.
Personas	{data_dir}/meta/personas/{accountId}.json	Every full audit.

These are the handoff to /meta-ads — write them even if the report itself is short. Otherwise downstream skills operate without business context and produce generic output.

If a {data_dir}/business-context.json exists from /google-ads-audit (no meta/ subdir), read it as a starting point — most fields (services, brand voice, differentiators, locations, seasonality) are platform-agnostic. Then write the Meta-specific version to {data_dir}/meta/business-context.json with any Meta-specific overrides (different creative angles, different audiences, different funnel events).

business-context.json schema (shared with Google Ads where fields apply): business_name, industry, website, services[], locations[], target_audience, brand_voice{tone, words_to_use[], words_to_avoid[]}, differentiators[], competitors[], seasonality{peak_months[], slow_months[], seasonal_hooks[]}, social_proof[], offers_or_promotions[], landing_pages{}, unit_economics{aov_usd, profit_margin, ltv_usd, source}, notes, audit_date, account_id.

Meta-specific extensions: meta_funnel_events{top_of_funnel, mid_of_funnel, conversion}, creative_inventory{concepts[], formats[], aspect_ratios[]}, custom_audiences{purchasers, abandoners, engagers, list_uploads[]}, pixel_health{pixel_id, capi_enabled, emq_score, last_event_at}.

personas JSON schema: {account_id, saved_at, personas: [{name, demographics, primary_goal, pain_points[], decision_trigger, value, meta_creative_angles[], visual_cues[]}]}. The Meta version adds meta_creative_angles (e.g. "before/after demonstration", "founder-led explainer", "UGC review") and visual_cues (objects, settings, emotions that resonate with this persona). See references/persona-discovery.md.

Policy freshness check (run first)

Read ../shared/policy-registry.json. For each entry where last_verified + stale_after_days < today:

• High-volatility → WebSearch the area for recent Meta Ads changes; compare to assumption. If drift, banner the report and suggest registry update.
• Moderate-volatility → one-line "may warrant a check" note.
• Stable → skip silently.

The Meta platform changes faster than Google Ads (Advantage+, attribution, learning behaviors) — check high-volatility entries every audit.

Phase 1 — Pull the audit dataset

Use a single runScript call with ads.graphParallel to fan out the queries an audit needs. Build the fan-out from this rubric.

A complete audit needs at minimum:

• Ad account info (/{accountId}) — currency, timezone, business id, spend cap, account status, balance.
• Pixel health (/{accountId}/customconversions + /{accountId}/adspixels) — pixel id, last activity, CAPI status, Event Match Quality (EMQ) score.
• Campaigns (/{accountId}/campaigns) — id, name, objective, status, daily/lifetime budget, special_ad_categories, buying_type, bid_strategy, created_time. Last 90 days.
• Ad sets (/{accountId}/adsets) — id, name, status, campaign_id, optimization_goal, billing_event, bid_strategy, daily_budget, lifetime_budget, attribution_spec, targeting (summary), promoted_object, learning_stage_info.
• Ads (/{accountId}/ads) — id, name, status, ad set, creative summary (image/video, primary text, headline, description, CTA), effective_status.
• Insights at campaign level (ads.insights({level:"campaign", date_preset:"last_30d"})) — spend, impressions, reach, frequency, cpm, link CTR, link clicks, purchases (or other primary action), purchase value, ROAS, CPA.
• Insights at ad set level — same fields, last 30 days.
• Insights at ad level — top 50 ads by spend; same fields plus video metrics (3-sec views, ThruPlays) for video creatives.
• Insights with breakdowns — placement (publisher_platform,platform_position), age/gender, device. Use these to spot placement losers and audience composition.
• Recent edit activity — when available via /{adsetId} last_modified or /{adsetId} change history.

Compute aggregates in the script, return summarized JSON. Don't return all rows — rank, slice, summarize. The agent narrates the result; the script does the math.

suggestImprovement is a useful cross-check for the server's heuristic surface — call it as a separate tool after the runScript pass if you want to compare your findings.

If a critical query errors out (auth, schema, API version), surface the error and stop — don't fall back to a degraded audit.

Skip scoring entirely if totalSpend == 0 or activeCampaigns == 0. Go straight to business context.

Phase 2 — Scope handling

If the user narrows the audit ("focus on one campaign", "campaign X", "just check creative fatigue"):

• Match campaign names by case-insensitive substring. If no match, list available campaigns and ask.
• Filter the in-memory dataset before scoring — no extra API calls.
• Account-level dimensions (Pixel health, attribution defaults) stay account-wide. Note "Scoped to: X" in the report.
• Skip Phase 4 (business context refresh) on scoped audits if business-context.json is fresh.

Phase 3 — Score

Score each of the 7 dimensions 0–5 using references/account-health-scoring.md. Overall = round(sum × 100 / 35).

Score	Label	Meaning
0	Critical	Broken or missing — actively losing money
1	Poor	Major waste or missed opportunity
2	Needs Work	Several clear issues
3	Acceptable	Functional, room to improve
4	Good	Well-managed, minor opportunities
5	Excellent	Best-practice

Scope-aware: campaign-level dimensions reflect in-scope data; account-level dimensions (Pixel + CAPI, attribution setup) score account-wide with a note on scope impact.

Encoded heuristics — apply these, they aren't obvious

• Pixel + CAPI is upstream of everything. EMQ < 7.0 means Meta can't match events well — Smart Bidding starves regardless of how good the creative is. STOP-condition input.
• Reported ROAS systematically overstates true ROAS. Cross-check Meta-reported numbers against Shopify / GA4 / MMM where possible. The gap is the modeled-conversion premium and is typically 20–40% in ecom.
• Frequency × CPM trend = creative diagnosis. Frequency > 3.0 with CPM rising ≥ 30% w/w is fatigue — recommend creative refresh, not budget cuts.
• One ad set carrying > 70% of a campaign is fragility, not concentration. When it fatigues, the campaign collapses.
• Audience overlap > 50% between sibling ad sets fragments signal. Consolidate; don't try to "fix" with bid caps.
• Special Ad Category misclassification is a takedown risk, not just a policy nit. Surface as Critical regardless of current performance.
• Manual placements without evidence is a sign of inherited-from-2018 thinking. Default should be Advantage+ Placements; deviations need data.

Pixel + Tracking Diagnosis Matrix

	EMQ < 5	EMQ 5–6.9	EMQ 7.0+
CAPI off	Critical — flying blind	Critical — most events lost	High — leaving 15–25% of events on the table
CAPI on, dedup off	Critical — duplicated and weak signal	High — duplicate counting risk	Medium — match quality improves with dedup
CAPI on, dedup on	High — match quality is the bottleneck	Medium — improve event_id coverage	Healthy

Phase 4 — Business context

Derive what you can from the data already pulled:

Field	Source
business_name	Ad account name (/{accountId} name field)
services	Top campaigns by spend, ad set names, top-converting ad creatives
locations	Targeting geo summary (countries / regions in active ad sets)
brand_voice	Top-performing ad copy (primary text + headline)
creative_inventory.formats	Mix of image / video / carousel observed in active ads
creative_inventory.aspect_ratios	Aspect ratios across active ads (1:1, 4:5, 9:16)
meta_funnel_events.conversion	Most common optimization event on top-spending ad sets
custom_audiences	Custom audiences referenced in active ad set targeting
pixel_health	From the Pixel detail call
website	Apex domain from active ad final URLs

Then crawl the website (homepage + about + 1–2 top landing pages, parallel WebFetch) and merge into the schema. See references/business-context.md for the full crawl procedure.

Always ask the user: differentiators, competitors, seasonality, AOV + profit margin (essential for ROAS-aware scoring). Ask for everything else only if data + crawl can't answer it.

Phase 5 — Personas

Discover 2–3 personas from creative performance (which angles convert), top-spending audiences, and landing-page content — all from the dataset already in memory. Persist to {data_dir}/meta/personas/{accountId}.json. Each persona must be grounded in observable evidence (a converting ad set, a converting creative angle, a landing-page section) — no inventing. See references/persona-discovery.md.

Phase 6 — Report

Lead with the verdict, then the top 3 actions (with dollar impact when possible), then the scorecard, then evidence for dimensions scoring 0–2 only. Cite specific campaigns, ad sets, ads, and dollar amounts. Cap at ~80 lines.

End with a single closing line after the handoff to /meta-ads:

Your audit history is saved to your NotFair account — view it at https://notfair.co.

Guardrails

1. Read-only skill. Diagnose; don't mutate. Every fix routes through /meta-ads. End the report with one handoff tied to the #1 action.
2. STOP condition — if Pixel health scores 0–1 (EMQ < 5 or CAPI off in an ecom account), recommend pausing scaling decisions until tracking is fixed before recommending anything else. Everything downstream is unreliable.
3. Always persist meta/business-context.json and meta/personas/{accountId}.json even if the report itself is short — downstream skills depend on them.
4. Name names. Every finding cites specific campaigns, ad sets, ad creatives, and dollar amounts. "Some ad sets are underperforming" is not a finding.
5. Never report Meta-reported ROAS without footnoting the modeled-conversion premium. "ROAS 3.2× (Meta-reported, 7DC1DV — typically overstates Shopify-attributed ROAS by 20–40%)" is honest. "ROAS 3.2×" is misleading.