AGENT SKILL

meta-ads-tracking-setup

Implement and validate Meta Pixel + Conversions API (CAPI) for maximum signal quality — the single biggest lever for Meta Ads performance via the Andromeda ranking system.

License 许可
MIT
最近更新
2026-09-16
Stars
★ 19
Forks
2

01 前言 Front Matter

name
meta-ads-tracking-setup
description
Implement and validate Meta Pixel + Conversions API (CAPI) for maximum signal quality — the single biggest lever for Meta Ads performance via the Andromeda ranking system.

02 资源文件 1

03 SKILL.md 正文

Why This Matters

Meta's Andromeda deep learning architecture uses signal quality as its primary input for ad delivery optimization. Weak tracking = weak signals = poor delivery = wasted spend. Getting Pixel + CAPI right is foundational — do this before any campaign optimization.

Process

  1. Audit current state — Check Events Manager for active Pixel, CAPI status, EMQ scores, event volume, and deduplication status
  2. Choose implementation path — Select CAPI method based on technical resources and budget (see Decision Tree below)
  3. Implement CAPI — Deploy server-side tracking alongside existing Pixel
  4. Configure event deduplication — Match event_id between Pixel and CAPI to prevent double-counting
  5. Optimize EMQ — Add all matching keys to maximize user match rate
  6. Validate — Use Test Events tool to confirm both Pixel and CAPI fire correctly
  7. Monitor — Track EMQ scores, event volumes, and deduplication rates weekly for 30 days

CAPI Implementation Decision Tree

Ask these questions in order:

Q1: What platform runs the website?

  • Shopify → Use native Shopify integration (free, built-in CAPI, minimal setup)
  • WordPress/WooCommerce → Use CAPI Gateway or server-side GTM
  • Custom-built site → Use server-side GTM or manual API
  • No website (CRM/offline events only) → Use manual API with action_source: system_generated

Q2: What's the technical budget?

Method Cost Dev Effort Best For
Native platform integration (Shopify, WooCommerce plugins) Free–$50/mo Low Ecommerce on supported platforms
CAPI Gateway $10–$400/mo hosting Low–Medium Teams wanting managed solution
Server-side GTM $10–$50/mo (GCP) Medium Teams already using GTM
Manual API implementation $500–$5K+ dev time High Custom platforms, maximum control

Q3: Are there offline/CRM conversions?

  • Yes → Must implement CAPI with stored fbc values (see fbclid Capture Workflow below)
  • CRITICAL (May 2025): Meta permanently discontinued the Offline Conversions API. All offline tracking now flows through standard CAPI with action_source set to physical_store or system_generated

Pixel vs CAPI Comparison

Dimension Meta Pixel (Client-Side) Conversions API (Server-Side)
How it fires JavaScript in browser Server-to-server HTTP POST
Blocked by ad blockers Yes No
Blocked by iOS ATT Partially No (server-side)
Blocked by cookie restrictions Yes (3P cookies) No
Data freshness Real-time Near real-time (seconds)
Data quality control Limited Full (you control the payload)
Events supported Web events only Web + offline + CRM + in-store
Required for Baseline tracking Signal recovery + offline events
Typical signal recovery Baseline Recovers 20–30% of events lost to browser restrictions

Both are required. Pixel alone misses 20-30% of events. CAPI alone misses real-time behavioral signals. Together they give Meta's algorithm the fullest picture.

Event Deduplication Setup

When both Pixel and CAPI fire for the same event, Meta must know they're the same event to avoid double-counting.

How it works:

  1. Generate a unique event_id on your server or in the browser (UUID or timestamp-based)
  2. Pass the same event_id to both the Pixel event AND the CAPI event
  3. Both events must share the same event_name (e.g., both say Purchase)
  4. Meta deduplicates by matching event_id + event_name

Implementation pattern:

// Browser: Pixel fires with event_id
fbq('track', 'Purchase', {value: 99.99, currency: 'USD'}, {eventID: 'order_abc123'});

// Server: CAPI fires with same event_id
POST to graph.facebook.com with event_id: 'order_abc123'

Validation: In Events Manager, check "Overview" tab → deduplication should show matched events. If CAPI event count far exceeds Pixel, deduplication may be failing.

EMQ Optimization Checklist

Event Match Quality (EMQ) measures how well Meta can match your sent events to Meta user profiles. Higher EMQ = better attribution = better ad delivery.

Targets:

  • 6.0+ = Good (minimum acceptable)
  • 8.0+ = Excellent (target this)
  • Below 6.0 = Degraded performance — fix immediately

Matching keys to send (priority order):

Key Parameter Impact Notes
Email (hashed) em Highest SHA-256 hash, lowercase, trimmed
Phone (hashed) ph High SHA-256 hash, include country code, digits only
fbc (click ID) fbc High Captured from fbclid URL parameter — critical for offline conversions
fbp (browser ID) fbp Medium Read from _fbp first-party cookie
IP address client_ip_address Medium User's IP, not server IP
User agent client_user_agent Medium Browser user agent string
First name (hashed) fn Lower SHA-256 hash, lowercase
Last name (hashed) ln Lower SHA-256 hash, lowercase
City (hashed) ct Lower SHA-256 hash, lowercase
State (hashed) st Lower SHA-256 hash, lowercase, 2-char code
Zip code (hashed) zp Lower SHA-256 hash
Country (hashed) country Lower SHA-256 hash, 2-char ISO code
Date of birth (hashed) db Lower SHA-256 hash, YYYYMMDD format
Gender (hashed) ge Lower SHA-256 hash, 'm' or 'f'
External ID (hashed) external_id Lower Your system's user ID, hashed

Rules:

  • ALL customer data must be SHA-256 hashed before leaving your server — Meta never receives raw PII
  • Send as many keys as you have — more keys = higher match rate = higher EMQ
  • Email alone typically gets EMQ to ~5-6. Email + phone + fbc gets to 7-8+
  • Read references/capi-event-parameters.md for full parameter specifications and hashing requirements

fbclid Capture & Storage Workflow

For CRM and offline events sent hours/days/weeks after the ad click, the fbc value is the critical link connecting the downstream conversion to the original ad.

Step-by-step:

  1. User clicks ad → lands on your site with ?fbclid=xyz in URL
  2. Your landing page JavaScript captures fbclid from the URL
  3. Store fbclid alongside the lead record in your CRM/database
  4. Also read and store the _fbp cookie value
  5. Construct the fbc parameter: fb.1.{timestamp}.{fbclid_value}
  6. When the CRM event fires (e.g., sale closes 14 days later), send it via CAPI with the stored fbc value
  7. Event acceptance window: Maximum 7 days from event time. Send CRM events as soon as possible after they occur

action_source Parameter Values

Every CAPI event requires an action_source telling Meta where the event originated:

Value Use When
website Event happened on your website (most common for web CAPI)
app Event happened in your mobile app
phone_call Event was a phone conversion
chat Event originated from a messaging conversation
email Event resulted from an email campaign
physical_store In-store purchase or offline event at a physical location
system_generated CRM event, backend event, or any system-triggered conversion
other Doesn't fit above categories

Post-May 2025 migration: If you were using the Offline Conversions API, change your integration to standard CAPI and set action_source to physical_store (for in-store events) or system_generated (for CRM/backend events).

Validation Workflow

Using Meta Test Events Tool:

  1. Go to Events Manager → your Pixel → Test Events tab
  2. Open your website in a new tab
  3. Perform a conversion action (e.g., submit a form, complete a purchase)
  4. Return to Test Events — confirm you see:
    • One Pixel event AND one CAPI event for the same action
    • Both showing the same event_id
    • Status shows "Deduplicated" (not "Duplicate" or missing match)
  5. Check EMQ score for the event — identify any missing matching keys

Common validation failures:

Symptom Likely Cause Fix
CAPI events not appearing Server not sending, API auth error Check server logs, verify access token
Double-counted events Missing or mismatched event_id Ensure same event_id in Pixel and CAPI calls
Low EMQ despite sending keys Hashing errors, wrong format Verify SHA-256 hashing, check lowercase/trim
CAPI events far exceed Pixel Pixel blocked but no dedup issue Normal — CAPI recovers blocked events. Verify dedup still works for events both capture
Events delayed >7 days Exceeds acceptance window Speed up CRM-to-CAPI pipeline. Events must fire within 7 days

Output Format

When asked to help with Meta tracking setup, produce this deliverable:

## Meta Tracking Implementation Plan

### Current State Assessment
- Pixel status: [Active/Inactive/Not installed]
- CAPI status: [Active/Inactive/Not configured]
- EMQ score: [X.X / 10]
- Event deduplication: [Working/Not configured/Failing]
- Events tracked: [List of standard and custom events]
- Missing matching keys: [List]

### Implementation Steps
1. [Step with specific technical action]
2. [Step with specific technical action]
...

### Matching Keys Action Plan
| Key | Current Status | Action Required |
|-----|---------------|-----------------|
| em (email) | Sending | None |
| ph (phone) | Not sending | Add to CAPI payload |
...

### Validation Checklist
- [ ] Pixel fires on all conversion pages
- [ ] CAPI fires for all conversion events
- [ ] event_id matches between Pixel and CAPI
- [ ] Deduplication confirmed in Test Events
- [ ] EMQ score ≥ 6.0 (target 8.0+)
- [ ] All available matching keys sending
- [ ] SHA-256 hashing verified (no raw PII)
- [ ] action_source set correctly per event type
- [ ] CRM events firing within 7-day window
- [ ] fbclid capture and storage working

### Expected Impact
- Estimated signal recovery: [X]% of currently lost events
- EMQ improvement: [current] → [projected]
- Timeline to stable data: 7-14 days post-implementation

Example

Input: "We're running Meta ads for a B2B SaaS company. We have the Pixel installed but no CAPI. Our EMQ is 4.2 and we track leads via a CRM. Some leads close 2-3 weeks after the ad click. How do we fix this?"

Output:

Meta Tracking Implementation Plan

Current State Assessment

  • Pixel status: Active
  • CAPI status: Not configured — this is your biggest gap
  • EMQ score: 4.2 / 10 — below minimum threshold of 6.0
  • Event deduplication: Not applicable (no CAPI yet)
  • Events tracked: PageView, Lead (Pixel only)
  • Missing matching keys: phone, fbc, fbp, IP, user agent

Implementation Steps

  1. Implement CAPI via server-side GTM — Best option for a custom B2B stack with your CRM. Set up server-side GTM container on Google Cloud ($10-$50/mo). Configure the Meta CAPI tag in server GTM to mirror Pixel events.
  2. Add event deduplication — Generate a unique event_id for each Lead event. Pass it to both the Pixel fbq('track', 'Lead', {}, {eventID: 'lead_xyz'}) and the server-side CAPI tag.
  3. Capture and store fbclid — Add JavaScript to capture fbclid from landing page URLs. Store it in a hidden form field so your CRM records it on the contact. Also capture the _fbp cookie value.
  4. Build CRM-to-CAPI pipeline — When a CRM deal closes, fire a CAPI Purchase event (or custom ClosedWon event) with action_source: system_generated, the stored fbc value, hashed email, and hashed phone.
  5. Fix the 2-3 week delay problem — Events must arrive within 7 days of occurring. Set up a CRM workflow to fire CAPI events at key pipeline stages (MQL, SQL, Opportunity) rather than waiting for close. This keeps events within the 7-day window AND gives Meta more mid-funnel signals.
  6. Add matching keys — Send hashed email + hashed phone + fbc + fbp + IP + user agent with every CAPI event.

Matching Keys Action Plan

Key Current Status Action Required
em (email) Not sending via CAPI Add to CAPI payload, SHA-256 hashed
ph (phone) Not sending Add to CAPI payload, SHA-256 hashed
fbc Not capturing Add fbclid capture to landing pages + CRM
fbp Not sending Read _fbp cookie, pass to server GTM
IP Not sending Auto-captured by server-side GTM
User agent Not sending Auto-captured by server-side GTM

Validation Checklist

  • [ ] Pixel fires Lead event on form submission
  • [ ] CAPI fires Lead event via server GTM
  • [ ] event_id matches between Pixel and CAPI for Lead events
  • [ ] Deduplication confirmed in Test Events
  • [ ] EMQ score ≥ 6.0 (target 8.0+)
  • [ ] fbclid captured and stored on the contact in your CRM records
  • [ ] CRM events (MQL, SQL, ClosedWon) firing via CAPI within 7 days
  • [ ] SHA-256 hashing verified for all PII fields
  • [ ] action_source = "website" for web events, "system_generated" for CRM events

Expected Impact

  • Signal recovery: ~20-30% more conversion events captured
  • EMQ improvement: 4.2 → 7.0+ (with email + phone + fbc + fbp)
  • More mid-funnel signals for Meta's algorithm to optimize against
  • Timeline: 7-14 days post-implementation for algorithm to recalibrate

Guidelines

  • Never skip CAPI. Pixel alone is insufficient in 2025. Browser restrictions, ad blockers, and iOS ATT mean Pixel misses 20-30% of events.
  • Never send unhashed PII. All customer data must be SHA-256 hashed before transmission. Sending raw emails or phone numbers violates Meta's terms and privacy regulations.
  • Don't ignore EMQ below 6.0. Sub-6 EMQ directly degrades ad delivery quality. Treat it as a critical issue.
  • Don't delay CRM events beyond 7 days. Events older than 7 days are rejected. Build pipeline stages that fire within the window.
  • Don't set up CAPI without deduplication. Double-counted conversions inflate metrics and mislead the algorithm.
  • Offline Conversions API is dead. If someone references it, redirect them to standard CAPI with action_source: physical_store or system_generated. Discontinued May 2025.
  • Instant Experience (June 2025): Meta no longer counts Instant Experience as a landing page view. Account for this in funnel metrics if using Instant Experience ads.
  • Attribution API change (October 2025): 7d_view and 28d_view attribution windows are no longer returned in the Ads Insights API. Update any reporting that pulls these windows.
  • iOS 14+ context: Aggregated Event Measurement (AEM) still governs measurement for iOS users. You can configure up to 8 prioritized events per domain.
  • Cross-reference: For Meta account auditing including tracking health checks, use the meta-ads-audit skill. For Google tracking (gtag, GA4, enhanced conversions), tracking setup is covered within the google-ads-audit skill.
  • Confidence: If the user's platform or tech stack isn't listed in this skill, recommend they consult Meta's official CAPI documentation for platform-specific integration guides rather than guessing at implementation details.

04 同仓库更多 Skills

narayan-metaflow
narayan-metaflow/metaflow-marketing-skills

technical-seo-ai-crawler-audit

Diagnose and prioritize the technical issues blocking a site from being crawled, indexed, ranked, and cited. Modern technical SEO has two audiences: Google's crawler/indexer and AI answer engines (GPTBot, PerplexityBot, ClaudeBot, etc.). Both need HTML-visible content and a clean crawl path.

市场营销营销推广
★ 19
narayan-metaflow
narayan-metaflow/metaflow-marketing-skills

seo-reporting-measurement

Build reports that stakeholders trust, rooted in data sources that are actually reliable. Modern SEO measurement has to reckon with a very specific constraint: **the primary data source (GSC) is ~75% incomplete**, and the AI visibility layer is **probabilistic**, not deterministic. Building reports that ignore these realities produces false confidence.

数据分析师市场营销营销推广
★ 19
narayan-metaflow
narayan-metaflow/metaflow-marketing-skills

seo-max-kt

Long-form SEO and AEO methodology bundle; read SEO_AND_AEO_KT.md for the full corpus.

市场营销营销推广
★ 19
narayan-metaflow
narayan-metaflow/metaflow-marketing-skills

seo-keyword-research-intent-mapping

Build a keyword universe that reflects how both humans *and* AI answer engines search. Traditional keyword research looks at standalone terms; modern research also maps the **fan-out sub-queries** AI tools generate when breaking down a user prompt.

市场营销营销推广资料检索与研究
★ 19
narayan-metaflow
narayan-metaflow/metaflow-marketing-skills

seo-aeo-content-strategy

Design content systems that earn both Google rankings *and* AI citations. The core insight: AI answer engines read content differently than humans — they extract the highest-information-gain sentences and treat H2 headings as prompts. The "ski ramp" structure optimizes for both.

市场营销内容创作者视觉设计营销推广
★ 19
narayan-metaflow
narayan-metaflow/metaflow-marketing-skills

prompt-picker

Build a prompt library that reveals how a brand shows up across the full buyer journey in AI search systems. The library is designed for actual LLM visibility tracking, not theoretical SEO keyword research.

Skill与智能体开发
★ 19