agentleFS
minhnv0807minhnv0807/ai-business-skills/skills/en/53-tracking-setup-global/SKILL.md

53-tracking-setup-global

Skill644 starsChanged 6 months agoMITView on GitHub

Use when conversion TRACKING must be set up or fixed before spending — Meta Pixel and Conversions API, Google Ads conversions with Enhanced Conversions, GA4, TikTok Pixel and Events API, server-side GTM, consent mode for GDPR and CCPA, UTM conventions, iOS ATT attribution windows, and a pre-launch verification checklist. Trigger on 'tracking setup', 'pixel setup', 'CAPI', 'GA4 events', 'conversions are not being recorded', 'Meta and Shopify numbers do not match'. Also use when launch is imminent and nothing has been verified. Not for — auditing a live account broadly, see `21-ads-audit-global`; analyzing data once it flows, see `13-data-analysis-global`; the campaign hierarchy, see `52-account-structure-global`.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 7 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈7

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Ask up to 4 questions:

    line 38
  2. Distinctive:

    Measure to revenue, not to lead: If 1,000 leads produce 10 sales, you must be able to tell whether the failure is in ads or in sales.

    line 52
  3. Distinctive:

    Keep a consent log — which user, which choice, when, which policy version.

    line 77
  4. Distinctive:

    Only one primary conversion action per objective should be enabled for bidding.

    line 104
  5. Distinctive:

    Verify the domain and configure Aggregated Event Measurement with a prioritized list of up to 8 web events per domain.

    line 111
  6. Distinctive:

    All must pass. One missing item means do not launch:

    line 156
  7. Distinctive:

    When ads manager, GA4, and backend numbers disagree, classify the gap before changing budgets or tags.

    line 179

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Tracking Setup (Global)

Operating rules: skills/en/references/operating-rules-global.md (esp. rules 2, 3, 7).

Ads without verified tracking is driving blind. Hard rule: do not launch ads until tracking verifies green. Every dollar spent before verification is unmeasurable. Set it up correctly the first time — changing events mid-campaign destroys historical comparability.

Full per-platform install steps and the consent mode parameter reference: read references/platform-tracking-setup.md.

Information gathering

Ask up to 4 questions:

  1. Which ad platforms and objective? Meta / Google / TikTok / LinkedIn / YouTube — lead (form or message) / purchase / traffic?
  2. What runs the website or landing page? Shopify / WordPress / Webflow / Framer / custom — and is there access to edit the head tag or install GTM?
  3. Which markets receive traffic? This decides the consent requirements: EU/UK means GDPR and ePrivacy, California means CCPA/CPRA, other US states have their own opt-out laws.
  4. Where does the sale actually close? On site (pixel-measurable) or offline via sales calls, demos, or DMs (needs CRM/offline conversion import)?

Principles

  1. No green, no launch. Verification comes before spend, every time.
  2. Consent is part of tracking, not a legal afterthought. In the EU/UK, tags must not fire before consent. Getting this wrong risks fines and, on Google, loss of remarketing audiences.
  3. One installation method only. GTM plus hardcoded tags equals duplicate events and doubled phantom conversions.
  4. Client-side plus server-side, deduplicated. Browser-only tracking loses a large share of events to ad blockers, ITP, and iOS ATT. Send the same event from both sides with a shared event ID.
  5. Consistent UTMs. Define the convention once, before anyone builds a link. GA4 cannot repair inconsistent casing or naming after the fact.
  6. Measure to revenue, not to lead. If 1,000 leads produce 10 sales, you must be able to tell whether the failure is in ads or in sales. That requires CRM or offline conversion import.
  7. Expect platform numbers to disagree. Ads managers use different attribution windows and modeled conversions. Pick one source of truth for decisions and use the others as directional.

Workflow

1. Choose the measurement architecture

LayerPurposeMinimum
Client-side pixel/tagFast setup, on-page eventsMeta Pixel, Google tag, TikTok Pixel
Server-side APIRecovers events lost to blockers and ITPMeta Conversions API, TikTok Events API, Google Enhanced Conversions
AnalyticsCross-channel truth, funnel analysisGA4 with key events configured
Consent layerLegal gate in front of everything aboveCMP plus Google consent mode v2
CRM / offlineConnects leads to actual revenueCRM with source fields, offline conversion import

Server-side GTM is the recommended container for anything beyond a simple store: it centralizes event forwarding, improves data quality, and keeps first-party cookie lifetimes usable under ITP.

2. Consent management (do this before any tag fires)

This is a first-class requirement, not an optional extra.

EU / UK — GDPR and ePrivacy:

  • Install a certified consent management platform. Meta requires a CMP for EU traffic; Google requires a Google-certified CMP for EEA/UK traffic to keep personalized advertising and audience features.
  • Implement Google consent mode v2. The four signals are ad_storage, analytics_storage, ad_user_data, ad_personalization. Default them to denied for EEA/UK before the banner is answered, then update on consent.
  • Non-essential tags must not fire before consent is granted. Consent must be as easy to refuse as to accept, and withdrawable.
  • Keep a consent log — which user, which choice, when, which policy version.

California — CCPA/CPRA (and comparable US state laws):

  • Provide a visible opt-out of sale/sharing of personal information.
  • Honor the Global Privacy Control browser signal as a valid opt-out.
  • On Meta, apply Limited Data Use for California traffic.

Email and SMS: CAN-SPAM in the US and GDPR in the EU require documented opt-in for marketing sends and a working unsubscribe. Any list uploaded as a Custom Audience or Customer Match list must have that consent on record — see 14-email-marketing-global.

Record the decision in the output file: which CMP, which regions gated, who verified.

Vietnam traffic: personal-data duties for Vietnamese visitors and customers are not covered here. See the VN legal hub skills/vi/references/phap-ly-marketing-vn.md section 2 and verify the source instruments on vbpl.vn / thuvienphapluat.vn.

Any change to live tags, pixel settings, consent configuration, or an audience upload is a write action: draft it as a change block (current / proposed / rollback / named approver), apply only after approval, read back, and log it (operating rule 7).

3. Install per platform

Full step-by-step checklists live in references/platform-tracking-setup.md. Summary of what must exist:

PlatformClient-sideServer-sideVerification tool
MetaPixel with standard events (PageView, ViewContent, AddToCart, Lead, Purchase)Conversions API with event_id deduplication, Event Match Quality >= 6.0Meta Pixel Helper, Test Events
Google AdsGoogle tag / GTM conversion actionsEnhanced Conversions (hashed email/phone), consent mode v2Tag Assistant, conversion diagnostics
GA4GA4 tag, key events marked as conversionsMeasurement Protocol or server-side GTMGA4 DebugView, Realtime
TikTokTikTok Pixel (ViewContent, AddToCart, SubmitForm, CompletePayment)Events API with deduplicationTikTok Pixel Helper, Events diagnostics
LinkedInInsight Tag with conversion rulesConversions API for CRM-side eventsLinkedIn Insight Tag Helper

Only one primary conversion action per objective should be enabled for bidding. Extra enabled conversions inflate counts and confuse smart bidding.

4. iOS ATT and attribution windows

Since iOS 14.5, a large share of iOS users opt out of app tracking, so platform-reported conversions are partly modeled rather than observed.

  • Meta's post-ATT default is a 7-day click / 1-day view attribution window. Compare periods on the same window or the trend is meaningless.
  • Verify the domain and configure Aggregated Event Measurement with a prioritized list of up to 8 web events per domain. The event you optimize for must be ranked highest.
  • Expect ads manager conversions and GA4 last-click conversions to differ, often materially. Decide up front which one drives budget decisions.
  • Server-side events plus Enhanced Conversions and good Event Match Quality recover a meaningful share of the loss. This is the main reason server-side is no longer optional.
  • Third-party cookie deprecation compounds this. Prioritize first-party data collection — email capture, account creation, CRM — over pixel-only audience building.

5. UTM convention

https://domain.com/lp?utm_source=[source]&utm_medium=[medium]&utm_campaign=[campaign]&utm_content=[ad_name]
ParameterConventionExample
utm_sourcemeta / google / tiktok / linkedin / klaviyometa
utm_mediumpaid_social / paid_search / display / email / organicpaid_social
utm_campaign[product]_[audience]_[market]_[MMYY]course-a_cold_us_0726
utm_content[creative-type]_[hook-type]_[version] — matches the ad namevid_pain_v1

Lowercase, no spaces, no special characters. utm_content must equal the ad name defined in 52-account-structure-global so ads manager and GA4 reconcile in both directions. Add utm_term only for search keywords.

6. GA4 event mapping

GA4 eventTriggerPlatform equivalent
page_viewAll pagesPageView (Meta), ViewContent (TikTok)
view_itemProduct or offer pageViewContent (Meta)
add_to_cartCart addAddToCart (Meta/TikTok)
begin_checkoutCheckout startedInitiateCheckout (Meta)
generate_leadForm submit or qualified messageLead (Meta), SubmitForm (TikTok), Conversion (Google)
purchasePayment confirmed, with value and currencyPurchase (Meta/TikTok/Google)

Each action fires exactly one event. Mark the event that matches the ads objective as a key event in GA4. Always send value and currency on purchase — without them ROAS cannot be computed and multi-market revenue cannot be normalized.

7. CRM and offline conversions

Required whenever the sale closes off-site (B2B, high-ticket, coaching, agency, phone or DM sales):

  1. Capture source on every lead. CRM fields: source, medium, campaign, ad content, GCLID or fbclid, timestamp, owner, stage, disqualification reason, closed revenue.
  2. Import offline conversions weekly. Meta offline/CAPI conversions and Google Ads offline conversion import via GCLID or Enhanced Conversions for Leads. This teaches bidding from real revenue instead of raw lead count.
  3. Reconcile monthly. Compare CRM lead count against ads manager lead count. A large gap (20% is a heuristic starting band, tune it per account) signals a tracking or attribution problem, not a sales problem.

Uploaded CRM data must be hashed and covered by consent.

8. Verify green before launch

All must pass. One missing item means do not launch:

  • Pixel/tag fires on landing page and confirmation page
  • Correct conversion event type selected — not PageView
  • No duplicate events (one action, one event in the events manager)
  • Server-side API live, deduplication working, Meta EMQ >= 6.0
  • Google Enhanced Conversions enabled and receiving hashed data
  • Consent mode v2 implemented; tags blocked before consent in EEA/UK; Limited Data Use applied for California
  • Test event: complete one conversion and confirm it is recorded (preliminary only, see Discrepancy triage)
  • End-to-end proof: one real order or lead traced from click to backend/CRM by its ID (order_id, lead_id, event_id)
  • Domain verified and Aggregated Event Measurement events prioritized
  • UTMs on all ad links and matching the convention
  • GA4 receiving paid sessions (confirm in Realtime)
  • CRM source capture live if the sale closes off-site

9. Ongoing checks

Daily (2 minutes): events flowing in the last 24 hours on every platform; Google Ads conversions recording; GA4 showing paid sessions.

Weekly: server-side match rate healthy against your own baseline (house heuristic: 70% as a starting band, verify against the platform's current docs); no duplicate events; lead counts across ads manager, GA4, and CRM within your tuned gap band (house heuristic: start near 20%); every live link carries a correct UTM; consent banner still firing correctly after any site release.

Discrepancy triage

When ads manager, GA4, and backend numbers disagree, classify the gap before changing budgets or tags.

  1. Source of truth first. Orders = backend order system; leads = CRM. Platform and GA4 figures are directional.
  2. Compare like with like. Same date range, timezone, currency, and attribution window.
  3. Expected causes (not defects): attribution window, click vs view-through, modeled conversions, timezone/currency, session vs user counting, ad blockers and consent declines, cross-device.
  4. Double-sender check: pixel plus server API without a shared event_id double-counts; a GTM tag plus a hardcoded snippet duplicates; a platform plugin plus GTM duplicates. Keep one installation method.
  5. Thank-you-page guard: fire the purchase or lead event once per order_id. Reloads, the back button, and bookmarked confirmation URLs all re-fire it, so dedupe server-side as well.
  6. Test Events are not proof. A green Test Events panel only shows one test event arrived. Proof is one real order or lead traced end to end by its ID.
  7. Label each finding observed / reported / inferred (operating rule 3). Missing data is unknown, not fail.
  8. No universal percentage thresholds. Any "acceptable gap" band (for example 20%) is a heuristic to tune per account; record it as such.

Verdict per gap: expected / investigate / fix. The Vietnamese pack has a detailed reference and a reconciliation-report template (skills/vi/53-tracking-setup/references/chan-doan-lech-so-lieu.md); reuse its structure.

Troubleshooting

ProblemCommon causeFix
Pixel does not fireTag missing on the page or blocked by consent gateCheck GTM preview and the consent state; confirm the tag is allowed after consent
Duplicate eventsGTM and hardcoded tag both installedRemove one; verify event_id deduplication
Low match rate / low EMQNo hashed email or phone sent server-sideAdd customer parameters to the server payload
Conversion not recordedWrong event name or wrong trigger pageCheck spelling and that the trigger is the confirmation page
GA4 far below ads managerDifferent attribution windows plus modeled conversionsCompare on the same window; treat one source as decision-grade
Remarketing audiences shrinking in the EUConsent mode v2 missing or misconfiguredImplement all four consent signals and a certified CMP
Leads arrive with no sourceNo CRM source captureAdd hidden UTM/click-ID fields to every form

Output structure

File name: tracking-setup-[product]-[YYYYMMDD].md

# Tracking Setup — [Product/Campaign]
Platforms: [list] · Site platform: [x] · Markets: [x] · Objective: [Lead/Purchase]

## 1. Measurement architecture   [client, server, analytics, consent, CRM]
## 2. Consent plan               [CMP, regions gated, consent mode v2, LDU, log owner]
## 3. Platform setup status      [table with status per item]
## 4. Attribution settings       [window per platform, AEM priority list]
## 5. UTM convention             [parameters and real project examples]
## 6. Event mapping              [GA4 to platform events]
## 7. CRM / offline conversions  [fields captured, import cadence]
## 8. Pre-launch verification    [green checklist, verified by, date]
## 9. Ongoing check schedule     [who, when, alert thresholds]

Related skills

  • 52-account-structure-global: build campaigns only after tracking is green; utm_content must match ad names.
  • 54-media-plan-global: the objective in the plan determines which events must be tracked.
  • 21-ads-audit-global: periodic tracking audit (match rate, duplicates, EMQ).
  • 12-landing-page-brief-global: the landing page must expose the events and form fields this setup needs.
  • 13-data-analysis-global: reads the clean data this system produces.

Quality checklist

  • Setup checklist complete for every platform in the plan
  • Consent plan explicit: CMP, consent mode v2 signals, regions, California opt-out, consent log
  • Server-side layer live with deduplication and a documented match rate
  • Attribution window stated per platform, with an agreed decision-grade source
  • UTM convention documented with real examples and matching ad names
  • Event mapping complete from page_view to purchase, with value and currency
  • Offline or CRM conversion path exists if the sale closes off-site
  • Green verification checklist completed, dated, and signed off by a named person
  • Daily and weekly checks scheduled with alert thresholds (house heuristics to tune per account: match rate starting band 70%, source gap starting band 20%)
  • The hard rule is restated in the output: not green means do not launch

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.