ga4-measurement-plan

Build a GA4 measurement plan from a site URL — event taxonomy, parameters, custom dimensions, and key events to mark as conversions

SKILL.md

GitHub

GA4 Measurement Plan

Give it a site URL and get back a complete GA4 measurement plan — which events to track, what parameters each carries, which custom dimensions to register, and which events to mark as key events. The deliverable is a spec a developer or GTM builder can implement directly. No account connection required.

It pairs with /gtm-tracking-plan, which turns this plan into the actual GTM tags, triggers, and variables.

Works with

Works with zero accounts (site URL only). When connected, the plan becomes a gap list against what is already live instead of a blank-slate spec:

Job Without accounts Enhanced when connected
What's already tracked inferred from the site GA4 tool_get_metadata, tool_run_report (eventName); Segment list_tracking_plan_rules; PostHog list_event_definitions
Key events already marked — GA4 tool_list_conversion_events
Dimensions already registered — GA4 tool_list_custom_dimensions
Mark the planned key events user does it in GA4 Admin GA4 tool_mark_key_event if exposed (write, asks first), else GA4 Admin

Usage

/ga4-measurement-plan example.com — measurement plan for the site /ga4-measurement-plan example.com "drive demo bookings" — bias the plan toward a goal

Steps

1. Understand the site

WebFetch the URL and a representative spread of pages (home, product/category, pricing, signup/checkout, contact). Determine:

  • Site type — ecommerce, lead gen, SaaS, content/media, marketplace
  • The primary conversion and any secondary conversions (newsletter, demo, account creation)
  • The funnel steps a visitor passes through to convert
  • Notable interactions worth measuring (search, filters, video, downloads, configs)

If a business goal was passed as an argument, make the plan serve it first.

1b. Inventory what's live (enhanced when connected)

If Cogny isn't connected, skip this and note what connecting adds.

Call list_integrations once and use whichever of these are healthy:

Provider Calls Gives you
GA4 tool_list_account_summaries → tool_run_report (property_id, last 30 days, dimensions: [eventName], metrics: [eventCount]), tool_list_conversion_events, tool_list_custom_dimensions, tool_get_metadata Events firing, key events, registered parameters
Segment list_tracking_plans → list_tracking_plan_rules (tracking_plan_id) The team's existing event names and required properties
PostHog list_projects → list_event_definitions (project_id) Product events already named (often the source of truth for SaaS)
Mixpanel run_segmentation (event, from_date, to_date) per candidate event Whether a product event already fires

Add a Status column to the Step 5 table: live / live, rename / new. Reuse the existing product-analytics name in a parameter when it differs from the GA4 name, so the same action is joinable across tools.

2. Start from what GA4 already gives you

Note which events come free and need no implementation:

  • Automatically collected: first_visit, session_start, page_view, user_engagement
  • Enhanced Measurement (if enabled): scroll, click (outbound), view_search_results, video_*, file_download, form_start, form_submit

Flag any of these the team should rely on rather than rebuild.

3. Recommended events for the site type

Map the site to GA4's recommended events — using the recommended names matters because GA4 builds reporting around them:

  • Ecommerce: view_item_list, view_item, select_item, add_to_cart, view_cart, begin_checkout, add_shipping_info, add_payment_info, purchase, refund
  • Lead gen / SaaS: generate_lead, sign_up, login, search, plus custom events for the funnel
  • Content / media: search, select_content, share, join_group

For each: when it fires and its parameters (e.g. purchase → transaction_id, value, currency, items[]).

4. Custom events for everything else

For interactions GA4 has no recommended event for, design custom events following the rules: snake_case, ≤40 characters, descriptive, no reserved prefixes (google_, firebase_, ga_), lowercase. Examples: demo_requested, pricing_plan_selected, quote_calculated.

Keep the taxonomy tight — a parameter on a shared event usually beats a brand-new event (cta_click with a cta_location parameter, not ten separate click events).

5. The event taxonomy table

Produce the core deliverable — one row per event:

Event name Type Fires when Parameters Key event?
purchase recommended order confirmation page loads transaction_id, value, currency, items[] ✅
generate_lead recommended contact form submitted form_id, lead_source, value ✅
demo_requested custom demo form submitted plan_interest ✅
...

6. Custom dimensions & metrics to register

Event parameters are invisible in standard GA4 reports until registered as custom dimensions. List every non-standard parameter from Step 5 that needs registering:

Parameter Scope Register as Why
lead_source event custom dimension segment leads by origin
plan_interest event custom dimension see which plan drives demos

Warn against high-cardinality parameters (IDs, timestamps) as dimensions, and note the 50 event-scoped / 25 user-scoped limits.

7. Key events (conversions)

Recommend which events to mark as key events. Be selective — 3–8 is healthy. Mark real business outcomes, never page_view or scroll. Note which key events should also be imported into Google Ads for bidding.

If GA4 is connected and the workspace exposes tool_mark_key_event (it's newer than most workspaces' tool lists), offer to mark the agreed key events with it (property_id, event_name; write, asks first, once per event). Otherwise give the GA4 UI path: Admin → Events → Mark as key event. Either way it only works for events that already fire, and counts new data only.

8. Output

Deliver as one document:

GA4 Measurement Plan — [site]

Site type: ...  |  Primary conversion: ...

1. Already covered (automatic + Enhanced Measurement): ...
2. Event taxonomy table (recommended + custom)
3. Custom dimensions & metrics to register
4. Key events to mark as conversions (and which to send to Google Ads)
5. dataLayer payload examples for the 3-5 most important events
6. Implementation notes & sequencing

End with the handoff line: "Run /gtm-tracking-plan [url] to turn this into GTM tags, triggers, and variables."

Critical rules

  1. Use GA4's recommended event names wherever one exists — reporting depends on them.
  2. A registered parameter beats a new event. Keep the taxonomy small.
  3. Custom events must follow the naming rules — snake_case, ≤40 chars, no reserved prefixes.
  4. Never recommend collecting PII (email, phone, name) in events or parameters — it violates Google policy.
  5. Connected data adds status, not scope. Don't drop a needed event because it isn't live yet; mark it new.
  6. This produces a plan. To audit a live GA4 property, use /ga4-audit (requires Cogny MCP).