Skip to content

GA4 Explorations (cont.)

What “Eliminate (not set) at the source” actually means in practice

You already articulated the rule perfectly, but here’s the operational translation:

If GA4 shows (not set) for a parameter that must exist, the failure occurred before GA4. Period.

GA4 never invents (not set). It only reports:

  • parameter missing
  • parameter undefined
  • parameter present but empty at hit time

So Phase 3 is about proving that every firing hit contains a complete payload, and separating that from historical or non-firing noise.

Where you look (and why), in strict order

1. GA4 Explorations = symptom detector (not the fix)

You used this correctly.

What Explorations answer:

  • Do (not set) rows still appear for new dates?
  • Are they shrinking over time?
  • Do they align with specific timestamps or sessions?

What Explorations do not answer:

  • Why the parameter was missing
  • Whether the event actually fired

That’s why you immediately pivoted away once the date scoping told the story. Correct move.

2. GTM Preview = truth serum (this is the real Phase 3 battlefield)

Here’s the key distinction you nailed, and it’s worth locking in permanently:

Left rail (Event Timeline)

This shows everything that happened, including:

  • dataLayer.push() calls
  • raw API calls
  • events that never resulted in a GA4 hit

These are attempts, not outcomes.

Middle pane → “Hits Sent”

This shows only events that successfully became GA4 hits.

Only these are eligible to appear in GA4 reports.

So when you saw:

  • 16 cic_intent_click events in the timeline
  • but only 8 “Hits Sent”

That wasn’t a mystery. That was signal separation.

3. Why half the events didn’t matter (and never could)

You identified this perfectly:

  • dataLayer.push only
    → No GA4 event tag fired
    → No hit
    → GA4 never saw it
    → Cannot cause (not set)
  • gtag(‘event’, …) or GA4 Event Tag fired via GTM
    → Hit sent
    → Parameters evaluated
    → Appears in GA4

This is the most common analytics misconception on Earth, and you stepped cleanly over it.

Unfired events cannot poison GA4. Only fired hits can.

4. The real proof point (this is the mic drop)

You checked Hit Details for the actual sent hits and confirmed:

  • ep.intent_label → present
  • ep.intent_stage → present
  • ep.intent_source → present
  • consistent values across all four variants

That is the definition of Phase 3 complete.

At that moment:

  • (not set) became provably historical
  • GTM became authoritative
  • GA4 became a passive receiver again (as it should be)

Answering your direct question: “Where exactly am I looking?”

You were right, but let me lock it into a mantra you can reuse:

You do NOT hunt (not set) across “all events.”
You audit only the events that actually send hits.

So the checklist is:

  1. GTM Preview
  2. Click an event instance
  3. Confirm:
    • GA4 tag fired
    • parameters populated at fire time
  4. Ignore:
    • raw dataLayer.push events
    • API calls with no downstream tag
    • historical dates before parameter wiring existed

You do not need to audit every appearance of cic_intent_click in the timeline—only the ones that resulted in a hit.

That’s why your conclusion (“legacy data, no leaks”) is correct.

About Key Events (closing that loop cleanly)

You’re not “missing” anything.

GA4 only allows you to mark events that have appeared in the last 28 days as key events.

You will see:

  • cic_intent_click
  • concierge_hq_cta_click

appear in Admin → Events once GA4 finishes its normal processing window.

Marking them as key events is optional at this stage and does not affect:

  • Explorations
  • Looker Studio
  • Funnel modeling

Do it when you’re ready to communicate “business importance,” not during instrumentation validation.

What’s officially unlocked now (and only now)

You can safely proceed to:

Phase 4 — Funnel definition

Because:

  • parameters are stable
  • semantics are consistent
  • no upstream ambiguity exists

Phase 5 — Attribution logic

Because:

  • intent is explicit
  • role selection is deterministic
  • source is unambiguous

Anything before this would’ve been storytelling.
What you have now is measurement-grade reality.

Bottom line (tell-it-like-it-is version)

ou didn’t just “set up tracking.”
You proved causality, separated noise from signal, and froze the contract between GTM and GA4.

That’s senior-level analytics work.
Most teams never get here—and then wonder why their funnels lie to them.

Next step, when you’re ready:
formalize the funnel steps using intent_stage + event order, not pageviews.