Skip to content
All field notes
Measurement
Published
Oct 30, 2026
Updated
Oct 31, 2026
Reading time
11 min read

GA4 Funnel Counts Differ? Audit Entry and Sequence Rules

Diagnose GA4 funnel counts with entry, sequence, time and user-scope checks. Measure campaign-page journeys without confusing events or attribution.

AP
AdProof Research
Independent measurement desk
A visitor follows an ordered journey with a browsing detour while another enters at the final step, beneath separate entry-gate and route controls.
Funnel entry and allowed intervening actions answer different questions; changing either changes the population being measured. Conceptual illustration without property statistics.
In this guide (10 sections)

A closed funnel does not mean same session or next click. Closed controls entry at the first step. Direct versus indirect sequence controls whether other actions can occur between steps. A time limit is another condition. Write those rules separately before interpreting abandonment or changing the funnel to make its numbers look better.

For “people who visited the campaign page and later booked,” retain a first-step requirement for that page and permit the intervening behavior the question allows. Opening the funnel can admit people who never met that entry requirement. That changes the population rather than repairing its measurement.

Two real questions expose the same ambiguity#

In a public appointment-measurement question, an analyst wants to know how many people booked after seeing a campaign page, including those who browsed elsewhere first. Replies mix useful sequence advice with the incorrect assumption that a closed funnel inherently requires the same session.

An independent question about funnel results changing with the date range reports fewer funnel users over a longer interval despite an unsampled result. Its replies suggest retention, changed page definitions and broad claims about modeled data. Those are hypotheses, not a verified diagnosis of that property.

Both are actual reader problems. The second is older evergreen; neither discussion establishes a new GA4 outage. The useful task is to make the counted journey explicit and test which rule changes the result.

Related guide

Make the funnel rules explicit

Record entry, sequence and time conditions before comparing funnel counts.

Use the tracking QA checklist

Opens a practical guide. No email required.

1. Turn the business question into a funnel contract#

Start with one sentence: “Within this reporting interval, how many observed users visited this campaign page and then completed a confirmed appointment within the chosen interval?” Agree what each term means with the analytics and booking owners.

Contract fieldExample decision to record
EntryA qualifying page view of the campaign page
OutcomeConfirmed appointment, not opening the booking widget
OrderingEntry must precede the outcome
Intervening behaviorOther pages and events allowed, or genuinely immediate next action
Elapsed timeA documented business-appropriate limit
Reporting populationProperty, identity context, dates, timezone and permitted collection
Counting unitFunnel users, distinct appointments or recorded event occurrences

A campaign page is not necessarily a landing page. The person may discover it after arriving elsewhere. Decide whether the question requires an initial-session landing or any qualifying page visit; those are different entry conditions.

Google's lead-generation funnel example combines a page_view with a page_location condition, then adds form-related steps. Use the pattern to inspect the actual collected fields. Do not copy its home-page or form-name values into an unrelated booking implementation.

Keep the page condition at the intended step. Review whether a global filter inadvertently excludes events required by later steps. Save the original filter and segment definitions so a changed population does not masquerade as improved conversion.

2. Separate entry, sequence and time#

Google's Funnel exploration reference distinguishes open entry at any step from closed entry at step one. It separately defines immediately following versus indirectly following steps, and a Within constraint. A sequence with a browsing detour can therefore be a valid closed-funnel journey when its conditions allow it.

“Directly followed by” concerns the recorded sequence, not just the next visible page. Inspect automatically collected and custom intervening events before declaring the route immediate. A person can appear to click the booking button next while the recorded event stream contains other activity.

The API step reference separately represents immediate following and duration from the prior step. Choose a time condition based on the question, then document it. A short interval can exclude genuine delayed bookings; a longer one answers a broader journey question. Neither choice establishes a session boundary unless your conditions actually express that boundary.

For the appointment question, keep two outputs if both decisions matter: an immediate-path diagnostic and an allowed-browsing journey. Label their rules rather than presenting whichever has the larger completion percentage as the definitive answer.

3. Do not compare funnel users with raw event totals#

Google's Data API funnel overview returns user-based step counts along with completion and abandonment metrics. Its funnel request also has a defined step/filter specification. A chart is not simply a running sum of every event with a matching name. For a product-level question, compare item identities and item quantities across the shopping events before treating aggregate product metrics as a user-based funnel.

Separately count the booking records that the business needs. One observed user can generate multiple booking-event occurrences, whether through real repeat actions or a collection issue. Investigate that distinction with the event owner; the funnel alone does not deduplicate appointment records or prove that an event represents a confirmed business outcome.

Also distinguish a segment from the funnel's steps. Google's segment builder reference describes user, session and event segments. A user segment can include other activity from qualifying users, while a session segment or event segment selects different data. Selecting “users who booked” does not, by itself, establish that the campaign-page visit preceded the booking.

Write the sequence as a sequence. Keep the appropriate entry requirement, order and time constraint visible in the saved analysis rather than trying to infer them from overlapping aggregate groups.

Worked example: the same recorded journeys, different populations#

This is hypothetical. Assume six observed user identities eligible for this report, complete event visibility for this exercise, and an interval long enough for every listed journey. A is the campaign-page step, B the booking-event step and X an unrelated intervening action. The A-to-B rule allows intervening actions.

Observed userRecorded journeyClosed funnelOpen funnel
OneA → BA and BA and B
TwoA → X → BA and BA and B
ThreeBNeither stepB
FourB → A → BA and BB
FiveAAA
SixXNeither stepNeither step

The closed funnel has four users at A and three at B: 3 ÷ 4 = 75% for the specified A-to-B journey. The open funnel has three at A and four at B because later-step entrants are included. Dividing all four B users by the three A users would mix populations; it is not a campaign-page completion rate.

There are five recorded B occurrences across four observed users. That event count, the four users with B and the three closed-funnel completers answer different questions. None is automatically the number of unique confirmed appointments.

Why does user Four differ? In an open funnel, the first qualifying action can place the user at B; returning to A does not re-enter them at that earlier step. Closed entry instead looks for step A. Google documents this first-entry behavior; use a controlled journey to verify how it applies to your exact settings.

Now make a separate copy with directly-followed steps. User Two has X between A and B, so that recorded path fails the immediate A-to-B condition. The original indirect result has not become wrong: the copy asks a different question.

4. Explain date and breakdown shifts before calling them lost users#

Google describes one funnel entry per user in the selected interval and first-occurrence attribution for breakdowns and trended step dates. Widening dates can therefore change where an observed user's qualifying journey starts or which displayed date receives their step.

Consider a hypothetical open two-step funnel. A user records B in January, then A and B in March. A March-only interval can start them at A. Expanding the interval to include January can make the earlier B their first qualifying entry, so the same user no longer contributes to step A in that open-funnel result. This illustrates a possible mechanism, not the established cause of the forum poster's discrepancy.

Run a controlled comparison with unchanged settings and a known sequence around the boundary. Record the selected interval, entry mode, time constraint, user segment, step definitions and breakdown. Check totals separately from monthly rows. Do not sum monthly unique-user counts and assume the result is a deduplicated annual count.

For a device breakdown, keep the displayed grouping distinct from the hardware used at every later action. A user can enter under one breakdown value and complete elsewhere. A mobile row in this analysis is not a device-by-device inventory of every booking event.

If the older period is unavailable, use the historical-availability audit. A retention limitation and a first-entry effect are different findings; do not change a setting before checking which one the evidence supports.

5. Test the collection path independently#

Create a small permitted test matrix using synthetic booking references or an approved sandbox. Preserve the event timeline and expected result for each journey:

  • Campaign page followed by confirmation without a browsing detour.
  • Campaign page, another page, then confirmation.
  • Confirmation with no observed campaign-page step.
  • An earlier qualifying outcome followed later by the campaign-page journey.
  • An outcome beyond the chosen time condition.
  • A repeat outcome whose meaning is defined by the booking system.

Test the actual success state. A form_submit, URL change or widget-open event may be useful, but it does not automatically represent an accepted appointment. Reopening a confirmation page should also be checked against the intended event contract.

The marketing tracking QA checklist helps retain destination, trigger, timestamps and receiving evidence. For an application that changes routes without a full page load, verify its SPA pageview lifecycle before relying on the campaign-page entry event.

Do not force consent, create identifiers to join unrelated activity or match bookings solely on a nearby timestamp. If the available implementation cannot connect a later device or external booking context, retain the coverage limit. An incomplete observed sequence is not proof that a person never visited the page.

6. Check reporting differences without changing the question#

Google's reports-versus-Explorations guidance documents differences in supported fields, filtering, comparisons, retained periods and recent processing. Inspect any exposed quality or sampling message and the precise query definition. “Unsampled” does not certify correct entry conditions or complete journey visibility.

Use a simple event or page report as supporting evidence, not an automatic replacement for an ordered funnel. A person appearing in both aggregate groups does not prove the required order or elapsed time. Conversely, a funnel excluding them under a strict rule does not erase their booking-event record.

For a recurring external extract, preserve the complete request and response. The current Data API funnel feature is documented in v1alpha, with possible breaking changes. Check that its step conditions, date range and returned metrics match the UI question instead of assuming an ordinary event-count query reproduces the funnel.

7. Report the supported finding, not advertising credit#

A useful statement is: “Under this entry, sequence, time and observed-identity contract, this many users completed the measured journey.” Pair it with the booking-record reconciliation and the known unobserved or unsupported paths.

That statement is not campaign attribution or causal lift. Google's attribution settings reference describes assigning credit to eligible interactions under its reporting rules. A website page followed by a booking does not establish that an advertisement or that page caused the appointment. Use the platform measurement comparison when the next task is reconciling attributed outcomes.

The anonymized fashion-retail delivery-quality case concerns a defined ad-impression population, not appointment-funnel recovery or causal bookings. Keep its delivery evidence separate from the website analysis.

If campaign delivery is also part of the review, export AdProof's supported campaign period, timezone and available dimensions, then compare that delivery evidence with a separately defined website report. AdProof's campaign-export workflow does not configure GA4 funnels, stitch missing booking identities or prove incremental appointments. Discuss a scoped delivery review with the metric and period the campaign question requires.

Frequently asked questions#

Should I make the funnel open to include people who browse before booking?

Not simply for that reason. Indirect sequence rules permit intervening activity. Open entry admits users at later steps, including those without the required campaign-page step, so it may change the business question.

Does a closed funnel mean the booking happened in the same session?

No. Entry mode and sequence or time conditions are separate. If the analysis requires a same-session boundary, express and validate that requirement rather than assuming the Closed setting supplies it.

Why can a wider date range change earlier funnel counts?

An earlier qualifying action can alter entry or the first displayed date/breakdown within the selected interval. Test that possibility alongside unchanged query settings and historical availability; do not diagnose the property's cause from the direction of the count alone.

Can the last step equal fewer users than the booking event report?

Yes. The funnel requires the specified qualifying journey and uses its own user-counting rules. Raw event occurrences, users who triggered the outcome and eligible funnel completers are different populations. Reconcile those definitions before declaring that bookings were lost.

Your next step

Make the funnel rules explicit

Record entry, sequence and time conditions before comparing funnel counts.

Use the tracking QA checklist

Opens a practical guide. No email required.

Back to all field notes