- Published
- Oct 31, 2026
- Reading time
- 10 min read
GAM Roadblocks: Reconcile Creative Impression Counts
Investigate unequal GAM roadblock counts by checking delivery settings, grouped requests, creative eligibility, rendering and the booked impression basis.

In this guide (10 sections)
If two creative sizes in a GAM roadblock report different impression totals, check the serving contract and counting unit before calling the roadblock broken. “All,” “As many as possible” and master/companion sets do not establish the same promise or reporting basis. Trace the intended group through the actual slot request, returned creatives and separately observed counting events.
A real publisher question about unequal two-slot roadblock counts compares a rectangular creative with a horizontal banner across several trafficking configurations. The publisher uses SRA and lazy loading and asks whether equal creative impressions are a fair acceptance test. The archived discussion contains old implementation advice, including claims about private GPT behavior. Treat it as question provenance, not a current installation recipe or proof of the historical cause.
The useful deliverable is a small page-level reconciliation: which group was intended, which slots participated, which creatives were returned, what each observer counted and which records remain unknown. Equal aggregate totals can hide different page populations, so they are not sufficient evidence of successful co-delivery either.
Write the promised group before comparing reports#
Ask the campaign owner to define the placement requirement in plain language. Does the agreement require two creatives selected together, two counted display impressions, simultaneous visible placements, or a particular master/companion sequence? Record which system is responsible for each observation.
A supported serving guarantee is not a guarantee that a person saw both placements. A lower slot may never enter the viewport; another creative may encounter a loading problem. Those are separate checks, with their own measurement definitions.
| Record | What to preserve |
|---|---|
| Commercial group | Intended placements, sizes, creative relationship and accepted counting source |
| Serving configuration | Display setting, network feature availability, targeting and effective version |
| Page implementation | Actual slots, definition order, request grouping and lifecycle |
| Creative inventory | Associated IDs, sizes, template and eligibility |
| Measurement | Exact report metric, counting event, dates, timezone and extraction time |
Do not substitute a setting's name for the agreement. If sales uses “page takeover” while operations books creative impressions, settle the conversion between those units before reviewing delivery.
Check the roadblock promise and counting unit
Roadblock modes do not all promise the same creative group or impression count. Discuss what the serving records can verify.
Discuss a measurement questionA short inquiry to scope relevant options; no campaign access is needed.
Distinguish the display modes#
Google's roadblock setup guide distinguishes “All” from “As many as possible.” The latter does not guarantee that multiple creatives serve together. Guaranteed roadblocks require SRA and the applicable enabled network feature. The guide also explains request-processing order and individual creative requirements.
The display-creatives reference documents other modes and creative-count settings, including specific Ad Manager 360 reuse functionality. Check the actual feature and supported combination rather than assuming every account exposes the same option. A missing setting is not itself evidence of a broken account.
Save the active line-item configuration and its history. If several candidate line items or tests ran at the same time, identify which records belong to each. Adding their totals together can conceal that no single configuration met the intended group contract.
For an “All” requirement, ask the administrator to verify network eligibility and the actual enabled feature. For an “As many as possible” arrangement, document the accepted partial-delivery behavior. Neither question can be answered from the line-item title alone.
Check master/companion counting separately#
Master/companion is not merely another label for two independently counted placements. Google's master/companion documentation describes an Ad Manager 360 feature in which the master serves first and companions do not serve without it. It specifies one line-item impression for the served creative set rather than one for every creative.
Google's roadblock setup guidance also distinguishes the booked quantity: ordinary non-master/companion creative delivery and a master/companion goal use different counting bases. Preserve that distinction when interpreting an export. Do not assume a line-item impression total can be divided equally between creative rows, or that the absence of such a division proves lost delivery.
Write a short mapping before performing arithmetic:
Arrangement and applicable feature:
Required creative group:
Booked unit and goal:
Reported line-item metric:
Reported creative-level fields, if supported:
Independent observation and its event definition:
Permitted conversion between those units:
If the report does not expose a needed creative-level observation, mark it unavailable and seek supported diagnostics. Do not manufacture a companion row or infer its count from the master's aggregate.
Inspect which slots actually reached the grouped request#
SRA is part of the request contract, not a magic equality setting. Google's GPT best practices describes grouping requests for page-level features and configuring slots before the initial request. Record the actual request group and slot mapping for the affected page cycle.
A page template can reserve space for two placements while only one slot is registered or included at the relevant time. Responsive layout, conditional components, deferred content and route changes deserve explicit checks. Record intended sizes and actual size mapping rather than relying on a screenshot of desktop layout.
Use the existing Prebid/GPT lazy-loading and SRA guide for the timing investigation. The present task adds the group-serving contract and counting reconciliation. Do not assume that the old forum claim “lazy loading and SRA cannot work together” describes current GPT support.
In an approved QA environment, save the slot definitions, their order, effective targeting and exact request containing them. Compare one successful control with the affected context. A bundled HTTP request can contain multiple slot opportunities; it is not one counted creative impression.
Avoid private GPT variables or undocumented correlator manipulation. If the supported integration cannot expose the required grouping evidence, ask its owner for an equivalent trace.
Verify creative eligibility before blaming the count#
Map the intended creative to each slot by supported identifiers, sizes and targeting. Confirm that the creative is active and associated with the actual line item in the test. A familiar image or creative name does not establish that the expected ID served.
For repeated same-size placements, review the configured creative count and supported reuse behavior with the trafficking owner. The fact that a creative is eligible for a size does not demonstrate that it can occupy every matching slot under every display mode.
Then inspect supported non-delivery diagnostics for the particular request. Preserve any exclusion or competition reason without promoting it into a campaign-wide cause. A simulated request is another context; it does not replace the page state of the original observation.
Keep this investigation separate from generic fallback coverage. A missing companion is not automatically repaired by a House creative, and filling the empty space with another line item does not establish that the booked roadblock delivered.
Follow returned creatives through their own evidence#
Use a per-slot ledger to record request, response, creative-code insertion and the reportable counting event. Google's GPT event reference distinguishes code insertion from completed resource loading. It also documents how refresh or slot destruction affects GPT's long-lived pageview. Preserve the supported lifecycle rather than assuming a later slot cycle belongs to the original group.
| Page-cycle observation | What it can establish |
|---|---|
| Both intended slots included in the relevant request | Request grouping for that cycle |
| Expected creatives returned for both slots | Serving decision evidenced for the pair |
| Code injected for a creative | That injection stage occurred |
| Supported counted-impression receipt | That observer recorded its defined event |
| Viewability evidence | A separate supported exposure condition, within measurement coverage |
An absent browser callback is missing evidence until its delivery and coverage are understood. Conversely, a successful callback should not be upgraded into a counted or viewable impression without the applicable definition.
If client diagnostics show both creatives returned while one report is lower, investigate the downstream handoff, report scope and counting basis. Our ad-server impression discrepancy guide provides the receiver and report reconciliation. Do not change group trafficking solely to force two totals to agree.
A hypothetical paired reconciliation#
Hypothetical example. Assume 100 comparable page cycles use an ordinary, non-master/companion arrangement with two distinct creatives. In an authorized illustrative ledger, both intended creatives were returned for all 100 cycles. The following table describes which defined impression records were subsequently matched. These are invented observations, not a GAM customer result.
| Matched downstream records per cycle | Page cycles |
|---|---|
| Both creative A and creative B | 90 |
| A only | 5 |
| B only | 0 |
| Neither | 5 |
| Total | 100 |
Creative A has 90 + 5 = 95 matched impression records. Creative B has 90. The total is 185 creative records across 100 page cycles. Ninety cycles have a matched pair; that is 90% of this fully specified illustrative sample.
The five-count difference is not the same as the number of cycles without a fully matched pair. There are ten such cycles: five A-only and five with neither record. Reporting only the difference would hide the common downstream gap. Nor do these figures prove that ten people failed to see both ads; the ledger counts defined observations, not people.
If you only had the two marginal totals, 95 and 90, you could not reconstruct the table's overlap. A paired page-cycle relationship or supported aggregate equivalent is needed. If linking is unavailable, report the aggregate difference and that limitation.
For a separate master/companion example, a goal of 100 line-item impressions refers to the set's documented counting basis. Do not automatically replace it with 200 because the set contains two creatives. Keep this goal arithmetic separate from the ordinary per-creative example above.
Define the acceptance test for the actual arrangement#
Agree a narrow statement before changing the configuration. For example: the two intended slots participate in the supported group request; the configured mode selects the expected creative set in the documented test context; and each observer's count is reconciled under its own definition.
The exact campaign agreement may require additional evidence. If it promises viewable co-exposure, a return trace alone is insufficient. If it books master/companion sets, per-creative sums are not automatically the goal. Escalate an undefined commercial unit instead of resolving it with an arbitrary discrepancy tolerance.
After an approved change, retain the old and new settings, deployment time, test environment, request references, counts and unresolved cases. Change one variable at a time where practical. Do not replace one uncertainty with several simultaneous priority, grouping and refresh changes.
The frequency audit addresses another distinct unit. Google's roadblock overview says a line-item delivery counts once toward its frequency cap even when multiple creative impressions are reported. Do not interpret creative totals as an equivalent number of capped user exposures.
Frequently asked questions#
Should two roadblock creative sizes always have equal impressions?
Do not impose that rule without the serving arrangement, report population and counting definition. Trace a paired sample. Unequal totals can identify a useful investigation, while equal totals alone cannot prove co-delivery.
Is As many as possible equivalent to All?
No. They express different serving behavior. Verify the applicable account feature, tag mode and configuration rather than relying on the word “roadblock” in a campaign name.
Does SRA mean every creative was viewed?
No. It concerns request grouping and supported page-level serving behavior. Creative return, counting and viewability require separate evidence.
Should master/companion goals equal the sum of every creative?
Not automatically. Preserve the documented set-level line-item counting basis and distinguish it from ordinary per-creative delivery before calculating a goal or discrepancy.
Compare a defined delivery population#
The anonymized fashion retail display case illustrates a documented impression baseline. It does not establish that a GAM roadblock, creative set or viewability promise was satisfied.
Where AdProof measures the campaign, compare available event or aggregate exports with the publisher's matching delivery scope, dates and timezone. Keep the group-request and creative-eligibility evidence in the owning ad server and implementation trace. Discuss the delivery comparison without assuming AdProof configures GAM roadblocks or reconstructs an unobserved page-level pairing.
Check the roadblock promise and counting unit
Roadblock modes do not all promise the same creative group or impression count. Discuss what the serving records can verify.
Discuss a measurement questionA short inquiry to scope relevant options; no campaign access is needed.
