First, date the step and size the expected loss
Plot daily users, sessions and key events across the weeks either side of the banner launch and confirm the step falls on the release day, not a day earlier or later, where a site deploy or campaign change would be the better suspect. Then take the consent platform's own report of how many visitors accept, decline or ignore the banner over the same days. Under basic consent mode, refusals and visitors who never answer send nothing, so the expected loss is close to that combined share. Under advanced consent mode, the observed loss is similar but cookieless pings still arrive, and modelled figures may fill some of it later.
Compare the two. A drop close to the decline share is the documented cost of consent and is a reporting matter, explained in the privacy, consent and measurement guide. A drop clearly larger than it, or losses among visitors who accepted, is an error to diagnose with the table below. If the property shows nothing at all after the launch, the sibling page on GA4 not showing any data is the faster route. For the tests, use Tag Manager (GTMTag management system.) preview with its Consent tab, and a clean browser with no stored choice.
Diagnosis table
Diagnosis
| Symptom | Check | Likely cause | Fix |
|---|---|---|---|
| Visitors who accept still produce no full events; storage stays denied | Consent tab after accepting: does an update event appear and flip the state | Consent defaults set to denied with no update after acceptance | Make the banner issue the update on acceptance and on later pages |
| No requests at all before a choice or after a refusal; modelling never becomes available | Whether the banner or a blocking trigger prevents Google tags loading | Tags blocked by the banner rather than adjusted by consent mode | Let consent mode adjust Google tags instead of blocking them, if the policy allows |
| Accepting visitors lose their first page view or landing source | Timeline order: consent update against the page view tag | Consent update firing after the page view tag | Hold tags for the update, or sequence the update before the page view |
| The drop matches the decline share and nothing is sent for refusals | The implementation mode recorded in the consent platform's template | Basic consent mode sending nothing for declining users | Accept and annotate it, or change mode by policy decision |
| Reports still show only observed data weeks later | The property's consent and modelling status in admin; reporting identity | Modelled data not yet available or the property not eligible | Wait for or meet the documented eligibility conditions |
| Small segments vanish and reports carry a threshold icon | The threshold indicator and the reporting identity setting | Reporting thresholds hiding rows | Widen the range, drop sensitive dimensions or change reporting identity |
Fix, verify and prevent each cause
Defaults that are never updated
The defaults declared before any Google tag set the starting state. If the banner records acceptance but never issues the update command, or issues it only on the page where the choice was made, Google's tags keep behaving as though storage were denied.
Fix: make the banner, or its template, issue an update when the visitor accepts and re-issue it on every later page from the stored choice, as the consent mode guide describes for default and update commands.
Verify: after accepting, the Consent tab shows an update and the state flips to granted, and the analytics cookie appears.
Prevent: include "accept, then navigate to a second page" in every consent test, since the second page is where a missing re-issue shows.
Blocking instead of adjusting
Many banners block scripts until consent, and some containers carry blocking triggers that do the same. The result is the basic pattern whether or not anyone chose it: nothing is sent before a choice or after a refusal, so there is no input for behavioural modelling.
Fix: if the privacy policy permits it, let consent mode adjust Google tags rather than blocking them, keep blocking for tags that must wait, and record the decision; the consent mode guide sets out how the two approaches differ.
Verify: before a choice, cookieless requests leave the page with storage shown as denied, and no analytics cookie is written.
Prevent: read which pattern the consent platform's template implements before launch, not after the numbers move.
The update lands too late
When the page view tag fires before the update, a visitor who accepts on the first page sends that page without storage, and the landing page and its source can be separated from the rest of the visit.
Fix: fire the consent defaults on the consent initialisation trigger, use the wait-for-update setting for visitors whose choice is already stored, and where needed hold the page view tag until the update event, using the trigger ordering and sequencing in the tags, triggers and variables guide.
Verify: in the timeline, the update precedes the page view tag for a returning visitor who already accepted.
Prevent: test with a stored choice as well as a fresh one, because the two paths fire in a different order.
Basic mode working as designed
Under basic consent mode, tags do not load until the visitor interacts with the banner, and a refusal means nothing is sent. The resulting drop is the expected loss, not a bug.
Fix: nothing in the container; agree how reports will present the change, and if the business wants modelled figures, raise the mode as a policy question using the privacy, consent and measurement guide.
Verify: the drop stays close to the decline share week after week.
Prevent: annotate the launch date everywhere the affected figures are reported, so later readers do not mistake the step for lost demand.
Modelling not there
Behavioural modelling fills part of the gap only when the documented conditions are met: the advanced implementation, consent mode across the site, enough traffic over the documented period, and a reporting identity that includes modelled data.
Fix: check the property's consent and modelling status in admin and close whichever condition is missing, as the privacy, consent and measurement guide explains.
Verify: reports indicate that modelled data is included once the property qualifies.
Prevent: treat modelled figures as estimates, label them, and keep an observed-only view for decisions that need it.
Thresholds hiding rows
With a reporting identity that includes Google signals or modelled data, small rows can be withheld and the report marked with an icon. After a banner launch, fewer observed users per row makes this more frequent.
Fix: widen the date range, remove demographic dimensions, or switch to the device-based reporting identity, which the reports and dimensions guide describes as a view-time setting.
Verify: the icon disappears and the missing rows return.
Prevent: build recurring reports at a grain large enough to clear thresholds.
Prevention: testing consent changes before release
Consent changes are releases and need the same discipline. Before any banner, template or policy change goes live, test three states on every distinct template, including checkout and logged-in pages: no choice yet, refusal, and acceptance followed by navigation. In each, read the Consent tab, the requests in the network panel and the cookies written. Repeat for each region whose defaults differ. Record the expected decline share and the expected reporting effect, and brief report owners before launch, not after. Annotate the release date in GA4Google's event-based analytics platform. and in every dashboard. Then watch the ratio of analytics key events to the order system or CRM for a few weeks; a stable ratio after the step means the loss is consent, a falling one means a fault. Channel AttributionAssigning conversion credit to touchpoints. shifts too, because channels whose visitors decline more often lose more credit. The consent banner QA checklist gathers these release checks, from the three-state test to the annotation and the briefing for report owners, into a list to tick before launch.
A banner launch that cost more than consent
A furniture retailer launches a banner and sees sessions fall 46% in a week. The consent platform reports that 31% of visitors decline and 6% never answer, so about 37% is expected. Preview shows the remainder: accepting visitors get an update on the first page, but the template never re-issues it, so every later page of their visit is sent with storage denied. The vendor template's setting for restoring stored consent was off. Turning it on and retesting the three states closes most of the gap; the following week sessions sit 39% below the pre-banner level, close to the expected loss, and the team annotates both dates.
Common mistakes
Testing only the accept path. Comparing post-launch numbers with pre-launch ones without an annotation. Treating every point of the drop as a bug. Inheriting basic or advanced mode from a template default. Reading modelled figures as observed ones.
