First, test the first page view with no choice made
The test is simple and must be done cleanly. Open a fresh browser profile with no stored consent and no extensions, so nothing remembers an earlier choice. Open Tag Manager (GTMTag management system.) preview with Tag Assistant, which the preview and debug help page describes, then load a page and do nothing: no click on the banner, no scroll that a platform might read as consent. Record three things. The cookies and storage the site has written. The requests that left the page for vendor domains, and for Google requests, the consent state parameter they carry. And, in Tag Assistant's consent view, the default state on the first event and which tags fired or were held.
Then sort what you found. A Google tag sending a request that shows storage denied, with no cookie written, is the advanced pattern of consent mode behaving as documented; whether the site permits it is a policy question for the privacy, consent and measurement guide. Everything else that stored data or reached a vendor before a choice is a gap. Note which template and region you tested, because several causes only show on one of them. If the opposite problem brought you here, too little data after the banner launched, the page on data drops after a consent banner covers it. Then use the table.
Diagnosis table
Diagnosis
| Symptom | Check | Likely cause | Fix |
|---|---|---|---|
| Google requests before any choice show storage granted; analytics cookie present | Tag Assistant consent view: is there a default on the first event | No consent defaults set on the consent initialisation trigger | Declare denied defaults on the consent initialisation trigger |
| Defaults exist but appear after the first tags in the timeline | Event order: the consent platform's template against the page view tags | Consent platform loading after tags have already fired | Move the platform's template to the consent initialisation trigger |
| A pixel or vendor tag fires before the choice while Google tags wait | The consent overview: additional consent settings on non-Google tags | Non-Google tags without additional consent checks | Require the matching consent types on every non-Google tag |
| Vendor requests leave the page but no container tag fired them | Page source for script tags outside the container snippet | Tags hard-coded outside the container | Move them into the container, or gate them in code on the same consent state |
| Correct for visitors in one country, wrong in another | The region list on the default command against where visitors are | Regional defaults not matching where visitors are | Set denied defaults for every region the policy covers |
| Cookies from unfamiliar domains appear although every tag shows held | Requests initiated by scripts that custom HTML tags inject | Custom HTML tags loading further scripts that set cookies | Gate the custom HTML tag on consent, or replace it with a template |
Fix, verify and prevent each cause
No default declared before the tags
Consent mode works only if the default state exists before any tag decides what to do. When no default is declared, Google's tags behave as though consent were granted and write their cookies on the first load. The consent mode guide describes the consent initialisation trigger, which fires before every other trigger, as the one place the default belongs.
Fix: declare denied defaults for the advertising and analytics storage types on the consent initialisation trigger, through the consent platform's template or a dedicated tag, following Google's consent mode documentation.
Verify: a fresh load shows the default on the first event in Tag Assistant, Google requests carry a denied state, and no analytics cookie is written.
Prevent: keep nothing else on the consent initialisation trigger, so the default tag is easy to see and hard to move by accident.
A consent platform that arrives late
Some consent platforms are loaded as an ordinary tag on a page view or window loaded trigger, or as a script lower in the page than the container. By the time they declare defaults, tags on earlier triggers have already fired with no state to respect.
Fix: move the platform's template to the consent initialisation trigger, or if the platform must be hard-coded, place it, with its default command, above the container snippet, as the consent mode guide recommends for hard-coded setups.
Verify: in the timeline, the default appears on the consent initialisation event and before every tag.
Prevent: record the consent platform's loading method in the container documentation and recheck it whenever the platform is upgraded.
Vendor tags that ignore the choice
Google's tags carry built-in consent checks; other vendors' tags in the container do not, unless someone adds them. A pixel, a chat widget or a RemarketingAdvertising to previous visitors or engagers. tag from another network fires on its trigger regardless of the banner.
Fix: open the consent overview in the container settings, find every non-Google tag with no additional consent requirement, and require the consent types it needs, typically advertising storage for ad pixels, as the consent mode guide explains.
Verify: in preview, before a choice, those tags appear as held for consent, and no request reaches their domains.
Prevent: make "consent types declared" a required field in the review of any new tag.
Scripts that never went through the container
A tag in the page template runs before or beside the container and knows nothing about its consent settings. Old analytics snippets, pixels added by a developer, platform plug-ins and app integrations are the usual sources, and they often duplicate tags the container already manages.
Fix: inventory scripts in the page source outside the container snippet, and either move each into the container where consent settings apply, or gate it in code on the same consent state, as the hybrid patterns in the guide to Tag Manager versus hard-coded tags describe.
Verify: a fresh load shows no vendor request that cannot be traced to a container tag held or permitted by consent.
Prevent: agree a rule on which tags may be hard-coded, and include the page source in the consent test.
Defaults set for the wrong place
The default command accepts a region list, so a site can deny by default where a banner is required and grant elsewhere. Mistakes here are invisible to a tester in the wrong country: a region list that misses a market, a granted default with no region that ends up applying where denial was intended, or a consent platform configured for one jurisdiction only.
Fix: set the region list to cover every market the policy names, with denied defaults for each, following the region guidance in the consent mode guide.
Verify: repeat the clean first-load test from each region with different defaults, using a connection that appears to be there, and confirm the default matches the policy.
Prevent: keep the region list in the consent policy itself and review both together when the business enters a new market.
Custom code that brings friends
A custom HTML tag can be held for consent and still cause trouble if it is allowed to fire for one purpose and then loads further scripts that set their own cookies. Vendor snippets that load a library, which then loads partners, are the common case; the container shows one tag, while the browser shows several domains.
Fix: check what each custom HTML tag injects, gate it on the strictest consent type its downstream scripts need, and prefer a gallery template where one exists, as the container governance guide recommends.
Verify: the network panel shows no requests to the injected domains before the matching consent is granted.
Prevent: record each custom HTML tag's owner, purpose and consent category, and treat a new one as a change that needs review.
Prevention: a consent check in every container release
Consent gaps come back with ordinary work: a new pixel, a platform upgrade, a template that loads a script before the container, a new market. The defence is a fixed test that runs before every container publish and every site release that touches templates. For each distinct template, and for each region with its own defaults, load the page in a clean profile, make no choice, and record cookies, vendor requests and the consent state on the first event. Then decline and record again, then accept. Compare the three records with the expected lists, and block the release on any unexpected cookie or request before consent. For a new banner or consent platform, the same test sits inside the consent banner QA checklist, which also covers the banner's choices and the reporting drop that follows.
Support the test with two standing controls. Keep the consent overview switched on and review it monthly, so every tag shows its consent requirement at a glance. And keep a register of every tag and hard-coded script with its owner and consent category, which the container governance guide builds into its audit routine. After publishing, confirm in the GA4Google's event-based analytics platform. property's admin area that consent signals are being received. The analytics troubleshooting hub lists the neighbouring consent symptom for under-collection.
A fashion retailer's pixels before the banner
A fashion retailer's privacy team finds, on a clean first visit with no choice made, 9 cookies and requests to 6 vendor domains. Tag Assistant shows denied defaults on the first event, so the Google tags are behaving as designed. The other findings trace to three gaps. Two ad pixels in the container have no additional consent requirement. A product review widget is hard-coded in the product template and loads its own analytics. And a custom HTML tag for an affiliate network, gated on analytics consent only, injects a partner script that sets advertising cookies once analytics consent alone is granted. The team adds advertising consent requirements to both pixels and the affiliate tag, and moves the review widget into the container behind functionality consent. The next clean first visit shows 2 strictly necessary cookies and no vendor requests beyond Google's cookieless pings.
Common mistakes
Testing in a browser that already holds a consent choice. Treating Google's cookieless requests in the advanced pattern as a leak, or treating a granted state before any choice as normal. Checking the container and forgetting the page source. Testing from one country only. Approving a custom HTML tag without reading what it loads.
