Why your GA4 conversions dropped overnight
The campaigns did not change, the site did not change, and yet Tuesday's conversions are half of Monday's. When GA4 falls off a cliff overnight, the cause is almost never the marketing. It is nearly always the measurement layer — and usually one of five things.
1. Someone touched the banner
The most common culprit. A CMP update, a new banner version, a redesign that moved the accept button — anything that changes how or when consent is granted changes how many tags fire. If the drop lines up with a deploy or a CMP release, start there. Check whether the default consent state is still set before your Google tags load, and whether the update silently switched you from advanced to basic behaviour.
2. A tag now fires before consent is ready
Tag order is fragile. If Google Tag Manager loads before the consent defaults are set, tags fire in a denied state or get blocked entirely, depending on your setup. The symptom is a partial drop rather than a clean zero: some browsers, some regions, some pages. The fix is structural — consent defaults first, everything else after. In GTM, the Consent Initialization trigger exists precisely for this.
3. A region rule flipped
Consent platforms let you vary behaviour by region: explicit consent in the EU, notice only elsewhere. One toggle in those settings can put your largest market behind a wall it was not behind yesterday. If the drop is concentrated in one geography, compare your region configuration against last week's.
4. The measurement moved, not the conversions
Check the business numbers before panicking: orders in the shop system, leads in the CRM, calls answered. If revenue held steady while GA4 fell, you have a measurement problem, which is annoying but fixable. If both fell, you have a different conversation ahead of you.
5. Modelling stopped
If a meaningful share of your reported conversions were modelled, losing modelling eligibility looks exactly like losing conversions. This happens when advanced consent mode gets downgraded, when the cookieless pings stop, or when traffic drops below Google's thresholds. The tell is in the timing: modelled data adjusts over a few days rather than instantly.
The fast diagnostic
One session with the network tab answers most of it. Load the site fresh, reject the banner, and watch what Google receives. Then accept and watch again. Compare both against what your reports assume is happening. On sites running Velo, the audit log shows every consent state change and every version of the banner, so the question of what changed on Tuesday has an answer you can read rather than reconstruct.