NOTE-001 · FIELD NOTE FILED 14 JUN 2026 · REVISED — F-201 F-204 F-402
Consent Mode v2 didn't lose your conversions. Your tags did.
ABSTRACT — Conversions fall 20 to 40 percent after a Consent Mode v2 rollout while real orders hold steady. Here is why the measurement drops, and the tag-gating fix that recovers most of it.
A Consent Mode v2 rollout almost always drops reported conversions, often by 20 to 40 percent in the first week. The orders themselves do not move. What moves is the measurement: consent-unaware tags stop sending, and nothing fills the gap. The fix is not to roll consent back. It is to make the tags consent-aware and let modeling cover the rest.
Update, June 2026. The 15 June change to Google’s Analytics-to-Ads data controls put more weight on this exact fault. ad_storage consent is now the sole control over whether Google Ads collects cookies and identifiers from the GA4 link, so a mis-wired banner that never sends ad_storage now loses Ads data outright, with no Google Signals fallback behind it. The fix below is unchanged. The cost of skipping it went up. Full breakdown: F-201 .
What you are actually looking at
Two numbers stop agreeing. The shop backend still ships the same volume of orders. GA4 and Google Ads report far fewer. People read the report, not the warehouse, so the panic starts there: “Consent Mode broke our tracking.”
Nothing broke. The consent banner now gates the tags, and the signals it sets default to denied. Tags that used to fire for everyone now wait for a yes that, for a big share of visitors, never comes.
The same three weeks as numbers, so the gap is legible and not just a shape:
| STATE | ORDERS (BACKEND) | CONVERSIONS RECORDED | GAP |
|---|---|---|---|
| Before CMv2 | ~1,000 | 1,000 | 0% |
| After CMv2, tags mis-gated | ~1,000 | 610 | −39% |
| After fix and modeling | ~1,000 | 960 | −4% |
Why the drop is this steep
Default consent is denied, and Consent Mode v2 is specific about what gets withheld. On top of the original ad_storage and analytics_storage, v2 added two signals: ad_user_data (may the hit carry user data for matching) and ad_personalization (may it feed remarketing and audiences). Until a visitor accepts, all four sit denied.
What the tag does next depends on how Consent Mode is wired. In basic mode the tag is blocked outright until consent, so a declined visitor sends nothing and there is nothing to model from. In advanced mode the tag still loads and sends a cookieless ping that Google can model from later. A tag that was never built for consent gets the worst of both: it either does not fire, or it fires without the storage and signals it needs, so Google Ads cannot tie it to a click. That is F-201 on the collection side and F-402 on the recovery side: the hit either never leaves, or it leaves with nothing for Ads to join on. Distinct from F-204 , where the click id is stripped in transit, before any consent decision enters the picture.
How steep it gets depends on your accept rate. A 60 percent accept rate with mis-gated tags loses roughly the 40 percent who declined, plus a slice of the accepters whose tags fired before the consent state was set. That is how a healthy shop posts a 39 percent drop overnight while selling exactly as much as it did yesterday.
The fix, and the honest ceiling
Three moves, in order. None of them touch the storefront.
- Wire the tags for consent, not around it. Advanced Consent Mode, tags set to read the consent state before they fire, the cookieless ping allowed on denial. This recovers the most, because it hands modeling a signal to extend instead of a silence.
- Let modeling do its job. Conversion modeling fills the denied-but-cookieless gap, but it needs volume and a clean signal to model from. Give it both and the recorded line climbs back toward the real one within days. Low-volume accounts may never clear the modeling thresholds, so expect a softer recovery there.
- Turn on Enhanced Conversions for the consented portion. For visitors who granted
ad_user_data, hashed first-party data tightens match quality and reclaims attribution Ads had written off. That is F-402 , and it is usually the last few points.
What you will not get back is 100 percent, and you should not pretend to. The visitors who declined and could not be modeled are genuinely unmeasured, not lost. A recovered line that sits a few points under the backend is honest. A recovered line that matches it perfectly means someone is firing tags through a consent wall, which is the original problem wearing a better suit.
If your conversions fell the week you shipped Consent Mode v2, this is almost certainly the shape of it. The orders are fine. The wiring is not, and the wiring is fixable.