What it signals, and what it never collects
Consent Mode is four flags. Your site sets them before any Google tag fires, then updates them the moment the visitor chooses. That's the whole mechanism — a signalling layer, not a compliance product.
ad_storage— may advertising cookies and device identifiers be written.analytics_storage— may analytics storage be written, including the identifiers GA4 stitches a session from.ad_user_data— may data go to Google for advertising purposes at all.ad_personalization— may that data then be used to personalise ads.
Why it is not the same thing as your cookie banner
Google's setup guide is explicit that consent mode doesn't save consent choices. Something else — a consent platform, or a banner your developer wrote — has to ask the visitor, store the answer and call the update. Consent Mode only carries the message.
The conflation costs money in both directions. A banner with no Consent Mode wiring blocks nothing on Google's side — the visitor clicks reject and the tag fires anyway. Consent Mode with defaults denied and no working update call does the opposite: it switches measurement off for everyone, including the people who said yes.
Whether you need a banner at all is a separate question with a different answer in India than in Europe. What the DPDP Act actually asks of your analytics is the place to settle that first.
Basic and advanced, and what each keeps
One decision, and it sets how much survives a refusal.
- Advanced is what every measurement vendor recommends and it's usually right — but it does mean Google tags load on every visit, including for the visitor two seconds from refusing.
- Basic matches a strict reading of "nothing loads before consent". You accept the blindness rather than pretending it isn't there.
| Behaviour | Basic mode | Advanced mode |
|---|---|---|
| When Google tags load | Only after the visitor interacts with the banner | Immediately, with consent defaulted to denied |
| What a refusal sends | Nothing at all — not even the consent state | The consent state plus measurements without cookies |
| Modelling it supports | General modelling, built from patterns across advertisers | Advertiser-specific modelling, built from your own consenting traffic |
| Who it suits | Counsel wants the least possible collection before a choice | You need bidding signals and have volume to model from |
Modelled conversions, and what they are not
When consent is denied, the hole gets filled with estimate. Google Ads looks at the relationship between clicks and conversions among traffic it can observe, then applies that relationship to the paths it can't. Modelled conversions then appear in campaign reports with the same granularity as observed ones — same Conversions column, same campaign rows, no asterisk.
GA4 does its own version under the name behavioural modelling, and at least marks affected reports with a data-quality icon. It also has volume thresholds — roughly a thousand denied events a day and a thousand consenting users a day, sustained over weeks. Most Indian SMB properties clear neither, so their refused traffic isn't modelled at all. It's simply missing. The precise thresholds and what a banner costs in measured conversions are worth reading before you sign for a consent platform.
The two sorts of number in that column differ in kind, not just in confidence.
- There's no person behind a modelled conversion. You can't open it, name it or match it to a row in your CRM.
- There's no click ID, so it can't be reconciled against an offline conversion import or a call log without double counting.
- It's a population estimate wearing an integer. "47 conversions" implies a precision the underlying figure doesn't have.
Why we won't settle a guarantee against a modelled number
Our SEO work is judged against one figure: the client's trailing-90-day count of qualified leads from organic search, frozen on day one. Miss it in 90 days and we keep working free until we beat it. That only means anything if both sides can audit the number, so it has to be a count of individual people someone can point at — which is why what counts as a qualified lead is agreed in writing first.
A modelled conversion can't take part in that. It's an estimate of a population, produced by a model neither party can inspect, and it moves when Google changes the model. Useful for reading a trend; hopeless as the thing an invoice turns on.
So on any account running Consent Mode we report two numbers side by side: the platform figure including modelling, and the CRM count. Where they disagree, the CRM count is what the guarantee settles against. It's the less flattering of the two almost always, which is rather the point.