Consent Mode Conversion Modeling: What You Stopped Seeing

When a visitor declines cookies, the conversion they go on to complete is not recorded the way it used to be. Consent mode lets Google's tags still send a signal without cookies, and Google then estimates some of the missing conversions using machine learning. Those estimates appear in the same Conversions column as counted ones. A modeled conversion is a statistical inference, not an observed event, and if your account is small you may not qualify for modeling at all.

That last sentence is the part that changes how you report. The drop you saw after the banner went up is partly real loss of measurement and partly a change in what the number even means. Neither of those is a campaign performance problem, and treating it as one leads to cutting a channel that never actually got worse.

This page covers what happens to your numbers. It is not legal advice, and it cannot tell you whether your setup satisfies the law where you operate. That is a question for your own legal advisers, and the answer differs by country, by what you collect and by what you do with it.

Which rules apply to you, and where they come from

Consent obligations are not global and they are not one rule. Three separate layers are usually in play.

Layer Who it binds Where it comes from
The law where your visitor is You, as the site owner National law implementing the EU ePrivacy Directive, UK PECR, US state privacy statutes, and others
Google's own product policy You, as a Google advertising customer Google's EU User Consent Policy
Your consent platform's configuration Your tags Whatever your banner is actually set to send

Google's EU User Consent Policy applies in the "European Economic Area, the UK and Switzerland" and requires consent for "the use of cookies or other local storage where legally required" and for "the collection, sharing, and use of personal data for personalization of ads" (Google EU User Consent Policy, checked 2026-09-07). It is a contractual obligation to Google on top of whatever the law requires, and the policy states that failure to comply may lead Google to "limit or suspend your use of the Google product".

If your visitors are in the United States or Canada, that policy is not what governs you, and the local picture is different in kind: US state privacy laws generally run on notice and opt-out rather than prior consent, and they vary state by state. Do not assume a banner built for the EEA describes your obligations, and do not assume the absence of one means you have none.

One nuance almost nobody writing about consent mode mentions. In the UK, the ICO's guidance on storage and access technologies sets out an exception for statistical purposes, which can cover analytics used only to improve your own service, on conditions that are narrow: the data must be used for that purpose alone and the output must be aggregate information that cannot identify people (ICO guidance on the use of storage and access technologies, checked 2026-09-07). The moment the same measurement feeds ad personalization or is shared with an ad platform for that purpose, you are outside it. So "analytics might not need consent" and "my Google Ads conversion tracking does not need consent" are two very different statements, and only the first one is ever arguable. Whether your own use falls inside the exception is, again, your lawyer's call and not ours.

Counted, missing, modeled: three different things

Once a banner is live, your conversion total is made of parts that used to be one thing.

  • Counted. The visitor consented, cookies were set, the event was observed and attributed in the ordinary way.
  • Missing. The visitor declined, no cookie was set, the conversion happened and nothing links it to the ad click. It is gone.
  • Modeled. Google estimates part of the missing group from the behavior of the consented group and adds the estimate to your reports.

Google Ads describes modeled conversions plainly enough: "Modeled conversions will appear in the 'Conversions' column and be reflected in all downstream reports that use this data", and modeled values appear in the conversion value column too (Google Ads Help, "About consent mode modeling", checked 2026-09-07). The same page states the limitation directly: "Google is unable to observe whether an unconsented conversion previously had an ad interaction", and notes that "some conversions that in reality occurred may not be accounted for".

In Analytics the equivalent is behavioral modeling, which "uses machine learning to model the behavior of users who decline analytics cookies based on the behavior of similar users who accept analytics cookies" (Analytics Help, "Behavioral modeling for consent mode", checked 2026-09-07). Google's own page warns you will "probably see differences when compared to reports that include only observed data (for example, higher user counts in reports that include modeled data)".

The practical rule: a modeled number is an estimate presented in the same column as a measurement. It is not a lie and it is not padding. It is also not something you can drill into, dedupe against your CRM, or hand to finance as a record of individual customers.

You may not be eligible, and small accounts often are not

This is the part that catches in-house teams with modest budgets, and it is worth checking before you plan any reporting around modeled data.

Product Published threshold
Google Ads conversion modeling Correctly implemented consent mode or IAB TCF v2.0, plus "a daily ad click threshold of 700 ad clicks over a 7 day period, per country and domain grouping"
Analytics behavioral modeling At least 1,000 events per day with analytics storage denied for at least 7 days, and at least 1,000 daily users sending events with analytics storage granted for at least 7 of the previous 28 days

Sources: Google Ads Help, "About consent mode modeling", and Analytics Help, "Behavioral modeling for consent mode", both checked 2026-09-07.

Read the Google Ads line carefully, because the qualifier does the damage: per country and domain grouping. An account spending across several countries can clear the total comfortably and still qualify in none of them individually. If you run deliberately narrow geography, as many small budgets should and as location targeting on a small budget argues, you may be spreading clicks below the threshold in every market at once. The Analytics side has its own trap: it needs a substantial volume of declined traffic as well as consented traffic, so a low-traffic property gets no modeling regardless of how the banner is configured.

Meeting the numbers is also not a guarantee. Google's Analytics page notes the model has additional internal criteria and may simply fail to train.

What this does to your reporting

Three consequences, in the order they will bite you.

A step change at the banner date is not a performance change. The week the banner went live, your measurement changed. Comparing the month before to the month after compares two different instruments. Annotate the date in every dashboard you own and refuse the year-on-year comparison across it until you have rebuilt a baseline on the new basis.

Reconciliation gets harder, not easier. A modeled conversion has no matching row in your CRM, because no individual completed it. If you match platform conversions against closed deals, expect the platform side to include a portion that cannot be matched by design. This is a separate cause from the mechanical ones such as a redirect stripping the click identifier, and it is worth ruling those out first, because they are fixable and this is not.

Bidding is affected before reporting is. Automated bidding consumes what the ad account records. Modeled conversions restore some of the signal it would otherwise lose, which matters most on accounts that have the least data to spare. If bidding is behaving as though it is guessing, read what Smart Bidding does when there are not enough conversions alongside your consent setup, because a small account below the modeling threshold is losing signal twice.

Attribution reporting is affected too, in a way that is easy to misread: with part of the picture inferred, the credit split between channels is partly inferred as well. The choices in attribution models in plain language still apply, but the underlying data is no longer wholly observed, and the honest version of any channel chart built on it says so.

How to report it without overclaiming

Say what the number is. In a footnote, in a column header, in one line of the commentary: "Conversions include modeled conversions where Google's thresholds are met." That sentence takes ten seconds and protects you from the meeting where someone tries to match the platform total to the sales ledger and concludes the marketing team is inventing results.

Keep a separate, unmodeled series if you can. Your own back office count of orders or qualified leads did not change when the banner went up, which makes it the only continuous series you have across the boundary. Trend that alongside the platform numbers rather than instead of them.

And separate the two questions when you present the drop. "We are measuring less" and "we are selling less" look identical on a chart and require opposite responses. The vocabulary problem underneath this is the same one described in key events, conversions and the words GA4 keeps renaming: once a single word covers several different objects, a meeting can spend an hour disagreeing about a definition while believing it is disagreeing about performance.

What we could not confirm

  • How much of the loss modeling recovers, for you. Recovery figures circulate, including from Google's own marketing material about early results. We found no figure on the Help pages above that applies to a specific account, and a recovery rate quoted for a global average tells you nothing about your property. We print none.
  • Whether your account is currently receiving modeled conversions. Google does not expose a simple flag for this in the places most advertisers look. Treat eligibility as unknown unless you can confirm it.
  • Whether any given configuration satisfies the law. Not our call, not a marketer's call, and not something any article can tell you. Compliance is a legal question about your specific data and your specific jurisdiction.

Both products change here more often than anywhere else in measurement, which is why this page carries a three month review date. If you are reading it long after 2026-09-07, check the thresholds and the policy scope against the sources below before relying on them.

FAQ

What is consent mode conversion modeling? It is Google using machine learning to estimate conversions that could not be observed because the visitor declined cookies, then reporting those estimates in your normal conversion columns. Google states that modeled conversions appear in the Conversions column and flow into downstream reports (Google Ads Help, checked 2026-09-07).

Are modeled conversions real conversions? The underlying actions are real. The attribution of them to your ads is inferred rather than observed, and the count is an estimate. Google states it "is unable to observe whether an unconsented conversion previously had an ad interaction". Do not treat a modeled conversion as a record of an individual customer.

Why does my account get no modeled conversions? Most likely you are below the published thresholds. Google Ads requires 700 ad clicks over a 7 day period per country and domain grouping, and Analytics behavioral modeling requires four-figure daily event and user volumes on both the declined and consented sides (both checked 2026-09-07). Narrow geography or low traffic can put you under the line in every market.

Do I need a cookie banner if my visitors are in the US? Different question with a different answer, and it is not one this page can settle. Google's EU User Consent Policy covers the EEA, the UK and Switzerland. US state privacy laws generally work on notice and opt-out rather than prior consent, and they differ by state. Ask your legal advisers about your own situation.

My conversions dropped the week we launched the banner. Did the campaigns get worse? Probably not, or not by that amount. Some measurement was lost and some of what remains is now estimated. Annotate the date, rebuild your baseline after it, and check an unmodeled series such as your own order or lead records before you change any campaign.

Sources, read 2026-09-07

Google Ads Help: "About consent mode modeling" (support.google.com/google-ads/answer/10548233). Analytics Help: "Behavioral modeling for consent mode" (support.google.com/analytics/answer/11161109). Google EU User Consent Policy (google.com/about/company/user-consent-policy/). Information Commissioner's Office: guidance on the use of storage and access technologies, including the exceptions section (ico.org.uk).

Nothing here is legal advice. Consent requirements differ by country and by use, and only your own legal advisers can tell you what applies to your site.

Similar Posts