Modelled conversions are conversions that an ad platform estimates rather than records directly, to fill gaps where it cannot see what happened after an ad click. Google Ads adds them to your conversion figures when, for example, a visitor declines cookies, switches device or uses a browser that limits tracking.
How modelled conversions work
Normal conversion tracking relies on joining an ad click to a later action, usually through a cookie or click ID. When that join cannot be made, the conversion is lost from observed data, even though the sale or enquiry happened.
Google looks at conversions it can observe, studies patterns among them (device, time, campaign type, browser and so on), and estimates how many of the unobservable clicks also converted. Those estimates are added to the Conversions column. They are blended in rather than shown as a separate row for each keyword or ad.
The main sources of gaps that modelling covers are:
- Declined consent. With consent mode in place, Google receives signals about visits where the person refused cookies and models conversions for them.
- Browser restrictions Such as Safari’s Intelligent Tracking Prevention, which shortens how long cookies last.
- Cross-device journeys Where someone clicks on a phone and buys on a laptop.
Modelling only switches on when an account has enough data; Google publishes minimum volumes of ad clicks and conversions per country and domain. Small accounts may not qualify.
Why it matters
In the UK, advertising cookies need opt-in consent under PECR, and a well-built banner gives a real “Reject all” option. Some visitors will take it. Without modelling, every one of their conversions vanishes from Google Ads, which makes campaigns look worse than they are and starves Smart Bidding of the signals it needs to set bids.
Modelled conversions put back an estimate of what was lost, so cost per conversion and return on ad spend sit closer to reality. They are still estimates. They work well across a whole account and over weeks, and poorly as evidence about one ad group on one day.
Google is not alone in this. Meta Ads also estimates some conversions it cannot observe directly, for instance from iPhone users who have opted out of app tracking, so the same caution applies when you compare the two platforms or add their figures together. Each platform models its own gaps in its own way, and neither will match your accounts exactly.
Common mistakes
- Removing consent mode, or never installing it, and losing modelling entirely.
- Assuming Google is “inventing” conversions and switching off bidding strategies that rely on them.
- Comparing Google Ads totals with CRM records line by line and expecting a perfect match.
- Expecting modelling in a small account that does not meet Google’s thresholds.
- Importing GA4 key events and also using the Google Ads tag for the same action, double counting before modelling even starts.
How to act on it
- Implement consent mode correctly, with UK visitors defaulted to denied until they choose, and check all consent signals update when they do.
- Add enhanced conversions where your privacy notice and consent set-up allow, so more conversions are observed rather than modelled.
- Reconcile Google Ads conversions with your CRM monthly, at total level, and record the typical gap.
- Judge campaigns over sensible periods so modelled figures can settle.
- Note the date of any change to your cookie banner or consent mode, so a later shift in reported conversions can be explained rather than mistaken for a change in performance.
Setting up consent-aware conversion tracking, and reading the results without over- or under-reacting, is part of how I handle PPC management for UK advertisers.