What Happens to Your Data When Visitors Reject Cookies
Every time a visitor clicks "Reject" on your cookie banner, a measurement gap opens. Google Analytics 4 can no longer tie that visitor's session to an ad click, track their journey through your site, or record a conversion event linked to their identity. For advertisers spending money on Google Ads, this gap translates directly into underreported return on ad spend.
Consent rates across Europe frequently sit between 30% and 70%, depending on industry and geography. That means a significant share of your traffic generates no usable conversion data at all.
Google's answer is conversion modelling - a machine learning process built into Consent Mode v2 that attempts to fill those gaps by estimating what unconsented visitors likely did, based on patterns observed from visitors who did consent.
How Consent Mode v2 Sends Data Without Cookies
Consent Mode v2 operates in two distinct modes, and the difference between them determines whether modelling is even possible.
Basic mode blocks all Google tags until a visitor grants consent. If they reject, no tags fire, no data is collected, and Google has zero signals to work with. Modelling under Basic mode relies on broad industry benchmarks rather than your site's actual behaviour, making it far less accurate.
Advanced mode loads Google tags on every page regardless of the visitor's consent choice. When consent is denied, those tags send what Google calls "cookieless pings" - minimal, anonymised signals that include the page URL, a timestamp, user agent information, and a flag indicating whether a conversion event occurred. These pings contain no cookie identifiers and cannot track an individual across sessions. They do, however, give Google enough aggregate data to build a site-specific model.
The distinction matters. If you have implemented only Basic mode, the modelled numbers you see in Google Ads are rougher estimates based on general patterns, not calibrated to your audience.
The Machine Learning Behind Modelled Conversions
Google's modelling process works by comparing two populations: visitors who consented (and whose full journey is tracked) and visitors who denied consent (and who sent only cookieless pings). The algorithm identifies behavioural patterns - pages viewed, time on site, device type, geography, referral source - that correlate with conversions among consented users. It then applies those patterns to the unconsented group to estimate a conversion probability for each session.
This is not a simple ratio. The model accounts for the fact that users who reject cookies may behave differently from those who accept them. A visitor arriving from a privacy-focused browser like Firefox or via DuckDuckGo may have different purchasing habits than one using Chrome with all defaults.
Google has stated that conversion modelling through Consent Mode recovers, on average, more than 70% of ad-click-to-conversion journeys that would otherwise be lost. Individual results vary widely.
Data Thresholds You Must Meet
Modelling does not activate automatically. Both GA4 and Google Ads impose minimum data requirements before their respective models start producing estimates.
| Platform | Requirement | Minimum Volume | Time Period |
|---|---|---|---|
| GA4 (Behavioural Modelling) | Events with analytics_storage='denied' | 1,000 per day | 7 consecutive days |
| GA4 (Behavioural Modelling) | Users with analytics_storage='granted' | 1,000 per day | 7 of the previous 28 days |
| Google Ads (Conversion Modelling) | Ad clicks per country and domain grouping | 700 total | 7-day rolling window |
Smaller websites often fail to meet these thresholds. If your site receives fewer than 1,000 daily sessions from users who deny consent, GA4 behavioural modelling will not activate. Meeting the threshold once is not enough either - it is a continuous requirement. A drop in traffic or a shift in consent rates can deactivate modelling without warning.
Even after thresholds are met, Google notes that training and activating the models can take additional time.
GA4 Reporting Modes: Observed vs Blended
GA4 gives you a choice in how modelled data appears in your reports. Under the "Observed" setting, reports show only data from users who accepted analytics cookies. No modelling is applied. Under "Blended," GA4 merges observed data with modelled estimates for unconsented users, producing higher - but less certain - numbers.
Switching to Blended does not retroactively change historical data. It applies from the point of activation forward.
The choice between these modes depends on your tolerance for uncertainty. If you rely on GA4 for financial reporting, Blended figures introduce a margin of error that should be clearly communicated to stakeholders. If you use GA4 primarily for directional insights and trend analysis, Blended mode offers a more complete picture.
Accuracy Limitations You Should Know
Conversion modelling is an estimate, not a measurement. Several factors affect how reliable those estimates are.
Consent Rate Skew
The model assumes that consented users are reasonably representative of unconsented users. This assumption weakens when consent rates are very low. If only 20% of visitors accept cookies, the consented group may be a poor proxy for the 80% who declined - particularly if consent fatigue or banner design pushes a non-random subset toward rejection.
Small Sample Sizes
Sites that barely clear the minimum thresholds may see volatile modelled numbers. A few hundred extra sessions on a given day can shift the model's output significantly. Google does not publish confidence intervals or error margins for modelled conversions, so you have no built-in way to gauge how reliable a specific number is.
Regional and Device Variation
Consent rates differ sharply by country. A site with traffic from both Germany (where rejection rates tend to be high) and the United States (where many states follow an opt-out model) will produce modelled data of uneven quality across regions. The model also considers device type, but mobile visitors - who convert at different rates from desktop users - may be disproportionately represented in the unconsented group.
No Visibility Into the Model
Google does not disclose the specific features or weights used in its conversion models. You cannot audit the model, test it against your own data, or adjust its assumptions. This opacity makes it difficult to validate modelled numbers independently.
Basic vs Advanced Mode: Impact on Modelling Quality
The implementation mode you choose has a direct effect on how much data Google can use for modelling.
| Feature | Basic Mode | Advanced Mode |
|---|---|---|
| Tags load before consent | No | Yes |
| Cookieless pings sent | No | Yes |
| GA4 behavioural modelling | Not available | Available (if thresholds met) |
| Google Ads conversion modelling | Industry benchmarks only | Site-specific calibration |
| Data accuracy | Lower | Higher |
Advanced mode provides richer signals because it collects cookieless pings from every visitor, including those who deny consent. Basic mode, by contrast, gives Google nothing to work with for unconsented sessions, forcing it to rely on generalised industry data rather than patterns specific to your audience.
The trade-off is a legal one. Some privacy professionals argue that loading tags before consent - even if those tags send only anonymised pings - may conflict with the ePrivacy Directive's requirement for prior consent before storing or accessing information on a user's device. The legality of Advanced mode under Article 5(3) of the ePrivacy Directive remains debated.
Practical Steps to Get the Most From Modelled Data
If you rely on Google Ads or GA4 for conversion tracking, a few practical measures can improve modelling quality.
First, implement Consent Mode v2 in Advanced mode through your CMP. Ensure your Google Tag Manager container fires the consent default command before any other tags load. Second, verify that cookieless pings are actually being sent by checking the Network tab in your browser's developer tools for requests to Google's collection endpoint when consent is denied.
Third, treat modelled conversions as directional indicators rather than exact figures. Build a reporting buffer by comparing Blended and Observed reports periodically, and flag the difference as your "modelling margin."
Fourth, consider supplementing Google's modelled data with privacy-preserving analytics tools that do not require cookies at all. Running a cookieless analytics tool alongside GA4 gives you a second data source against which to validate modelled trends.
Finally, optimise your cookie banner design to improve consent rates ethically. A well-designed banner with clear language and equal button prominence increases the consented data pool, which in turn makes the model more accurate. Avoid dark patterns - regulators are actively fining sites that manipulate users into accepting cookies.
Frequently Asked Questions
Does conversion modelling work without Consent Mode v2?
Google Ads has used conversion modelling for years to fill gaps from cross-device journeys and Safari's ITP restrictions. Consent Mode v2 extends this to consent-related data loss specifically. Without Consent Mode v2 implemented, Google Ads can still model some conversions, but the estimates are less accurate because they lack the cookieless ping signals that Advanced mode provides.
How accurate is Google's conversion modelling?
Google reports that Consent Mode modelling recovers over 70% of lost conversion journeys on average. Accuracy varies by site, depending on traffic volume, consent rates, and how representative the consented user group is. Google does not publish confidence intervals for individual accounts.
Can small websites use conversion modelling?
GA4 behavioural modelling requires at least 1,000 daily events from users who denied consent and 1,000 daily users who granted consent, sustained over 7 days. Many small websites do not meet these thresholds. Google Ads conversion modelling requires 700 ad clicks over a 7-day window per country and domain grouping.
Is Advanced Consent Mode legal under GDPR?
The legality is debated. Advanced mode loads Google tags before consent and sends anonymised pings when consent is denied. Some privacy experts argue this conflicts with Article 5(3) of the ePrivacy Directive, which requires consent before accessing or storing information on a user's device. Others contend that anonymised, cookieless pings fall outside the Directive's scope. No DPA has issued a definitive ruling on this specific question.
What is the difference between Observed and Blended reporting in GA4?
Observed reporting shows only data from users who accepted analytics cookies. Blended reporting combines observed data with modelled estimates for users who denied consent, resulting in higher but less precise figures. The choice depends on whether you prioritise data completeness or data certainty.
Do modelled conversions affect Google Ads bidding?
Yes. Google Ads automated bidding strategies such as Target CPA and Target ROAS incorporate modelled conversions into their optimisation. This means your bid strategy is partly influenced by estimated, not directly measured, conversion data.
Take Control of Your Cookie Compliance
If you are not sure which cookies your site sets or whether your Consent Mode implementation is working correctly, start with a free scan. Kukie.io detects, categorises, and helps you manage every cookie - so your visitors get a clear choice, and you stay on the right side of the law.
