Consent mode and conversion data: what an incomplete implementation actually costs you

Updated: 29 July 2026

Short answer: you are not losing data, you are losing bid quality

The most common misunderstanding about consent mode is treating it as a legal task. It is a performance task. Smart Bidding and Performance Max learn from conversion signals. When some of those signals never arrive, the algorithm optimises against an impoverished picture and bids up the wrong users. The symptom is a rising cost per acquisition at a stable cost per click, which is exactly where nobody looks for the cause.

The gap is measurable. With a basic implementation, reported conversions fall by as much as 20 to 40% compared with advanced mode, per Google’s own 2024 figures. Practitioners report that a correct advanced implementation preserves up to 65% of the conversions that would be invisible in basic mode.

Three things to check today if you are unsure about your own setup: which mode consent mode is running in, whether behavioural modelling is active at all, and whether the CMP integration has quietly stopped working since the last update. That third point is the most common silent failure.

What consent mode is, and what it is not

Consent mode is a communication protocol between the consent management platform (CMP) on your site and Google’s tags. Its job is to tell the tags what the user agreed to, and to adjust tag behaviour accordingly.

Version two added two parameters: ad_user_data, consent to send data to Google, and ad_personalization, consent for remarketing. Without them the implementation does not meet current Google and DMA requirements.

What consent mode is not, since this is where most of the myths sit:

• It does not replace your consent banner. It is an intermediate layer between the cookie banner and Google’s tags. You still need a certified CMP that displays the banner and collects consent.
• It does not bypass GDPR. Cookieless pings in advanced mode contain no personal data. They are anonymous signals without identifiers, which Google uses statistically rather than individually.
• It is not optional inside the EEA. Consent Mode v2 has been mandatory since March 2024, and since July 2025 Google has been actively disabling features on non-compliant accounts.

Basic mode versus advanced mode

This is the single decision with the largest effect on campaign performance, and it is most often made by accident during plugin configuration.

There is one thing many write-ups skip: modelling is not automatic and is not available to everyone. The thresholds are specific. Behavioural modelling requires a minimum of 1,000 daily events from consent-refusing users over 7 days, plus a minimum of 1,000 daily events from consenting users, which at a 50% consent rate means roughly 2,000 unique daily visitors. Smaller sites may not qualify for behavioural modelling at all.

If your site runs a few hundred sessions a day, the promise that “modelling will recover your data” is untrue in your specific case. Knowing this before you decide changes the whole measurement strategy: at low volume the answer is server-side measurement and offline conversion import, not modelling.

What happens to the numbers after implementation

The most common call to an agency goes: “we implemented it as recommended and conversions dropped.” Usually they did not drop. The definition of what gets counted changed.

What to expect:

• GA4 sessions and conversions shift. After implementation, session and conversion counts in GA4 can fall by 15 to 40% versus pre-implementation data, driven by the elimination of double counting and the absence of tags on refusal.
• Conversion rate can fall without any performance deterioration. A 3.2% rate can drop to 2.1% after moving to advanced mode, not because the store performs worse, but because the denominator, the session count, grew by modelled visits.
• ROAS can look lower. This applies where campaigns were previously being optimised against undercounted conversions.
• The conversion paths report will stop reconciling with everything else. GA4’s conversion paths report shows observed data only, and modelled conversions are not included there.

That last point is a frequent source of conflict between the PPC team and the analyst, because two reports in the same tool show different numbers and both are correct.

When modelling should not be trusted

An honest implementation includes where the method fails. Observations from agency portfolios suggest modelling performs well at high volume, above 1,000 conversions per month, with stable behavioural patterns, while for niche or seasonal campaigns, or after a sudden change in the offer, the model can over- or under-state conversions by 20 to 40%.

Attribution is especially fragile. When a user arrives via a paid ad and converts a week later through a direct visit, without cookies the model can assign the conversion to the wrong channel. That feeds straight into budget decisions, because a channel that looks unprofitable may simply be losing attribution.

This is an observation from a single agency portfolio rather than independent research, so treat it as directional. The direction is nonetheless consistent with how the method is built: statistical models work better on large, stable samples.

Implementation audit: seven checks

Ordered cheapest to most expensive. The first three take under twenty minutes combined.

1. Confirm the mode. In your CMP and GTM configuration, establish whether you are running basic or advanced. If nobody in the company knows, it is usually basic, set as the plugin default.
2. Confirm the pings are going out. On refusal you should see requests to google-analytics.com/g/collect without the _ga parameter. If they are absent, advanced mode is not working regardless of what the plugin dashboard shows.
3. Confirm modelling is switched on. In GA4 go to Admin, Data Settings, Data Collection, and verify that Google signals and conversion modelling are active.
4. Estimate the scale of loss. Compare GA4 session counts against unique server requests from logs over the same period, since the percentage difference approximates data loss. A reference point: typically 30 to 50% for B2C sites with high mobile traffic.
5. Reconcile Google Ads conversions against transactions in your commercial system. A gap above 15% needs explaining. A gap above 40% usually means the problem sits in the consent layer, not in attribution.
6. Check whether Enhanced Conversions are implemented. The platform recommendation is deploying both mechanisms together, Consent Mode v2 alongside Enhanced Conversions.
7. Set up a recurring regression test. The most common cause of silent data loss is not a bad implementation but its decay over time: CMPs update, GTM changes, and nobody checks whether the integration still works, so a silent failure means silent data loss for weeks.

Steps 1 to 3 can be run by a PPC specialist alone. Steps 4 to 7 require analytics access and are usually part of a Google Analytics and GTM audit, since they touch configuration outside the ads account.

Verification is not immediate. Implementation takes 2 to 4 hours for a standard GTM site, but verifying it works should run at least 7 days, since that is how long Google needs to begin filling gaps with modelled conversions.

What is changing in the law, and what you should not implement yet

Care is needed here, because this is an area where a legislative proposal is easily mistaken for binding law.

The position as of July 2026:

• The Digital Omnibus package was published on 19 November 2025 and proposes pulling cookie and tracking rules out of the ageing ePrivacy Directive and folding them into the GDPR itself, retiring the long-stalled ePrivacy Regulation.
• The mechanism is two new articles. Article 88a restructures how consent for terminal equipment access works, and Article 88b makes browser-level consent signals legally binding on controllers. Article 88a applies six months after entry into force.
• The cookie portion is not adopted. The GDPR and cookie portion remains a proposal under negotiation, and final text is expected by late 2026 or early 2027, with current obligations remaining fully in force in the meantime. Further, the Council’s own text currently drops the core cookie-consent provisions, so the direction of travel is not settled.
• The AI portion is further along. The AI Omnibus, postponing AI Act high-risk deadlines, was adopted by the Council on 29 June 2026 and takes effect in July.

A methodological caveat: sources give conflicting dates for the formal withdrawal of the ePrivacy Regulation, some citing February 2025 and others February 2026. The discrepancy does not change the practical conclusion, but verify the date at the primary source, meaning European Commission communications, before using it in any legal document.

The practical takeaway: the most common error is acting on proposals as if they were law, when the cookie and breach changes have not been adopted. Build against today’s rules and prepare for tomorrow’s. The one change worth preparing for early is readiness for machine-readable consent signals, because that preparation is neutral to how the negotiation resolves.

If the consent layer is a deferred project in your organisation, treat Consent Mode implementation as a precondition for campaign performance rather than a legal department task.

Frequently asked questions

Does consent mode let me track users who refused?
No. On refusal, tags do not write cookies and do not identify the user. What is sent are anonymous signals without identifiers, used statistically for aggregate-level modelling.

How many conversions will modelling recover?
It depends on volume. Published estimates cite recovery of a substantial share of events invisible in basic mode, but all of them assume the volume thresholds are met. Below those thresholds modelling never activates, and no percentage is the right answer.

Why do my Google Ads and GA4 conversion numbers disagree?
This is normal and stems from different attribution windows, different attribution models, and the fact that modelled conversions enter different reports in different ways. A gap in the low tens of percent is expected. A gap of several tens of percent warrants an audit.

Is installing a CMP plugin enough?
No. The plugin displays the banner and collects consent, but it must be correctly integrated with GTM, and advanced mode has to be enabled deliberately. The presence of a banner is not evidence of a working implementation.

Should I wait for the Digital Omnibus before implementing?
No. Current obligations apply in full, and Google has been disabling advertising features on non-compliant accounts since July 2025. Waiting means real data and feature loss today in exchange for an uncertain regulatory benefit later.

Sources

• SSAK Agency, Consent Mode V2 guide 2026, March 2026 → https://ssak-agency.pl/consent-mode-v2-poradnik/
• Artur Smolicki, Consent Mode v2 in Google Ads: GDPR and privacy, 2026 guide, March 2026 → https://artursmolicki.com/blog/consent-mode-v2-w-google-ads-rodo-i-prywatnosc/
• PremiumAds, Consent Mode v2 and Analytics: before and after comparison, May 2026 → https://premiumads.pl/premiumblog/consent-mode-v2-a-analytics-porownanie-danych-przed-i-po/
• PremiumAds, Consent Mode v2 and Analytics: what changes in the data, May 2026 → https://premiumads.pl/premiumblog/consent-mode-v2-a-analytics-co-zmienia-sie-w-danych/
• PremiumAds, GA4 and Consent Mode v2: how user consent changes reports, May 2026 → https://premiumads.pl/premiumblog/ga4-a-consent-mode-v2-jak-zgoda-uzytkownika-zmienia-raporty/
• Secure Privacy, EU Digital Omnibus: What Article 88a Changes for Cookie Consent (2026), July 2026 → https://secureprivacy.ai/blog/eu-digital-omnibus-what-article-88a-changes-for-cookie-consent-2026
• PrivacyForge, EU Digital Omnibus: 2026 GDPR & AI Act Changes, June 2026 → https://privacyforge.io/resources/blog/eu-digital-omnibus-gdpr-ai-act-2026
• Usercentrics, What To Know About The EU’s Digital Omnibus Package, February 2026 → https://usercentrics.com/knowledge-hub/eu-digital-omnibus-package/
• Biscotti CMP, ePrivacy Directive vs GDPR: EU Cookie Rules Explained 2026, July 2026 → https://www.biscotti-cmp.com/en/blog/the-eprivacy-directive-vs-gdpr-navigating-the-eus-cookie-rules

Data currency and limitations note. Several figures here, particularly the scale of data loss and modelling accuracy, come from agency write-ups based on their own campaign portfolios rather than independent research. Treat them as directional. The legislative status of the Digital Omnibus shifts during negotiation and must be verified before use in any document with legal consequences. This article reflects the position in July 2026 and requires quarterly review. It is not legal advice.