Consent mode comes with a reassuring story attached: users decline tracking, a gap appears in your data, and Google's modelling fills that gap with statistical estimates. The story is accurate. What is usually left out is that Google publishes specific volume thresholds the modelling requires, and many smaller sites will not reach them.
What Consent Mode Actually Does
Consent mode is a signalling layer. Instead of blocking Google tags outright when a user declines, it lets tags fire in a restricted state that sets no cookies and sends no identifiers, carrying instead flags describing what the user permitted.
Consent Mode v2 uses four parameters: ad_storage, analytics_storage, ad_user_data and ad_personalization. Each should default to denied before any tag fires and update when the user chooses. Mapping three of the four correctly produces partial data loss that will not necessarily be visible in surface-level reporting, which is the more dangerous failure mode, because everything appears normal.
Two Different Systems, Two Different Thresholds
A distinction worth getting right, because the two are frequently conflated: behavioural modelling in GA4 and conversion modelling in Google Ads are separate systems with separate requirements.
For behavioural modelling in GA4, Google's documentation states that a property needs to collect at least 1,000 events per day with analytics_storage set to denied for at least 7 days, and to have at least 1,000 daily users sending events with analytics_storage set to granted for at least 7 of the previous 28 days.
For conversion modelling in Google Ads, Google's documentation states a daily ad click threshold of 700 ad clicks over a 7 day period, per country and domain grouping, alongside a correct consent mode or TCF v2.0 implementation.
Google is also explicit that clearing these bars is necessary rather than sufficient. Its documentation notes that meeting the external prerequisites does not guarantee eligibility, because the underlying model applies further criteria - including the ratio of new to returning users and user-to-session counts - to maintain accuracy.
Why This Lands Differently in a Small Market
Greece has roughly 8.64 million internet users in total, according to DataReportal's Digital 2026 report. Measured against that ceiling, a site sustaining a thousand consenting users per day is not a typical local business site. It is a large publisher, a major retailer or a significant marketplace.
Most Greek eShops, B2B sites and service businesses operate well below that. The reasonable conclusion is one that rarely gets stated: for many advertisers in a market this size, modelling is not a safety net that underperforms - it is one that may never deploy at all.
This is where advice written for large markets travels badly. "Implement consent mode and modelling will recover most of what you lose" is reasonable guidance for a high-volume advertiser and misleading for a small one, for reasons that have nothing to do with implementation quality.
What You Get Instead, and Why It Is Manageable
Below the thresholds, your reporting shows observed conversions only. You are undercounting, roughly in proportion to the share of users who decline.
This deserves less alarm than it usually gets, on one condition: that you know it is happening.
An undercount you understand is still a usable measurement. Absolute numbers are wrong, but relative comparisons largely hold - campaign against campaign, month against month - provided your consent rate is stable. What damages decision-making is not the undercount but discovering it late, after months of budget decisions were made on numbers assumed to be complete.
Three things follow. Track your consent rate as a metric in its own right, since it acts as a multiplier on everything else you measure. Expect platform-reported conversions to sit below back-end reality, and reconcile against orders or CRM records rather than against the ad platform alone. And be careful comparing any period before your consent implementation with any period after it, because that comparison largely measures your banner rather than your marketing.
How This Site Is Set Up
This site runs Consent Mode v2 with all four parameters defaulting to denied before any tag fires, a banner on every page rather than only the homepage, and analytics loading only after an explicit accept.
I verified it the only way worth verifying anything: loading the site in a clean browser profile and checking what was actually set. Before accepting, no analytics cookies are present. After accepting, the expected GA4 cookies appear. The weather widget geolocates by IP only after consent and otherwise falls back to a fixed Athens location.
This site is nowhere near the modelling thresholds and never will be, which is rather the point. It is in the same position as most of the businesses reading this, and the sensible response was to make the measurement honest rather than to rely on a model that was never going to run.
The Compliance Side Is Not Theoretical
It is tempting to treat consent purely as a measurement inconvenience. Regulators across Europe have not treated it that way, and enforcement activity through 2025 and 2026 has included substantial penalties relating to cookie practices. Under Article 83(5) of the GDPR, the maximum is €20 million or 4 percent of worldwide annual turnover, whichever is higher.
The recurring theme in published decisions is the distance between a banner that looks compliant and one that behaves compliantly: tags firing before consent is given, decline options that only dismiss the banner without changing what loads, and interfaces where refusing is made harder than accepting. Nothing here is legal advice, and anyone with a specific concern should take proper advice - but a banner that hides itself while tags fire underneath is not a subtle distinction.
What to Do
- Verify the default state yourself. Open your site in a clean browser profile and inspect storage before interacting with the banner. If an analytics or advertising cookie already exists at that point, the implementation is not doing what you think.
- Confirm all four parameters. Three out of four produces silent partial loss. Check each individually.
- Check your volumes against the published thresholds. Compare honestly. If you are below them, plan around observed data rather than expecting a correction.
- Instrument the consent rate. It scales everything else in your analytics.
- Reconcile against the back end. Orders, leads and CRM records are the reality to check platform reporting against. Below the thresholds, this stops being good practice and becomes the main way to know what happened.
Consent mode is worth implementing properly, both because it is the compliant path and because it preserves more signal than blocking tags outright. Just do it with an accurate picture of what it will and will not recover at your size.
Sources
Google Analytics Help, behavioural modelling for consent mode eligibility requirements; Google Ads Help, about consent mode modeling; DataReportal, Digital 2026: Greece. Thresholds are as published by Google at the time of writing and may change - check the current documentation before relying on them. This article is not legal advice.