Cookies, consent and the ePrivacy mess: why selling into Germany and Austria needs a different tracking setup
TL;DR
Difficulty: Intermediate
- Since 5 February 2026 the UK allows first party analytics without consent, under strict conditions. Germany and Austria allow only consent or strictly necessary, nothing else.
- A valid consent is a clear affirmative act: two equal buttons on the first layer, no pre ticked boxes, nothing non essential fired before the press.
- Google consent mode v2 is a platform condition for EEA traffic, not a law. Modelling recovers part of what refusals hide, never all of it.
- Your ad account sees conversions times the consent rate. Hold it against your own database every month or you will cut Germany for pressing reject.
A shop in Brighton that sells into Germany and Austria has one website and two sets of cookie law, and since February 2026 they have pulled further apart. I go through what each market requires with the primary sources, draw what fires when a visitor accepts or refuses, and work through a swimwear shop's June to show in pounds how far measured and modelled conversions sit from the truth.
The point: one checkout, three markets, two rulebooks
A shop in Brighton that sells into Germany and Austria has one website and two sets of cookie law. The British rules moved on 5 February 2026, when the Data (Use and Access) Act 2025 rewrote the cookie regulation and opened a door for first party analytics without consent.1 The German and Austrian rules did not move. There, anything that is not strictly necessary waits for a yes, and the regulators have spent years spelling out what a yes looks like.23
That gap is the whole article. Build your tracking for the British rule and point it at German traffic, and you are breaking the law in your second biggest market. Build it for the German rule everywhere, and you give up something the UK now lets you keep. Either way, the numbers in your ad accounts will be wrong by a predictable amount, which arithmetic fixes and a new tool does not.
I go through the law in each market with the primary sources, show what a valid consent looks like, draw what fires when a visitor says yes or no, and work through a Brighton swimwear shop's June to put measured against modelled conversions in pounds. The business numbers are made up. The law is not.
What the UK law says now
The British rule is regulation 6 of the Privacy and Electronic Communications Regulations, PECR for short. Since 5 February 2026 it says, in substance: subject to Schedule A1, a person must not store information, or gain access to information stored, in the terminal equipment of a subscriber or user.1 Terminal equipment is your visitor's phone or laptop. The rule is about storing and reading, not cookies as such: a pixel, a fingerprint, local storage and a cookie are all the same thing to it.
Schedule A1 holds the exceptions, and that is where the change lives.4 There are two old ones and two new ones that matter to a shop:
The consent exception: you tell the visitor clearly and comprehensively what the storage or access is for, and they consent. Consent here means consent in the UK GDPR sense, which I come to below.
The strictly necessary exception: storage or access essential to a service the visitor asked for, now with examples in the schedule itself: security, fraud and fault detection, authentication, recording the visitor's selections.4
The statistical purposes exception, new in 2026: no consent is needed if the sole purpose is collecting statistics on how the site is used in order to improve it, the information is not shared with anyone except to help with that improvement, the visitor is given clear information, and the visitor is given a simple, free means of objecting and does not object.4 The Information Commission (the regulator was the ICO until September 2026, and its website still uses that name) adds that the results must be aggregate data that is not personal data, that the exception does not let you monitor or track individual visitors, and that it does not apply to anything related to online advertising.5
The appearance and functionality exception, also new: adapting the look or the functions of the site to the visitor's preference, again with information and a simple means to object. The ICO is explicit that this is not about adapting content to a person's interests or behaviour.5
Two ICO lines are worth pinning to the wall. Legitimate interests is not a route round regulation 6: anything that fits no exception needs consent.6 And online advertising has never been exempt and still is not,5 whatever the ICO's separate advertising review brings one day.7
The penalty changed too: the maximum PECR fine rose from £500,000 to £17.5 million or 4% of worldwide annual turnover.8 The ICO says 99% of the top 1,000 UK websites now meet its banner standard.7 In my experience the long tail of smaller shops is nowhere near that.
What the law says in Germany and Austria
Both countries implement the same European rule, article 5(3) of the ePrivacy Directive 2002/58/EC as amended in 2009. Storing information, or gaining access to information already stored, in a user's terminal equipment is allowed only if the user has consented, having been given clear and comprehensive information. The only carve outs are storage or access for the sole purpose of transmitting a communication, and storage or access strictly necessary for a service the user has explicitly requested.9 There is no statistical exception in that sentence, and neither country has invented one.
Germany: section 25 of the TDDDG. Paragraph 1 says storage in, and access to, the end user's device is only permitted if the end user has consented on the basis of clear and comprehensive information, and that the information and the consent must follow the GDPR. Paragraph 2 lists the two exceptions: transmission of a message, and storage or access that is strictly necessary so that the provider of a digital service can provide a service the user explicitly requested.10 Until May 2024 the law was called the TTDSG. The name changed, the section number did not, and the German regulators say the new section 25 matches the old one in substance.2
The German regulators' joint guidance, the Orientierungshilfe Digitale Dienste of the Datenschutzkonferenz, last updated in November 2024, reads that section with a straight face. The legislator kept section 25(2) deliberately close to the European wording and added nothing. The necessity test is technical, a cookie cannot become necessary because the business model needs it, and even simple visitor counting is not automatically part of the service the visitor asked for.2 The guidance refuses to bless "reach measurement" as a category at all: the term is too vague and the answer depends on the exact configuration and purpose.2
Austria: section 165(3) of the Telekommunikationsgesetz 2021. Providers must tell the user what personal data they will collect, on what basis, for what purposes and for how long, and collection is only permitted once the user has actively consented on the basis of clear and comprehensive information. The same two exceptions follow: transmission, and what is strictly necessary for a service the user explicitly requested.11 The Austrian regulator, the Datenschutzbehörde, summarises it in its cookie FAQ: for every technically unnecessary cookie, consent must be obtained, and pre set or pre ticked boxes in the banner are not permitted.3
The German fine for breaching section 25, under section 28 of the TDDDG, tops out at EUR 300,000.12 That sounds mild next to the British number until you remember who brought the Planet49 case to the Court of Justice: a federation of consumer organisations, not a regulator.13
What strictly necessary means, and what it never means
The phrase does the heavy lifting in all three countries. The ICO's test: the storage or access must be essential to the service the visitor requests, so that without it the service could not technically be provided, judged from the visitor's point of view, not yours.56 The German guidance says the same.2
What passes, in all three markets: a session cookie that keeps the basket together once the visitor has put something in it, the cookie that authenticates a logged in customer, a load balancing cookie, security cookies that detect repeated failed logins, and the cookie that remembers the visitor's consent choice itself.52 The German guidance adds a useful subtlety: the basket and payment cookies become necessary when the visitor actually puts something in the basket or starts paying, not while they are merely browsing, and a language or colour preference needs no unique identifier at all, just the value "language: de".2
What does not pass, anywhere: analytics that identify or follow individual visitors, advertising, remarketing, social media plugins that track, and anything that shares data across sites or devices.52 The Austrian FAQ lists the technically unnecessary services as the ones that need consent and names the basket as the classic necessary one.3
An example: Sealane Swimwear in Brighton, which I use throughout, has a basket, a customer login, a consent banner, Google Analytics 4, Google Ads conversion tracking and a Meta pixel. Before any press on the banner it may set the basket cookie once something is in the basket, the login cookie once someone logs in, and the cookie that stores the consent choice. Everything else waits. In the UK, Sealane could also run a self hosted analytics tool without consent under the statistical exception, provided it shares the data with nobody, keeps the output aggregate, says so plainly and offers an opt out. I do not see how Google Analytics 4 fits, because the data goes to Google and feeds advertising. In Germany and Austria the question does not arise: analytics waits for a yes.
What a valid consent looks like
All three markets borrow the consent definition from the GDPR, article 4(11): a freely given, specific, informed and unambiguous indication of the person's wishes, by a statement or by a clear affirmative action.14 Recital 32 adds the sentence that killed a decade of banner design: silence, pre ticked boxes or inactivity do not constitute consent.14 Article 7(3) adds that it must be as easy to withdraw consent as to give it.14
The Court of Justice applied that to cookies in Planet49, case C 673/17, on 1 October 2019. A German company ran a promotional lottery with a pre ticked box consenting to advertising cookies. The Court held that consent is not validly given by way of a pre checked box the user must deselect to refuse, that this is true whether or not the information in the cookie is personal data, and that the visitor must be told how long the cookies run and whether third parties can read them.13
The European Data Protection Board's consent guidelines of May 2020 close the remaining gaps. Access to a site must not be made conditional on consent to cookies, the so called cookie wall, because that consent is not freely given. Scrolling or swiping through a page will not under any circumstances count as a clear affirmative action.15
And then the regulators on the reject button, which is where most banners still fail. The ICO's checklist: visitors can reject non exempt storage and access as easily as they can accept it, and a banner that only offers "continue to use our website" is not valid consent even with an OK button next to it.16 The German guidance: the option to refuse must be clearly recognisable as an equal alternative to the option to consent, which it says is the case when, for example, a button comparable in size, colour, contrast and typeface sits next to the consent button, and a "Settings or reject" button that leads to a second layer is not enough.2 The Austrian FAQ: not giving consent must be just as easy as giving it, and the visitor cannot be made to find the refusal on a second or third layer.3
So the banner that is legal in all three markets is dull: two buttons of the same weight on the first layer, no pre ticked purposes, the purposes and third parties named, nothing non essential set until a button is pressed, and a footer link that reopens the banner. How a nudge becomes a liability is in my piece on dark patterns in consent design.
Google consent mode v2 is a platform rule, not a law
Consent mode is Google's mechanism for telling its tags what the visitor decided. The tags read four signals: ad_storage and analytics_storage for cookies, and since late 2023 ad_user_data for sending user data to Google for advertising and ad_personalization for personalised advertising. The last two are what people mean by "v2".17
Google's EU user consent policy requires legally valid consent for cookies or local storage where the law requires it, and for the use of personal data for ad personalisation, from users in the European Economic Area, the UK and Switzerland, with records kept.18 For EEA traffic Google tied the policy to features: to keep using its tags for measurement, ad personalisation and remarketing, you must collect consent and share the signals with Google.19 That is a contract condition, not a statute, but for a shop selling into Germany and Austria it has the same effect.
Consent mode comes in two shapes.17 Basic: the Google tags do not load until the visitor has interacted with the banner, and if they refuse, no data goes to Google at all. Advanced: the tags load with every signal defaulting to denied, send cookieless pings while consent is denied, and send full measurement only once the visitor grants it. The advanced shape is what gives Google's modelling its own material on the visitors who refused.
For German and Austrian traffic, that difference is a legal one. The advanced shape loads Google's script before the visitor has decided, and while consent is denied the script has the browser send pings with a timestamp, the user agent and the referrer to Google.17 The European Data Protection Board reads the European rule as covering exactly that: JavaScript that instructs the browser to send information clearly falls within it, and collecting header data such as the user agent can bring it into play.20 The German guidance says that while the banner is showing, no scripts that may access the device are loaded where that access needs consent.2 Section 25 TDDDG has only its two exceptions, and modelling Google's ad conversions is neither the transmission of a message nor a service the visitor asked for.10 Austria's section 165(3) implements the same European rule with the same two exceptions, so the same EDPB reading applies there.11 I found no German regulator statement that names cookieless pings, either way. So for German and Austrian visitors basic mode is my default, and advanced mode only if the shop's own lawyer sets out in writing why it is lawful. In basic mode Google still models refused conversions, with a general model rather than one specific to your account.17 That brings us to the numbers.
What fires when: the flow
Here is the sequence I build for a shop in all three markets. The first layer is identical everywhere. The branches differ in what the shop keeps for itself.
Read it top to bottom. The first two boxes happen before anyone touches anything, and they are where most sites fail: the tag manager loads, the pixels fire, and the banner is decoration. The accept branch is ordinary tracking. The reject branch is where the work is, and my rule for it comes below. Both branches land in one report, which has to say which numbers were observed, which were modelled and which came from the database. The engineering of event tracking under consent mode is its own article.
Measured versus modelled: the Brighton numbers
Now the arithmetic, because this is where consent stops being a legal question and becomes a budget question.
Worked example: Sealane Swimwear's June in numbers
Sealane Swimwear is an invented shop in Brighton run by Maya. It sells to the UK, Germany and Austria, in that order of size. In June it bought 20,000 Google Ads clicks at £0.40 each, so £8,000 of spend: 10,000 clicks and £4,000 in the UK, 8,000 clicks and £3,200 in Germany, 2,000 clicks and £800 in Austria. Maya's own database, which matches orders to the landing session, says those clicks produced 400 orders: 200 in the UK, 160 in Germany, 40 in Austria. The average order was £65, so £26,000 of revenue from £8,000 of ads. Every one of those numbers is made up for the example.
Now the banner. For the example I assume 60% of British visitors press accept and 45% of German and Austrian visitors do, and I come back to those assumptions below. For the first pass I also assume refusers buy at the same rate as accepters, and then I remove that assumption.
Google Ads can only see a conversion when the click id survived in a cookie it was allowed to set. So the conversions it observes are the true conversions multiplied by the consent rate:
Here is the number of conversions the ad platform observes, is the true number of conversions from those clicks, and is the consent rate, the share of visitors who accepted. This is a plain rate applied to a count, so if you want the groundwork it is in Foundations: ratios and rates.
Market by market: UK, . Germany, . Austria, . Observed total: 210 of 400 orders, which at £65 is £13,650 of the £26,000 that actually happened.
Google's modelling fills part of the gap. Its help page says modelling estimates the missing ad to conversion paths from the journeys of consenting users and from historical trends, deliberately errs on the side of under prediction, needs the cookieless pings to calibrate, and only switches on once an account passes 700 ad clicks a day over seven days per country and domain grouping.21 It gives one illustration: at a 50% consent rate, modelling produced an 18% uplift in reported conversions.21 For the example I assume a 20% uplift on the observed figure. UK, . Germany, . Austria, . Modelled total: 252 orders, £16,380.
The bars are revenue in pounds from the same 20,000 clicks, counted three ways: what Google Ads shows with consent mode alone, what it shows once modelling kicks in, and what Maya's database knows. The platform is not lying. It reports what it was allowed to see plus a cautious estimate, and even then the account shows about 63% of the real revenue. Nobody at the platform will ring Maya to tell her.
Now watch what that does to a decision. Return on ad spend is revenue divided by spend:
is the return on ad spend as a plain multiple, is the revenue the clicks produced and is what the clicks cost. The maths is the same ratio as before, so Foundations: ratios and rates covers it, and the reason the average ratio is the wrong one to scale a budget on is in Foundations: marginal versus average.
Actual ROAS is the same in all three markets by construction: UK , Germany , Austria . Observed ROAS is not. UK . Germany . Austria . If Maya has told Google to chase a target ROAS of 2.0, or if her agency reads the dashboard on a Monday morning, Germany gets cut, not because German customers buy less but because they press reject more often. That is the most expensive consent mistake I see, and it is invisible unless you hold the platform's number against your own.
Correcting for the bias, and why you still need your own database
Now remove the simplifying assumption. Google's own page says consented users are typically two to five times more likely to convert than unconsented ones.21 Part of that is real: someone who presses accept is more engaged. Part of it is an artefact: some of the "unconsented" conversions are simply not visible. Either way, dividing the observed number by the consent rate overstates the truth. If consenting visitors convert times as often as refusing visitors, the true count is:
is the true number of conversions, the observed number, the consent rate and the ratio of the consenters' conversion rate to the refusers' conversion rate. With the bracket collapses to and you are back to dividing by the consent rate. The derivation is two conditional rates added together, which is Foundations: probability territory.
Germany, observed 72, . With : , the true figure in the example. With : . So the honest estimate from the platform's side is "somewhere between 116 and 160 German orders", which is a wide range to run a budget on. Maya's database says 160, and the database does not need an estimate because it counted.
Server side tracking without consent: my renaming rule
A visitor who presses reject still places an order. The order is a contract between them and the shop, the shop records it in its own database under the GDPR's contract basis, and no cookie is read or written to do so.14 That record is the shop's first party truth and it is not a regulation 6 matter, because nothing is stored on or read from the device beyond the strictly necessary session.
The question is what, if anything, the shop sends onward from that record. Here is the rule I apply, and I state it as my practice rather than as a reading of the law. Every event from a visitor who refused is renamed out of retargeting. The consented order goes to the platforms as the standard event, purchase to GA4, the conversion action to Google Ads with the click id, Purchase to Meta with the matching identifiers. The refused order goes as purchase_nc, stripped of every identifier: no click id, no cookie id, no hashed email, no user agent beyond what the transport needs. That event name is never mapped to a conversion action the platform optimises on, and it is never eligible for an audience. It exists so the total in the report matches the database, and for nothing else.
Why a rename rather than a quiet toggle: the people who run ad accounts change, and a named event is harder to misuse by accident than a flag someone forgot. Nothing from a person who said no should ever become a retargeting audience, and the name enforces that when I am not in the room. The policy backdrop is Google's own: ad personalisation requires consent, and records of it.18 An event with no identifier has nothing to personalise with.
For Germany and Austria I go a step further and keep the refused stream inside my own systems. In January 2022 the Austrian Datenschutzbehörde, deciding the first of the 101 complaints noyb filed after Schrems II, found that a site's use of Google Analytics sent personal data to Google in the United States without an adequate basis.22 The EU US Data Privacy Framework has since restored a transfer ground to certified US companies.23 The regulators' appetite for data about refusing visitors landing at an American advertising company has not changed with it. So in those markets the refused orders stay in the shop's database and its own reporting, and the platforms see nothing of them, because in basic mode no tag fires after a no. The plumbing is in the technical guide to server side tracking.
Consent rates: what to expect, and why the number lies a little
People ask me for the consent rate they should plan on, and the honest answer starts with a definition. Commanders Act's 2025 privacy barometer, which looked at 1,301 banners on sites around the world in May 2025, reports an average consent rate of 80.54%. But it defines consent rate as the share of visitors who accepted among those who made a choice, and it reports separately that 32.41% of visitors made no choice at all. The share who consented out of all visitors, which the report calls opt in, was 55.03%.24 Same banners, same month, and the headline number is 25 points higher than the useful one.
For a shop, the useful one is the opt in: the share of all visitors whose conversions the platform can see. That is the in the formulas above. The visitors who made no choice are refusers as far as your ad account is concerned, because nothing non essential fired. So when a vendor quotes you a consent rate, ask which denominator, and ask whether the banner behind the number had an equal reject button on the first layer. A high consent rate is sometimes a well designed banner and sometimes an illegal one.
That is why the example used 60% and 45%: assumptions the published figures make plausible, not measurements. The right number for your shop is the one your own banner produces, measured as opt in over all visitors, market by market, and fed into the correction above every month. Where consent is logged properly, that is one query.
What I would build, in order
If Sealane came to me, the order of work would be this, and it is the same for any shop selling from Britain into the German speaking markets.
First, the banner: two equal buttons on the first layer, purposes and third parties named, nothing non essential before a press, a footer link to change the choice, a consent log. One banner for all three markets, because the German and Austrian standard is the strict one and a British visitor loses nothing by meeting it.
Second, the exceptions, per market. In the UK, decide whether a self hosted, unshared, aggregate analytics tool is worth running under the statistical exception, with its notice and opt out. In Germany and Austria, do not. Keep the strictly necessary list short and written down, with the reason each item is on it.
Third, consent mode v2, in the basic shape by default: no Google tag before a press, nothing to Google after a no. Advanced mode only if the shop's own lawyer has set out in writing why its cookieless pings are lawful in Germany and Austria. I run the same setup for British visitors rather than split the site in two.
Fourth, the shop's own measurement: orders matched to landing sessions, a monthly reconciliation of platform conversions against database conversions per market, and the correction formula applied so nobody cuts Germany for pressing reject. On Solidus the order record already carries what is needed.
Fifth, the renaming rule, in the code and on one page the next marketer will read.
The old version of this article said that in the German speaking markets compliance is a trust signal. I still think so, and I still cannot prove it with a number, so it stays an opinion. What I can prove is the arithmetic: a shop that measures against its own database makes better budget decisions than one that reads the dashboard, and in Germany and Austria the gap between the two is wider still than in Britain, because the unobserved share there is 55% rather than 40%, about a third more. That is reason enough.