Privacy & GDPR

EDPB Anonymisation Guidelines: When Is Analytics Data Anonymous?

The EDPB anonymisation guidelines set three tests: no record isolation, no linkage, no inference. What they mean for hashed IDs and web analytics.

By Jasper Koers Published 7 min read

On this page
  1. What the EDPB anonymisation guidelines say
  2. The three tests: record isolation, linkage, inference
  3. Hashed data, masked IPs and fingerprints: Example 6
  4. What this means for ClearAnalytics
  5. What to do with the guidelines now
  6. Where ClearAnalytics fits

At its plenary of 7 and 8 July 2026, the European Data Protection Board (EDPB) adopted Guidelines 02/2026 on anonymisation. They replace the anonymisation opinion of the old Article 29 Working Party from 2014 and give one answer to a question every analytics vendor likes to answer for itself: when is data really anonymous? The EDPB's answer is a three-part test. Data can safely be considered anonymous only if there is no record isolation, no linkage and no inference.

The EDPB announced the guidelines on 8 July, together with new guidelines on web scraping for generative AI and the final version of its blockchain guidelines. The anonymisation guidelines are open for public consultation until 30 October 2026. For web analytics, the most quotable part is a short example about fingerprints, and it closes a door that many "anonymous tracking" claims rely on.

Not legal advice. This article summarises a draft EDPB guidance document for website owners. Ask your DPO or a lawyer how it applies to your own processing.

What the EDPB anonymisation guidelines say

The starting point is the GDPR itself: anonymous data is information that does not relate to an identified or identifiable natural person, and the GDPR does not apply to it. The guidelines turn that into two questions:

  1. Does the information relate to a person, by its content, its purpose or its effect?
  2. Is that person identified or identifiable, using means reasonably likely to be used?

Three points in the legal analysis matter for anyone who runs analytics.

  • Identifying means distinguishing. The guidelines define identifying someone as distinguishing them from others in a given context so they can be treated differently. A name is not required.
  • The answer depends on who holds the data. Following the Court of Justice ruling in EDPS v SRB (C-413/23 P, 4 September 2025), data can be anonymous for one party and personal for another. But a processor working on behalf of a controller takes over the controller's perspective. If the data is personal for the website, it is personal for its analytics processor too.
  • Labels must be honest. The EDPB tells controllers not to use descriptions like "anonymous", "de-identified" or "de-personalised" if people are actually still identifiable.

The three tests: record isolation, linkage, inference

The core of the guidelines is a technical framework. It can be applied in a "simplified" way, which ignores differences between who could re-identify the data, or in a "contextual" way, which assesses each relevant party's actual means. Either way, the same three criteria apply.

Criterion When it is met What it looks like in analytics
No record isolation The data contains no unique combination of values that relates to a single person A raw event log with an ID per visitor, plus browser, OS, screen size and country, usually fails: many records are unique
No linkage No record can be linked to a record about the same person in another dataset A stable visitor ID that also shows up in an ad network's or CRM's data fails
No inference No specific and meaningful conclusion about one person can be drawn from the data A report that says "1 visitor from this town read the page about debt advice" can fail even though it is a count

If all three are met, the EDPB says the data "can be safely considered anonymous". If one fails, that is not automatically the end: further analysis is needed, in particular whether isolated records allow a person to be singled out. The guidelines give the example of a survey with thirty questions where every record is unique, but nothing in the answers can be traced back to a respondent.

Two details are easy to miss:

  • Aggregates are not a free pass. The EDPB notes that aggregate data "will usually" meet the no record isolation test, but it spends several examples on inference from aggregates: subtracting one salary total from another, or a survey result where every answer was low. In analytics terms, a report with a count of one is not anonymous just because it is a table.
  • Anonymity can expire. Re-identification techniques improve and more data becomes available, so the EDPB calls it good practice to reassess anonymised data periodically.

Hashed data, masked IPs and fingerprints: Example 6

Paragraph 25 of the guidelines explains that identifiers do not have to be about the person directly. A device's hardware identifier can identify the person using it, and a website can "fingerprint" visitors from their browser, operating system, screen resolution and time zone, and use that to recognise them from page to page and across visits.

Example 6 then describes a setup that is common in ad tech. Several websites use the same advertising provider. They do not share names, IP addresses or emails. Instead, they send the pages visited along with a pseudonym based on device fingerprints. The provider uses those pseudonyms to link the visits across sites, profile the users and pick ads.

The EDPB's conclusion: both the raw fingerprint and the pseudonym derived from it identify the individual, for the websites and for the third party. The provider never sees a name, but it can, and does, distinguish the person and treat them differently.

That has direct consequences for common "privacy-friendly" claims:

  • Masking the last part of an IP address reduces precision, but if the masked IP is combined with a user agent into a stable identifier, the result can still single out a visitor.
  • Hashing a fingerprint or an IP does not make it anonymous. A hash is a pseudonym. The guidelines list "pseudonyms that can easily be reverse engineered" as a simple case of data that is not anonymous, and an unsalted hash of an IPv4 address can be reversed by trying all of the roughly 4.3 billion possible addresses.
  • Removing names and emails is not enough when the dataset still lets you follow one visitor around.

This is also the practical difference in pseudonymisation vs anonymisation. Recital 26 GDPR says pseudonymised data that can be attributed to a person with additional information is still personal data. Pseudonymisation is a security measure that lowers risk. It does not take you out of the GDPR.

What this means for ClearAnalytics

We should apply the same yardstick to ourselves. ClearAnalytics counts unique visitors with a hash of a truncated IP address, the user agent and a salt that changes every day. The hash is a pseudonymous identifier for one day. During that day it lets us recognise that two pageviews on the same site came from the same browser; that is exactly why it works for counting unique visitors. Measured against the EDPB's tests, we do not claim that this hash is anonymous data.

What we can say is what reduces the risk, and how that maps onto the three criteria:

  • The IP address is not stored. It is used to compute the hash and then discarded.
  • The identifier rotates daily. A visitor gets a new hash tomorrow, so the pageview records cannot be used to follow someone across days or build a long-term profile. That limits linkage over time.
  • No cross-site linking. Each site's data is kept separately, the hash is not shared with anyone, and visitor data is not used for advertising. That is the opposite of the Example 6 setup.
  • Nothing is stored on the device. No cookies and no localStorage.
  • The reports are aggregated. What you see in the dashboard are counts of pages, sources, countries and devices, not visitor profiles.

The pageview records behind those reports do carry the daily hash, so they are pseudonymous data, processed in the EU on the site owner's behalf. The aggregated reports are where the anonymisation question really sits, and the inference test is the one to watch: tiny counts on sensitive pages deserve care, whatever tool you use. We explain the counting method in detail in how we count unique visitors without cookies.

What to do with the guidelines now

The guidelines are a consultation draft, so details can still change before the EDPB adopts a final version. The direction is clear enough to act on:

  1. Check your vendor's wording. If a tool says it collects "anonymous data", ask which of the three criteria its raw data passes, and from whose perspective. Our GDPR analytics checklist has the follow-up questions your DPO will ask.
  2. Map identifiers. List every ID your analytics stack sets or derives: cookies, user IDs, hashes, fingerprints. Note how long each one lives and who else receives it.
  3. Look at linkage first. A stable identifier that also reaches an ad network or a data broker is the Example 6 pattern. That is where the risk is highest.
  4. Watch small numbers. Reports filtered down to one or two visitors can fail the no inference test, even if every row is a count.
  5. Comment if it affects you. The public consultation runs until 30 October 2026.

Where ClearAnalytics fits

If your goal is to measure traffic without building visitor profiles, the guidelines reward short-lived identifiers, no third-party sharing and aggregated output. ClearAnalytics is designed around those choices: no cookies, a daily-rotating hash instead of a persistent ID, no IP storage and EU hosting. See the features for what it measures with that data.

Frequently asked questions

What are the EDPB anonymisation guidelines?

Guidelines 02/2026 on anonymisation, adopted by the European Data Protection Board on 7 July 2026. They explain when data no longer relates to an identified or identifiable person under the GDPR and give a three-criteria framework for testing it. They are open for public consultation until 30 October 2026.

Is hashed data personal data under the GDPR?

Usually, yes. A hash of an IP address, email or fingerprint is a pseudonym, which can still single out one person and link their records. The EDPB guidelines use a fingerprint-based pseudonym as an example of data that still identifies individuals.

What is the difference between pseudonymisation and anonymisation?

Pseudonymised data replaces direct identifiers with a code but can still be attributed to a person, so the GDPR still applies. Anonymous data no longer relates to an identifiable person from the relevant perspective, so the GDPR no longer applies to it.

Is aggregated analytics data anonymous?

Often, but not automatically. The EDPB notes that aggregate data usually passes the no record isolation test, but small counts and combinations of reports can still reveal information about one person.

  • edpb
  • anonymisation
  • pseudonymisation
  • gdpr
  • web-analytics

Back to blog Read as markdown

Enjoyed this article?

Try ClearAnalytics for free and get privacy-first analytics for your website.