Blog

AI in online advertising creates no exemption from UK GDPR and PECR

The UK ICO states in AdTech 3.0 that LLMs and agentic AI create no new privacy regime for online advertising, but do require demonstrable documentation per activity.

· By

Wooden meeting table with loose colourful sticky notes and sketches on the left and a neat stack of stamped and signed papers with a pen on the right.
The ICO states that AI in online advertising grants no exemption but requires demonstrable per-activity documentation under the UK GDPR and PECR.Image: IamVera.ai — original editorial illustration

The UK regulator ICO states in its AdTech 3.0 report that language models and agentic AI create no separate privacy regime for online advertising: PECR, the UK GDPR and the Data Protection Act 2018 still apply. Organisations must record demonstrably per advertising activity which data, inferences, roles, legal basis, consent and controls they use.

On 13 November 2026 the Information Commissioner's Office (ICO) published the report AdTech 3.0: Online advertising in the age of LLMs and agentic AI. It examines how large language models (LLMs) and agentic AI — software that independently plans and carries out tasks — are changing the planning, buying, optimisation and presentation of online advertising. The core message is sober: the report describes no new set of data protection obligations. Our analysis is that existing rules must therefore be applied more precisely and better substantiated.

What exactly does the ICO state in the AdTech 3.0 report about AI and online advertising?

The ICO describes that AI is now being deployed or explored for setting campaign goals, audience selection, optimisation, coordination across advertising platforms and displaying advertisements alongside conversations in chat services. According to the regulator, existing obligations under PECR, the UK GDPR and the Data Protection Act 2018 continue to apply, as relevant to the processing activity. Organisations must reassess their roles, data use and processing activities.

What matters is what the ICO does not say: AI creates no exemption. The regulator states in its guidance on storage and access technologies that using those technologies for online advertising purposes requires consent, subject to applicable exceptions. An AI model steering the campaign does not change that requirement.

Why does AI change the way organisations must demonstrate consent and data use?

Our analysis is that richer inferences and agentic systems with less human intervention may make it harder to reconstruct afterwards which personal data led to which advertising decision. According to the report the rules remain the same; in our assessment what mainly increases as a result is the need for demonstrable substantiation. That is, in our assessment, the real shift in the report: from "is it allowed" to "can you demonstrate how exactly it happened".

In earlier supervisory signals the ICO showed that this line is broader. In its notice on agreements with AI developers the regulator describes that agents carry out tasks, use tools and access external systems with limited human control. Our analysis is that an advertising agent could, for example, independently combine audiences or shift budgets; for such actions the substantiation and authority limits must remain verifiable.

The context has been known for some time. In its advice to government on possible exceptions to the consent requirement the ICO pointed out that programmatic advertising often shares identifiable and sometimes sensitive information with many third parties. That is a policy proposal, not applicable law: Subject to any applicable statutory exception, the current PECR consent requirement remains in force. AI may, in our assessment, increase these supply-chain risks, because data is combined faster and in more places.

What does this mean for directors, lawyers and CISOs working with AI advertising?

Our analysis is that marketing and data leads deploying AI should strengthen their demonstrability. It is therefore wise, per advertising use case, to update a current overview or, where applicable, the record of processing activities with data, inferences and legal basis. Because using storage and access technologies for advertising requires consent subject to applicable exceptions, an AI integration that tracks silently affects the entire chain; we recommend that the lawyer re-test the consent and tracking logic before an AI function goes live, with an explicit fail-state when consent is absent. Our analysis is that agentic systems with limited human intervention can pose additional risks for oversight and complaint handling. We recommend that organisations record authority limits, logging and auditable data flows, insofar as that is appropriate and feasible, so that campaign decisions can be investigated more readily. Our analysis is that an AI advertising use case may, through richer inferences, carry a high risk to data subjects. We recommend that the controller assess whether a DPIA is required and carry it out before the campaign starts when an applicable high-risk threshold is reached; inferences about sensitive characteristics or vulnerable groups are relevant risk factors here.

As a summary of that analysis, these are the priorities:

  • Update the record of processing activities per AI advertising activity, with data, inferences, legal basis and parties involved.
  • Re-test the consent and tracking logic and ensure the chain stops when consent is absent.
  • Record the roles of platforms, intermediaries and suppliers: who is controller, who is processor.
  • Carry out a DPIA as soon as richer inferences make a high risk likely.
  • Bound the authority of agentic systems and retain auditable logs of campaign decisions.

Which documentation do organisations record per AI advertising activity?

Our analysis is that no new legal document is needed for this, but that organisations should document existing obligations coherently and demonstrably per processing activity. That means data inventory, roles, purposes, legal basis, consent, risk analysis and controls do not exist in isolation but align per use case.

A workable documentation set per AI advertising activity contains:

  1. which personal data is used and which inferences the model makes;
  2. which parties perform which role in the chain;
  3. which legal basis and which PECR consent apply;
  4. how audience and campaign decisions come about;
  5. which model or system changes have been made and when;
  6. what evidence there is of human or organisational control.

For those wanting to delve deeper into the underlying concepts, we have set out the rules and methods of pseudonymisation under the GDPR, as well as the question of whether token IDs from LLM inputs are personal data. Those working with independently acting systems will find concrete starting points in our analysis of authority limits for AI agents. For related background, see our hub on AI privacy and the GDPR.

Sources and references

  1. AdTech 3.0: Online advertising in the age of LLMs and agentic AIInformation Commissioner's Office · 2026-11-13
  2. How do the rules apply to online advertising?Information Commissioner's Office · 2026-10-15
  3. ICO advice to government on online advertising exception(s) to regulation 6 consent requirementsInformation Commissioner's Office · 2026-05-20
  4. ICO secures changes from leading AI developers as scrutiny extends to AI agentsInformation Commissioner's Office · 2026-10-08

Sources: The article relies on the ICO report AdTech 3.0 (13 November 2026) and on additional official ICO guidance and notices about online advertising, PECR consent and agentic AI.

← All articles in this topic ← All articles