Agentic AI ads keep the old rules, but labels and dashboards alone won't prove compliance

Resultsense

Source published: Site published:

Original article · فارسی

The main finding of the Information Commissioner’s Office (ICO) in AdTech 3.0: Online advertising in the age of LLMs and agentic AI, published on 13 November, is that agentic AI and large language models (LLMs) “don’t create a new set of data protection requirements for online advertising”. This was the main point of our news report about the 51-page document. But reading the full report shows that it asks companies to do more than the headline suggests. It covers seven areas, from accountability to risk assessments. It looks at the evidence an advertiser might use to show that an AI-run campaign follows the rules. It explains that no single piece of evidence may be enough. Examples include a performance dashboard, a supplier's label saying data is “aggregated”, a purpose described as campaign optimisation, a signal about a group of people, or an impact assessment approved before launch. The question for UK advertisers, agencies and publishers is whether they can get this evidence when they increasingly depend on suppliers to explain how these systems work.

Strategic Insight: The report clearly says that it “doesn’t create new legal requirements”. In our view, this makes its examples more useful, not less. They show how the regulator already applies PECR (the Privacy and Electronic Communications Regulations) and the UK GDPR to agentic systems. The findings will also help shape the ICO's upcoming guidance on automated decision-making (ADM) and agentic AI.

Why did the ICO write a report that changes no rules?

Organisations asked for it. Companies told the ICO that clearer advice on how to follow existing rules would help them use these technologies and business models responsibly. The regulator asked outside researchers to study the market and spoke to advertisers, publishers, agencies and technology providers.

The ICO uses “AdTech 3.0” to describe a direction the industry is moving in. It is not a product or an official standard. The main change is not simply that advertising systems use AI, because machine learning has been used in online advertising for some time. The change is that people can now talk to advertising systems in everyday language, and organisations can give systems tasks that they carry out with some freedom to make decisions. The report also explains that these tools often work on top of existing systems for buying and selling online ads, rather than replacing them.

The real story is where the evidence sits

The ICO's 2019 update report into adtech and real-time bidding raised concerns about consent and legal grounds for using data, transparency, special category data and complicated supply chains. The new report says agentic AI does not make these problems go away. According to the ICO, what changes is which organisations hold the information needed to deal with them. When an agent turns a goal into actions, communicates with other systems or uses information from several sources, different organisations may each hold information about a different part of the process.

The report's executive summary explains what this means. Companies now depend more on suppliers to explain how data is used. They need evidence showing where data comes from, what it is used for, who is responsible and what controls are in place, “rather than relying on labels or contract-based assurances alone”.

What a firm might rely onWhat the ICO saysReport section
A dashboard of campaign performanceWatching performance alone is not meaningful oversight of personal data use4.1
A person who signs off the system’s outputSome human involvement does not by itself stop a decision being solely automated4.1
A supplier label such as “aggregated” or “privacy-preserving”Labels do not explain what personal data a system uses, its origin or how decisions are made4.5
A contract naming each party’s roleController or processor status follows what actually happens to the data4.5
A stated purpose of “campaign optimisation”Broad wording may not adequately describe the processing4.6
A cohort-level outputWhether it is proportionate turns on the whole chain of processing4.6
An impact assessment done before the system goes liveIt may reflect only one point in the system’s life4.7

Strategic Reality: A company cannot hand its responsibility over to software. The report says: “An organisation can’t transfer its responsibility by having an agentic AI system handle campaign activities or optimisation.” But the advertiser is not automatically responsible for everything either. The ICO warns companies not to assume that the advertiser is responsible for all the data processing simply because it set the campaign goal.

What does an AI-run campaign look like in the ICO’s examples?

The report looks at five different situations, including these three.

In chatbot advertising, an LLM responds to a prompt. In the examples examined by the ICO, it also sends information about the request, such as its likely topic or what the person seems to want, to a separate advertising system. That system then chooses an ad. The report says it is important to tell the difference between an ad shown next to an answer and an ad included in the answer itself. In the examples the ICO studied, advertisers could pay for an ad space next to an answer, but they could not pay to change the answer created by the chatbot. For more background, see our earlier article about advertising in ChatGPT.

In industry protocols, AdCP (Ad Context Protocol) allows an advertiser to describe a campaign's goals in everyday language. Agents can then contact publishers, compare advertising opportunities, arrange purchases and make changes. The ICO mentions one early AdCP campaign on connected TV (CTV). It says this does not show that the approach has become standard. Another system, ARTF (Agentic Real Time Framework), puts an advertiser's agent inside the technical system of an ad seller. When the research took place, ARTF was still a proposal that no organisation had used commercially.

In walled gardens, large platforms control their own services, advertising systems and user data. The advertiser sets a goal and a budget. The platform's system then chooses the audiences, ad placements and spending. Some companies describe this as “supervised autonomy”. Advertisers watch dashboards showing combined campaign results. In some cases, platforms have removed manual controls that advertisers could previously use.

Critical Context: In our view, this is where two important parts of the report come together. In some walled gardens, advertisers only see dashboards showing combined campaign results. Elsewhere, the same report says that “simply monitoring campaign performance won’t provide meaningful oversight” of how personal information is used. The report does not explain what an advertiser should do if a platform refuses to share more information.

Where conversation changes profiling

Profiling has been common in adtech for a long time. The report says that using conversations for this purpose is “a related but distinct development”. A single conversation or chat history may reveal what someone is going through, what worries them or what they plan to do. It may also allow a system to understand meanings that are not directly stated. A prompt might openly mention a health condition, religious belief or political opinion. A system might also guess one of these things without the person ever saying it.

The ICO looks at what an organisation is trying to do and how it uses the information, rather than whether its guess is correct. An inference will generally count as special category data if an organisation aims to work out a special category trait, or treats someone differently because of it. This applies “regardless of how confident it is that the inference is correct”. In one example, someone asks a chatbot about symptoms. The service guesses that the person may have a health condition and uses this information to choose which ads to show. This involves processing special category data. The report repeats the position set out in 2019: using special category data for online advertising requires the person's explicit consent.

⚠️ Warning: Simply avoiding instructions that tell an agent to target people based on health, religion or politics does not settle the issue. In the ICO's example of expanding an audience, patterns found by a system could act as clues to these sensitive characteristics. What matters, the report says, is whether the advertiser uses the system to infer special category information or to treat people differently because of it.

Consent has to travel down the chain

When an advertising service stores information on a person's device or reads information from it, regulation 6 of PECR applies. Consent must also meet the UK GDPR standard, whether or not a chatbot is involved. The ICO's guidance on technologies that store or access information, which the report refers to, says companies need effective ways to tell relevant third parties when someone withdraws consent. This helps make sure that nobody continues using the data. The report found that more automation could make it harder to pass on and follow consent and withdrawal signals. But it also sees a possible benefit: agentic systems could pass these signals in clearer and more consistent formats. AdCP has a system for labelling how data is handled, which could help. However, the ICO says this would depend on whether the organisations involved actually use and respect the system.

Who is affected?

The report's examples affect different parts of the advertising supply chain in different ways.

GroupWhat the report means for themWhat is still unknown
Advertisers and agenciesRecord their role, what the agent was authorised to do and what is known about its actionsHow much detail walled-garden platforms will share
Chatbot and walled-garden operatorsLet people understand whether their interactions shape the ads they see, and keep paid ads distinct from generated contentWhat form those explanations will take
Ad tech suppliers and protocol buildersTheir customers need evidence of data sources, purposes and controls, not labels aloneWhether data labels in protocols will be accurate and respected

For people using chatbots, a conversation may feel like getting advice rather than using a service that sells advertising. They may not realise that two separate systems are involved in choosing an ad. The ICO says that a general notice explaining that a service “uses AI to personalise advertising” is not enough. It does not tell people which of their information affects the ad or which organisations help choose it.

What this means for advertisers buying agentic tools

The ICO explains what companies should check when buying AdTech 3.0 services from suppliers. These checks could cover where a system gets its data, what it does with that data, how much freedom an agent has to act and what limits apply to information passed between systems. The ICO also points to its AI audit framework. This can help companies work out and record which organisations are controllers and which are processors in an AI-based supply chain.

Companies also need to describe their purposes clearly. The report explains that when an agent “optimising” a campaign might move money between publishers, change an existing audience or create new guesses about people. Each action may use different data and have a different purpose. The report gives the example of connecting an agent to a new source of consumer data part-way through a campaign. It calls this “purpose creep”.

Implementation Note: The report gives a similar example involving a DPIA, or data protection impact assessment. An organisation assesses an advertising system that chooses audiences using a defined dataset. Later, it adds an agentic feature that can access new consumer data. The original assessment may still cover some of the data processing. But reviewing it helps the organisation check that it still describes the risks correctly.

The open challenges

Visibility inside closed platforms

The report says that responsibility belongs to the organisations that decide why and how personal data is used. They are expected to understand and supervise any systems they build, buy or operate. It also explains that advertisers using some walled gardens only see combined results on dashboards while the platform makes the actual decisions. However, it does not explain what advertisers should do if a platform refuses to provide more information.

Words that mean different things

Discussions with the industry found that companies use different words to describe these systems. The ICO says this makes it harder to explain how data is used and divide responsibilities in contracts. One tool described as “agentic” might follow instructions inside a single advertising system. Another might work with several suppliers and data sources. The ICO's advice is to look at what the system actually does, not what it is called. In our view, this means buyers must find out how the tools work for themselves.

Data that hides where it came from

The report describes “data washing”. This means combining or changing personal information so that its original source and the rules for using it become harder to see. The result may then be presented as an insight about an audience or as an aggregated signal. The ICO says that changing personal data does not remove the need to check how it was first collected. An output based on a group of people may reveal less information to an advertiser, but creating that group may still have involved using personal data.

Assessments for systems that keep changing

New permissions, data sources, features or connections to other systems can change what an agent does with data and how risky it is. The report refers to article 35(11) of the UK GDPR, which requires a DPIA to be reviewed when the level of risk changes. It describes review as something that should continue throughout the system's life. In our view, this means that for agents that learn and change their behaviour, checking risks must be an ongoing task, not something done only before launch.

Reality Check: Much of the market described in the report is still at an early stage. ARTF had not been used commercially, consumer shopping agents were not yet established in UK advertising, and fully agentic buying was not common across CTV when the ICO wrote the report. The regulator is explaining how it views these systems before several of these models become widely used.

What to watch

Four developments will help show whether the report's examples become expectations that companies must meet:

  1. The ICO’s agentic AI and ADM guidance and its code of practice on AI and ADM. The report says its findings will help shape both. If the guidance repeats that watching campaign performance is not the same as overseeing data use, or that “campaign optimisation” is too broad a purpose, advertisers will have stronger reasons to ask platforms for more information.
  2. How chatbot services explain their ads. The ICO suggests that a conversational service could let people ask questions about what influenced an ad. Whether operators build this feature or provide only a general notice about AI will show how far transparency expectations are put into practice.
  3. Consent signals in agent protocols. The report sees AdCP’s labelling system as a possible way to pass consent and restrictions between systems, if the organisations involved use and respect it. Whether participating advertising systems start using these signals will be a practical test.
  4. Joint working with other regulators. The report says chatbot and agentic advertising may also raise issues for regulators responsible for advertising standards, consumer law, competition and online safety. The ICO will share evidence when these areas overlap. Joint statements about chatbot advertising would show whether the different regulators are moving towards similar rules on disclosure.

Companies looking for a completely new set of rules will not find one in this report. Its value comes from its examples, which show how familiar responsibilities from the real-time bidding era apply when software makes advertising decisions. Companies will be in a stronger position if they can explain what their agents were allowed to do, what data they used and why. They should be able to provide this information without depending on a supplier's dashboard or label to explain it for them.


Source: Information Commissioner’s Office, AdTech 3.0: Online advertising in the age of LLMs and agentic AI, 13 November 2026. Also referred to: the ICO’s Update report into adtech and real time bidding (2019) and its technology guidance pipeline.

This strategic analysis was written by Resultsense, a UK-focused AI news and analysis publication. We will follow whether the ICO's upcoming guidance on agentic AI turns these examples into expectations that advertisers can use when asking platforms for information. Read more analysis at Insights, or get in touch.