Outlook
Part of What three 2027 pressures mean for content agency retainer budgets
Before you brief AI work to content agencies, check the data rules
How AI applications change content agency briefs in England, covering new data law, ICO guidance on AI and what to write into a statement of work.
What to take away
- A brief must name the AI applications, the tools, the retention period and every sub-processor before client content is processed.
- The client is normally the controller and the agency the processor, so Article 28 UK GDPR terms apply to each AI step.
- The Data (Use and Access) Act 2025 changed automated decision-making rules, added recognised legitimate interests and created a new Information Commission.
- Expect a lawful basis, a data protection impact assessment for high-risk use, and clear disclosure of AI involvement.
The rules a brief has to cover
Your statement of work must name the AI applications, the model routes, the prompt rules and the sub-processor chain. Those items belong in the contract, before any client content is processed.
Back them with the processor duties the agency carries. The content agencies advertising rules in England guide covers CAP Code checks when a brief also carries claims.
The ICO guidance on artificial intelligence and data protection covers lawful bases, transparency and accountability, with companion documents on decision logging and risk assessment.
Client briefs and audience data are controlled by the client; the agency processes them on instruction under Article 28 terms covering security, breach assistance and deletion. If the agency tunes a model on client data for its own purposes, it becomes a controller for that work.
What the Data (Use and Access) Act 2025 changed
The Data (Use and Access) Act 2025 amends the UK GDPR and the Data Protection Act 2018. It adds a safeguards-based route for some automated decision-making, sets out recognised legitimate interests, and establishes the Information Commission as the regulator's new body.
The automated decision-making provisions matter most for content work, because personalisation and audience scoring often inform decisions about individuals. Review any master agreement drafted before Royal Assent in 2025.
Example: five clauses for the statement of work
Put these five items in the statement of work, not the pitch deck.
Five clauses for the statement of work
- Which AI applications may be used
- What data may enter a prompt
- Who reviews output before publication
- How model use and prompts are recorded
- What happens if law changes
Five clauses for the statement of work
- Named tools and model routes. Suggested wordingonly the AI tools listed in Schedule 2, with 10 business days notice before any addition.
- Prompt and input rules. Special category data, unpublished client material and third-party personal data must not be entered into any AI prompt.
- Retention and deletion. Prompt inputs and outputs are kept for 30 days, then purged within 14 days of contract end.
- Sub-processor and transfer controls. Sub-processors are listed in Schedule 3, and transfers outside the UK rely on the IDTA or the UK Addendum.
- Human review. Outputs that touch personal data are checked by a named editor, and incidents are notified within 24 hours.
Retention should follow the storage limitation principle in the ICO guide to data protection for organisations. The content agencies commercial contracts in England guide shows how these clauses sit inside a wider master agreement.
For example, an agency on a 6,000 pound monthly retainer might allocate 900 pounds of that to review and compliance time. That is an illustrative allocation, not a market rate, and it belongs in the brief.
The four AI application buckets and their data risk
Group the work before you name tools. Four buckets cover most agency activity.
AI use buckets and data risk
Low risk
- Research and summarisation
- Rarely special category
- Drafting and editing
- Named person reviews
- Personalisation
- Audience data involved
- Measurement
- Scores content performance
High risk
- Research and summarisation
- Keep client material out
- Drafting and editing
- Rarely special category
- Personalisation
- ICO expectations bite
- Measurement
- Personal data involved
| Bucket | Typical content work | Personal data risk |
|---|---|---|
| Drafting and editing | Copy generation, rewriting, headline variants | Low if prompts exclude client material |
| Research and summarisation | Desk research, transcript summaries, competitor scans | Low to medium |
| Personalisation | Dynamic copy, subject lines, landing page variants | High, usually built on audience data |
| Audience analytics | Segmentation, propensity scoring, sentiment analysis | High, often profiling under Article 22 |
The first two buckets rarely involve special category data when prompts stay clear of client material. The last two usually do, which is where ICO expectations bite hardest.
Who is accountable when AI mishandles personal data?
The client is usually the controller and the agency the processor, so liability follows the contract and the facts. The ICO can act against either party, and individuals can claim compensation under Article 82.
An AI vendor that stores prompts on its own servers is a sub-processor, so it needs documented authorisation.
A sub-processor added without written authorisation breaches the processor terms, whatever the tool default settings allow.
How do you keep the brief current?
Set triggers, not a review date. Reopen the brief when the ICO publishes new guidance, when the client changes its privacy notice, or when a new model enters the workflow.
Compliance capacity is a buying criterion rather than an optional extra, which the content agencies trends and outlook for England in 2027 treats as central.
Common questions
Which guidance should the brief cite by name?
The ICO guidance on AI and data protection, Explaining decisions made with AI, and the AI and data protection risk toolkit. Cite the Act where automated decisions are in scope.
Does an agency need consent to use AI on client content?
Not always. It depends on whether personal data is processed and which lawful basis applies. Public copy is a different case from audience data used for personalisation.
Who is liable if an AI tool mishandles personal data?
The client is usually the controller and the agency the processor, so liability follows the contract and the facts.
How often should the AI clauses be reviewed?
Use triggers, not a calendar. Contract renewals, new tools and new ICO guidance are sensible points to revisit the wording.
