How do agencies protect client confidentiality when using AI?

What clients assume stays private

Agencies hold some of the most sensitive material a client has, often months before anyone else sees it:

• Unreleased products: Packaging designs, product names and launch dates for a campaign still under embargo.

• Campaign and media plans: Budgets, targeting strategy and creative routes a competitor would value.

• Crisis and PR material: Holding statements, issues briefings and internal correspondence that could cause real harm if exposed.

• Customer data: Mailing lists, event delegate lists, CRM exports and survey responses.

Clients rarely think about where this goes once a brief lands. They assume it stays private, and your AI policy should make sure it does.

Account type matters more than the tool

AI tool terms differ between consumer and business versions, including whether inputs may be used to train models. A free personal account and a business account from the same provider can treat your data very differently.

• Approve a short list: Decide which business tools and account types staff may use for client work, and put it in writing.

• Check training and retention settings: Confirm whether inputs are used for training, how long they are kept, and who at the provider can access them.

• Ban personal accounts for client work: A designer pasting a launch brief into a personal account on their phone is one of the easiest ways for confidential material to leave your control.

• Review access when people leave: Shared workspaces can hold months of client prompts and uploads.

What should never go into an AI tool

Even with approved accounts, some material should stay out unless the client has explicitly agreed:

• Embargoed and price-sensitive information: A PR team drafting a results announcement for a listed client should not put the numbers into any external tool.

• Sensitive personal data: Health, financial or similar information about real people.

• Raw customer lists: Use anonymised or synthetic samples when testing segmentation or personalised copy.

• Credentials and access keys: Never paste client logins or keys into prompts.

Personal data and UK GDPR

UK GDPR applies whenever personal data is entered into an AI tool, and the organisation whose staff enter it remains responsible for it. If an events coordinator uploads a delegate list to draft personalised follow-up emails, your agency is accountable for that processing. Putting client data into AI tools under UK GDPR covers the data protection side in depth.

NDAs and contracts still apply

Confidentiality duties under client contracts and NDAs apply alongside data protection law. Reread your NDAs with AI in mind. Some prohibit sharing confidential information with any third party, which may include an AI provider. Others require client consent before you use subcontractors or external services. If the wording is unclear, ask the client rather than guess.

These duties also extend to anyone working on the account. Freelancers using AI on agency work explains how to apply your rules to them.

Answering security questionnaires with confidence

Larger clients now ask agencies directly how they use AI. Prepare a standard answer that covers:

• Tools and account types: Which business tools you use, and confirmation that personal accounts are not permitted for client work.

• Data handling: Whether inputs are used for training, where data is stored and how long it is kept.

• Restricted material: Your list of what staff never enter into an AI tool.

• Oversight: Who approves new tools and how staff are trained.

A clear, consistent answer reassures procurement teams far more than a vague one, and saves you rewriting it for every pitch. Contract and data protection duties depend on your circumstances, so take advice on your own situation.