← All research

The Big AI Secret · ongoing research

The client's contract can veto the agency's AI

The call evidence shows that permission to use AI can depend on the client, not just the business doing the work. Contract bans, agreements silent on AI and personal preferences create different starting points for otherwise similar tasks.

16 November 2026 · 4 minute read

An agency can have the skills, the tools and a willing team. The client can still say no. In one agency account from The Big AI Secret call evidence, some client contracts prohibit AI. The accompanying call summary records no formal written AI policy within the agency itself. Here, an explicit restriction comes from outside the business doing the work.

“And we have some of our clients have, you know, strict no AI rules within our contracts. So we know, you know, everybody knows which clients they can use it for, which they can't, and to the degree to which they can use it.”Agency and marketing, under 25 staff

That changes what readiness means. Knowing how to use AI is not enough. Staff also need to know whose work they are doing and what that client permits. The speaker describes that knowledge as shared. The extract does not show how it is recorded, kept current or passed to someone taking over an account.

This week's focus is client permission across agency and media work. The evidence is selected, anonymised call extracts, not a measure of how common these practices are. Roles and the total business sample behind this selection were not supplied. The accounts reveal different permission arrangements, but cannot establish whether those arrangements work consistently.

Silence does not settle the question

An explicit ban gives staff a clear starting point, even if applying it takes care. Another agency account described a less settled position. Client agreements matter to its decisions, but those agreements do not mention the technology being assessed.

“We don't have clear policies, but we do feel we do have to evaluate what we use AI for in the light of data protection agreements we have with our clients. None of them talk about AI.”Agency and marketing, under 25 staff

The important word is evaluate. This speaker does not treat the absence of AI wording as an automatic yes. They described assessing proposed uses against existing obligations. That is a different activity from checking whether a tool is available or whether its output is good enough.

The extract does not establish what those agreements legally allow. Nor does it tell us who makes the assessment or what happens when the answer is unclear. It does show a practical burden: staff must connect an existing agreement to a new way of working, without clear AI policies to guide them.

Client permission

A tool being available does not tell staff whether they may use it for this client.

Waiting for the client to ask

A media speaker described a more cautious personal starting point. Rather than using AI unless a client objects, they prefer to wait for the client to request it.

“Well, with my clients, honesty, I prefer not to use AI as far as they don't ask me.”Media, under 25 staff

This is a stated preference, not evidence of a contractual requirement or a sector-wide rule. That distinction matters. A client ban and a supplier's choice to wait can produce the same visible outcome, no AI used, for different reasons. An adoption assessment that asks only whether AI was used would miss that difference.

It would also be a mistake to label this hesitation as a lack of capability. The quote concerns permission and the client relationship. It says nothing about whether the speaker could complete the work with AI. A business may deliberately leave a capability unused because it does not consider permission established.

An opt-out is a different arrangement

Another agency account described giving clients a choice about recorded communications. The call summary supplies that context for a short statement about allowing clients to opt out.

“And yeah, we give clients the opportunity if they don't want to, you know, for us to do that, to remove it.”Agency and marketing, under 25 staff

This should not be stretched into evidence of permission for every AI activity. The described choice concerns a particular practice. Agreement to recording communications would not, on this evidence, establish agreement to AI-generated creative work or other uses of client material. Permission needs a defined scope before staff can act on it reliably.

The contrast with waiting for a request is useful. An opt-out offers a route to decline a proposed practice. Waiting for a request makes the client's initiative the starting signal. Neither account gives enough detail to judge how clearly the choice is explained or how reliably it is honoured.

What a director needs to be able to trace

The management question is not simply whether the business has an AI policy. It is whether someone picking up a client task can find the relevant permission, understand its limits and identify who can resolve uncertainty. A general policy may set the business's standards. It cannot, by itself, establish what a particular client has agreed.

A useful follow-up would trace a recent client task from the agreement to the decision about AI use. Where was permission found? What activity did it cover? What would happen if the client changed its position? Those checks would test the reported practices rather than merely collect another assurance that everyone knows the rules.

The finding is narrow but consequential. In these accounts, AI use is partly governed by the relationship with the client. Capability explains what a business can do. Permission helps explain what it actually does, and why apparently similar work may follow different rules.

Contributors Julia Chepurna, Sarah Bassett, AI Analysts

Published by Sarah Bassett

Where does your business stand?

Six questions, two minutes, no account needed to see your stage.

Find your stage →