Ask three people to find the latest product description, a good example of the brand’s voice and the right logo for a dark background. You’ll learn quite a bit about where the brand actually lives.

Some of it is probably in a brand guide. Some is in the last campaign deck. Some is in the head of the person who remembers why a particular phrase stopped being used.

When a team starts producing more with AI, those gaps get easier to see. You can generate twenty variations of an ad in less time than it used to take to make two. If the brief pulls from an old product page, you’ve just made twenty variations of the same mistake.

What I’d put in the shared context

  • Voice. A few examples that show how the brand actually writes in different situations, with notes on why they work. “Friendly and clear” is a reasonable start. An example of how you explain a declined application or a price increase tells you rather more.
  • Product facts. What the product does, who it’s for, current offers and approved claims, with links back to wherever those facts are maintained. Include the limits people are most likely to miss.
  • Visuals. Approved assets, layouts and examples that show how the brand adapts across formats. A tiny display ad and a landing page are doing different jobs, even when they’re part of the same campaign.

That’s what I mean by a context layer. It gives the team, and any AI tools they use, a common place to start. I’d start with references for one real campaign and see what the team still has to chase down.

The bit that needs looking after

A product changes. An offer expires. A visual that made sense in one campaign becomes a template for six others. If the shared reference never changes, the system gets very good at reproducing decisions the team has already moved on from.

I’d give each part an owner and use changes in the work as prompts to maintain it. When the product team changes an offer, update the reference. When a reviewer corrects a claim three times, check whether the brief or the shared source is teaching people the wrong thing.

Where my thinking is today

Before I scale an AI-assisted creative workflow, I want to know what it will scale. A few trusted references that people keep using and correcting are worth a lot. They also help the person reviewing the work see what was intentional and what simply turned up in the output.