Brand guidelines are the theory, but AI needs to learn the practice.

In conversation with global marketers at our recent roundtable events, the question has come up several times: once you have chosen your AI tool, how do you tell it what it needs to know about your brand, when most of that knowledge has never been written down?

Brand guidelines exist for a reason, as do tone of voice documents. But the criteria that actually decide whether an asset is right in a given market lives in the heads of the handful of people who have worked on that brand and that market for years.

So what does the system need to know to be useful? Not only to produce assets, but to move assets through review, approvals and out to market faster, without adding to the workload of your team. How can you set the system up to actually deliver on the efficiencies it promises?

The basics: what AI needs to know

In our experience, an AI producing or adapting global campaign assets needs minimum five in-depth inputs before it can be useful in the context of global campaigns:

1. Brand rules and tone of voice, including what is fixed and what can flex.
2. Legal and regulatory restrictions that apply market by market.
3. Technical specifications for every channel, platform and format you deliver into.
4. Approval criteria, meaning what good looks like and who decides on it.
5. Cultural knowledge of each market, so the work is made and delivered in a way that resonates locally, which goes well beyond language.

What brands aren't capturing in AI

Underneath the five basic inputs sits a layer that decides whether the output is genuinely usable or just plausible. In a recent survey of localisation professionals, missing context was the most cited operational failure, ahead of quality and terminology (Crowdin, 2026). Much of what's worth capturing across judgement, habit and institutional memory has never been written down in a brand guidelines document. Here are a few worth including:

• Capture the reason behind a rule, not just the rule itself.

• Track what has already been rejected, and why.

• Teach the system what is obvious in market and invisible from head office.

• Capture the unwritten approval logic.

• Keep it current, in one place, with one owner.

What cutting corners can cost you

This is a significant piece of work, and an ongoing commitment. If you fail to capture the nuance in the system and keep it up to date, the knock-effects undermine the very efficiencies you are using the AI system for:

• Standards slip under deadline pressure, and work goes out that the team knows is not properly on brand.
• Errors are more likely to get through, because nothing in the system catches them and there is less time to catch them by hand.
• You are more likely to publish work that is not quite right for the market, which is the failure that costs most and surfaces latest.
• The corrections get applied to the asset rather than to the instruction. So quality plateaus and you get the same errors, just at higher volume.
• The team loses trust in the tool, and the wrong conclusion gets drawn, because everyone decides the technology is not ready when the technology was fine and the brief was not precise enough.

How to get started

AI didn't create this problem but it has exposed it. This knowledge was always what held quality together across markets, but it was being carried around in people's heads, invisibly, and it worked because a human looked at every asset before it went out. Take the human out of every check and the gap stops being a risk and becomes the output. Understanding what your best people already know, and getting it written down, is and should be the priority investment. That's what will still be worthwhile in five years, no matter what you are using to produce marketing campaigns by then.

Building and maintaining this kind of AI-enabled knowledge base is part of our operating system at Freedman. If it is something your team is wrestling with, come and talk to us.