Your Brand Guidelines Need a Second Reader Now — and It Doesn't Read Between the Lines
For years a brand guide had one kind of reader: a designer, a copywriter, a new hire. Someone who could look at "warm but professional" and fill in the rest from context, from the last project, from a conversation in the hallway. That reader is no longer the only one. Increasingly the first draft of an email, a product description or a social post is written by a model, working from whatever the guide actually says. And a model doesn't read between the lines. It reads the lines.
Adjectives are not instructions
Look at a typical brand voice page: friendly, confident, human, innovative. A person reads those as a mood. A model reads them as a request to produce the most statistically typical "friendly, confident" text it has seen, which is exactly the generic copy that makes brands blur together. Practitioner guides on keeping AI content on-brand keep arriving at the same diagnosis: vague descriptors produce inconsistent output, and specific rules work better.
This is the quieter cousin of the problem we described in why AI isn't a brand strategy. The tool isn't the issue. The brief is.
What a second reader needs
Three things turn a mood into something that can be followed:
- Rules you can check. How long sentences run. Whether you use contractions. Which words you use for your own work, and which ones you never would.
- Examples, good and bad. Two or three real passages that sound like you, and two or three that don't, with a line on why. Examples teach faster than description, for people and for models.
- A list of refusals. The claims you never make, the clichés you avoid, the tone you won't take with a frustrated customer. A voice is defined as much by what it leaves out, the same logic as positioning being what you refuse.
Express is the part that gets repeated
Identity isn't the logo. It's what gets repeated, in every channel, by every hand, until people can recognise it without the name attached. When the hands multiply, including the automated ones, the guide stops being a reference document and becomes the thing that decides whether that repetition stays consistent. Teams that run AI at scale tend to report the same drift: more people, more tools, more interpretations of the same loose rule.
Keep a human at the end of it
Guidelines reduce drift. They don't replace judgment. Someone who knows the brand should still read what goes to a customer, and what they correct is the best material for the next version of the rules. The guide is never finished. It's a record of what the brand has learned it sounds like.
This is an Express post — the second of our four stages: Define, Express, Build, Grow. If your guidelines describe a mood but no one, human or machine, could follow them, /branding is where that gets made specific.
Common questions
Why does AI-written content sound off-brand even with a brand guide?
Usually because the guide describes the brand in adjectives, such as "friendly" or "confident", that a person interprets from context and a model interprets as the average of everything it has read. Without concrete rules and examples, the model fills the gap with generic language.
What should a brand voice guide for AI include?
Specific rules (sentence length, words you use and words you ban, how you address the reader), a handful of real examples of on-brand and off-brand writing, and a clear statement of what the brand never says. Examples teach a model faster than descriptions do.
Do I still need a human to review AI-written content?
Yes, especially for anything customer-facing. Guidelines reduce drift, but they don't replace judgment. The reviewer's corrections are also the best source for improving the rules over time.
