The Brand Guardrails Your AI Tools Are Missing

By Midya U · Midya U Advisory

In brief: Most institutional brand reviews check whether the facts, spelling and logo are right — almost nobody checks whether a campaign sounds and behaves like the brand, which is how two reviewed, approved campaigns (Callaway Golf and Converse) still went out wrong in August and September 2026. A one-page brand guardrail — five personality adjectives, banned topics, claims that need a source and one named owner of tone — closes the gap a style guide leaves open, and it's what stops an AI drafting tool from filling that gap with the average voice of the internet.

Callaway Golf ran a social campaign with Good Good Golf that showed a woman golfer being knocked to the ground, and MediaPost reported on September 1 that at least three marketing staff at Callaway and Good Good were leaving over that ad. Then Converse posted a sneaker ad where, critics said, the lighting and the pose evoked a Klan hood and a lynching. On September 18 it pulled the ad and said, "We understand why this image is deeply upsetting and recognize that we got this wrong."

Neither was an AI mistake. People made both, and people approved both. If your team is producing far more content than it was two years ago, with the same two reviewers, that should worry you.

Why do bad campaigns get past a review process?

Because most reviews check the wrong things.

In a university setting, a typical sign-off asks whether the facts are right, spelling and grammar are correct, the logo is current and, occasionally, whether faculty has seen it. Almost nobody in that chain is asked whether this sounds and behaves like us. That one gets answered by instinct, which depends on who's reviewing that campaign.

Julie Reeves, a VP analyst at Gartner, made this argument in Marketing Dive on September 9, using Callaway as the example. Her view is that positioning alone gives reviewers too little to judge creative by, and that brand personality, written down as five adjectives or fewer, gives them something objective to check against.

What's the difference between a style guide and a guardrail?

A style guide covers how things look. A guardrail covers the things your brand won't say or do, which most brands leave out entirely.

The usual institutional brand book runs to sixty pages on logo clear space, the approved shade of blue and whether we use the Oxford comma, and says nothing about whether we joke about tuition, how we talk about a program being cut, or which claims we never make without a source attached.

For years that gap was covered by a long-serving writer who just knew. That writer is now usually reviewing AI drafts, with less time per piece.

Why does this matter more once AI is writing the first draft?

An AI tool has no instinct about your brand. It only has what you gave it, and in most teams that's nothing, or a PDF of the brand book that it reads as a list of colours.

So it fills the gaps with the average voice of the internet, which is upbeat, a bit breathless and fond of the word "journey." Your reviewers then catch what they can. As I wrote in Why Faster Drafting Didn't Make You Faster, the review queue is already the slowest step in most institutional teams, and AI just pushed more volume into it.

I'd be careful not to oversell the fix. A one-page guardrail doc probably wouldn't have stopped the Converse image. The problem there was what the photo looked like; no wording rule would have flagged it, and I don't know of a checklist that reliably catches that kind of thing. What it does catch is the steady drift in tone that sets in when most of a month's copy is drafted by a tool that has never met you.

What goes on a one-page brand guardrail?

Keep it to one page, because it has to fit inside a tool's custom instructions and an agency brief.

• Five personality adjectives, each with a "which means" and a "which doesn't mean." "Warm, which means we name the student's actual worry. Which doesn't mean exclamation marks."
• Words and phrases you don't use. Start with the ones your reviewers already cross out.
• Claims you never make without a source. Employment rates, rankings, salary figures.
• Topics you don't joke about. Money, immigration status and mental health are a reasonable start for most institutions.
• Two examples of on-brand copy and two of off-brand copy, taken from your own recent work.
• The name of the person who has the final word on tone. One person, named.

Is a list of banned words enough?

No, and it's usually the first thing a team writes.

A don'ts list is easy to build because every reviewer already has one in their head. Give it to an AI tool on its own, though, and you get copy that breaks none of your rules and still doesn't sound like you, because the tool reaches for the next synonym along. What moves the output is real examples of your writing at its best. If you only have time for one section of the page, write the on-brand examples first and add the banned words after.

Where would I start?

Pull the last ten pieces your team published and ask your two most experienced reviewers to mark every line they'd have changed, and why. The reasons they give will repeat, and those repeats are your adjectives and your banned words. Put the result into every AI tool your team uses and into your next agency brief.

If the reviewers can't agree on the adjectives, the problem may be older than AI, and How to Tell If Your Brand Predates Your Market is the better place to start. The guardrail exercise is also one of the first things I do in a Marketing Function Audit & Reset.