Guide
Why your AI writes off-brand.
The drafts are fluent, fast, grammatically spotless — and none of them sound like your company. It isn’t the model. It’s three structural gaps, and none of them closes with a better prompt.
Cause one: your guidelines aren’t machine-readable
Your brand lives in a PDF, a slide deck, maybe a Notion page. Those formats were designed for human eyes. The AI tools your team writes with have never seen them — and even pasted in, a 60-page document is prose to skim, not rules to follow. Frontify's research captures the paradox: almost every company has guidelines, and the overwhelming majority of marketers still struggle with off-brand AI content. The guidelines exist. They're just unreadable by the things doing the writing.
Cause two: the brand isn’t where the drafts happen
Even a perfectly written brand document fails if it lives in a drive while the writing happens in Claude, ChatGPT and Cursor. Every draft starts from what the model guesses your company sounds like — corporate-generic, competent, interchangeable. The brand arrives later, as your feedback, after the words exist. By then you're not guiding; you're correcting. At AI volume, correcting is a full-time job.
Cause three: everyone prompts differently
Some people paste voice instructions into every chat. Most don't. A few maintain personal system prompts — each a private, slightly wrong copy of the brand. Canva's own research says it plainly: five people using the same AI writing tool produce five different voices. The inconsistency isn't carelessness. It's what happens when the standard lives in individual habits instead of a shared source.
What won’t fix it
- Longer prompts. Instructions decay — they get trimmed, forgotten, and reinterpreted per person. A prompt is a habit, not a system.
- More review. Your hours are the scarcest resource in the building; AI volume grows and your calendar doesn't.
- Another training session. It helps the people who were already careful. The drafts that hurt you come from everyone else.
What does: the brand as data, in the tools
All three causes share one fix. Structure the brand as a living brand system — machine-readable, current, versioned — and connect it to the tools where drafts are generated. A published SPOKEPROS brand is a live brand server: Claude, ChatGPT and Cursor read the voice rules, boilerplates and banned phrases before writing, and everyone's tool reads the same ones. The three gaps close at once: readable, present, shared.
That's the difference between telling every writer and every tool what the brand is, forever — and telling the system once.
Tell the system once.
Import your brand, publish it, connect your tools — on the Free plan, in one sitting.