Copy is the great unsolved problem of most design and product teams.
Not because people can't write. There are plenty of good writers at most companies. The problem is that good writing requires context, and context lives in too many places. The brand guidelines that took months to write sit in a folder no one opens. The voice and tone doc is four versions old and nobody's sure which one is current. The product manager writes "discount" when the brand standard is "savings." The new hire sends a partner email that sounds nothing like the website. And the designer who knows all of this stuff, who built the guidelines, who wrote the quick reference cards, fields the same three questions every single week.
This is a solvable problem. And if I were to solve it, here's exactly how I would do it.
Step one: build something worth connecting to
An AI assistant is only as good as the knowledge base behind it. If the source material is thin, outdated, or scattered, the AI will confidently give you thin, outdated, scattered answers. So before you touch any AI tooling, you need a real standards manual.
I would build this in Confluence, because that's where the team already works. The goal is a single space that serves as the undisputed source of truth for anything related to how the brand communicates. Not a doc, not a deck, not a Notion page that one person maintains in a corner. A real, structured, living system.
The core of that system would include:
A Writer Quick Reference — the thing a writer opens when they just need a fast answer. What words do we use? What do we avoid? What are the names of our products, stated exactly right? This page needs to be scannable in thirty seconds and accurate enough to trust completely.
Core Content Principles — the deeper explanation of why the brand communicates the way it does. Voice, tone, the values that should come through in everything. This is less of a lookup document and more of an orientation for anyone who wants to understand the brand at a conceptual level.
Platform-Specific Guidelines — because the brand voice sounds different on a website than it does in a transactional email, and both of those sound different than a social caption. These pages give channel-specific direction without having to reinvent the principles each time.
Terminology Standards — a definitive record of the right way to say things. Not "discount," savings. Not "subscribers," members. Not "We Salute" (two words), "WeSalute" (one). These distinctions sound small until someone gets one wrong in front of a hundred thousand people.
[Screenshot: Confluence space overview showing page structure and navigation]
Build this first. All of it. The AI comes later.
Step two: create the AI context artifact
Here's the part nobody talks about enough when people describe building AI-powered workflows: the AI doesn't learn your brand by magic. You have to give it the context it needs every time it starts a new conversation, because it doesn't carry memory between sessions.
The solution is a context artifact — a single, comprehensive Confluence page that describes everything the AI needs to know to operate correctly. Think of it as the system prompt your AI assistant reads at the start of every conversation.
This page would include the brand's mission and audience. It would describe the brand voice and tone in plain terms the AI can apply. It would spell out the most critical rules, the absolute non-negotiables, at the very top so they're impossible to miss. And it would describe the organizational structure of the whole Confluence space so the AI knows where to look for specific types of guidance.
[Screenshot: AI context artifact page structure]
Every time you start a new conversation with the AI, you point it at this page first. Everything it does from that point forward is informed by this context. When the guidelines change, you update one page. The AI's behavior updates automatically.
This is the architecture that makes the whole thing maintainable. Without it, you're re-explaining your brand from scratch every single conversation.
Step three: build the team interface
The AI knowing the brand standards is only half the system. The other half is how your team interacts with it, especially the team members who shouldn't need to know anything about prompting an AI to get useful answers.
I would build this as a request workflow inside Confluence itself, using a set of structured templates that guide people toward asking good questions without needing to know how to do that on their own. The templates would cover the most common types of requests:
Copy Review: Someone pastes in a piece of marketing copy and asks whether it's on brand. The template prompts them to include the channel, the audience, and any specific concerns they have. The AI reviews against the guidelines and provides specific, actionable feedback.
Brand Questions: Someone has a strategic messaging question that doesn't have an obvious answer in the quick reference. The template helps them frame the question clearly. The AI reasons through it against the principles and recent decisions, then provides a recommendation.
Terminology Clarification: A product name is in question. A new category needs a label. Someone's not sure which word applies to a specific context. The template captures the specifics, the AI provides a ruling with rationale.
Guideline Updates: Someone on the team wants to propose a change to the brand standards. The template structures the proposal. The AI can surface conflicts with existing guidelines, note related decisions, and document the reasoning once a decision is made.
[Screenshot: request template examples in Confluence]
Each of these lives in a "New Requests" folder. When the AI processes a request, it moves the page to a "Resolved" folder and uses an @mention to notify the person who submitted it. They get an email with a link to the response. The whole interaction happens inside the tool the team already uses.
[Screenshot: resolved request with AI response and notification]
Step four: automate the notifications
The part of this system that makes it actually feel like a product rather than a manual workaround is the notification layer.
Confluence has automation rules built in. You can set a trigger on "page edited" that fires whenever the AI updates a request page. That trigger sends an @mention to the requester, which generates an email with a direct link to the response. The person who submitted the request doesn't have to check back. They get notified when their answer is ready, the same way they'd get notified about anything else in Confluence.
Setting this up properly is a bit of configuration work but it's not complicated. The result is a system that the team experiences as responsive and reliable, not as something they submitted a request into and then had to go hunting for later.
[Screenshot: Confluence automation rule setup]
Step five: add a discovery layer
One thing I'd add that most people wouldn't think to include upfront: a separate section of the space for exploratory, not-yet-official brand thinking.
Call it something like "Brand Discovery & Ideation." Label everything in it clearly as experimental. The purpose is to have a home for strategic brand questions that are being actively researched but haven't been resolved yet. New audiences the brand might expand to serve. Terminology experiments. Positioning concepts being pressure-tested.
Keeping this separate from the official guidelines matters a lot. You don't want the AI citing exploratory thinking as if it were established policy. The clear organizational separation makes that impossible. The discovery folder is for thinking. The guidelines are for doing.
[Screenshot: discovery folder structure and labeling]
What this actually gives you
Let me be concrete about the problem this solves, because I think it's worth naming precisely.
Before a system like this, brand compliance is largely dependent on individual knowledge. The people who know the brand standards well apply them correctly. Everyone else is guessing, or asking the person who knows, or both. As the team grows and the guidelines evolve, that individual knowledge becomes a bottleneck.
After a system like this, brand compliance is systematized. Anyone on the team can submit a question and get a reasoned, guideline-backed answer within twenty-four hours. Decisions get documented so the next person who asks a similar question can reference the precedent. The guidelines stay current because there's a clear process for updating them. And the designer who built the whole thing stops fielding the same three questions every week.
The AI isn't replacing the brand expertise. It's extending it. The judgment that went into building the guidelines still drives everything. The AI just makes that judgment accessible to everyone, all the time, without the person who holds it needing to be in every conversation.
The part I'd do differently in hindsight
If I were starting over, I would build the request templates more narrowly. The first version of something like this tends to have templates that are too broad, trying to accommodate every possible type of question in one form. The result is that people either fill in too little and the AI doesn't have enough to work with, or they fill in too much and the AI gets lost trying to address ten different things at once.
Narrower templates that ask for exactly the right amount of context produce much better answers. It's more upfront design work, but it pays off quickly.
I would also establish a more explicit process for graduating discovery content into official guidelines from the start. Research accumulates fast, and without a clear path from "we're exploring this" to "this is now official," you end up with a discovery folder full of good thinking that never fully makes it into the standards.
Copy bottlenecks are one of those problems that seem small until you're inside one. The question that gets asked four times a week. The inconsistency that makes it into the campaign. The new team member who writes beautifully but has no idea how the brand actually talks.
A system like this doesn't eliminate the need for good writers or good judgment. But it does make sure that good judgment is never stuck in one person's head. And that, in my experience, is worth building.
Even if I'm not saying I built it.
What did you think?
Share with friends