Why your team's AI output all sounds slightly different
Espen Oddvik · Founder
A while back I reviewed a stack of drafts from one team. A proposal section, two client emails, a product one-pager, all AI-assisted, all competent. And all written, as far as I could tell, by four different companies.
One draft was formal to the point of stiffness. One was chatty and used three exclamation marks. One called the product by its internal codename. The fourth promised a response time the company had stopped offering a year earlier.
Nobody had done anything wrong. That's the uncomfortable part.
The short version: a team's AI output sounds different from person to person because each person briefs the model from their own memory of the company, and the model fills every gap with averages. It stops when the company keeps one reviewed set of facts about itself and delivers them into every AI session automatically, which is what Company Brain does through MCP for Claude, ChatGPT, Copilot and Cursor.
The rest of this post is how that plays out in practice.
Where does the drift come from?
When a person briefs an AI, they type what's in their head. The senior consultant carries the firm's voice after a decade and briefs it in without thinking. The person who joined in March carries whatever the onboarding deck managed to transfer, which is less. The AI fills every gap with the most average possible answer, because averages are what it's made of.
So each teammate gets output shaped by their own private version of the company. Five people, five versions. The drift is what happens when the source of truth is five separate memories.
And here's the question worth sitting with: before AI, how did your company sound consistent? Mostly through slowness. Fewer documents got produced, senior people touched more of them, and the voice passed through the same few hands. AI removed the slowness and kept everything else. Volume went up tenfold and the quality gate stayed the same size.
Why the drift compounds
Small drift doesn't stay small, because output becomes input. The chatty email gets pasted into the next prompt as an example of "how we write". The codename escapes into a client deck, and a customer starts using it. The outdated promise gets made twice more before anyone catches it, and now honoring it is a conversation.
Clients rarely name what they're noticing. They just register that the proposal sounds like one company and the follow-up email sounds like another, and file it somewhere under polish. For a business that sells trust, that filing cabinet is expensive.
How do you keep a team's AI output consistent?
I've seen three fixes attempted. Two of them are half-fixes.
Tightening review catches drift after it's produced, and turns your most senior people into a bottleneck. Writing a shared style document for AI prompts helps the people who read it, on the days they paste it. Both fight the symptom.
The fix that holds is boring: give the company one set of written facts about itself, let one person own and review them, and deliver them into every AI session automatically so nobody's memory is the source anymore. Same facts for the ten-year veteran and the March hire. When the response time changes, it changes in one place, and the old promise stops appearing anywhere.
That's the shape of the thing we built. Company Brain holds your company's facts in one reviewed place and delivers them to Claude, ChatGPT, Copilot and Cursor through MCP. If your team's drift shows up mostly in campaigns and client copy, the marketing teams walkthrough shows what those facts look like in practice.
The team with the four-company draft stack runs this way now. Their drafts still vary, because people vary, and that's fine. But they all sound like the same company wrote them.
Which, finally, it did.