AI does not care about your org chart.

Why does AI content drift off brand across teams?

Because AI lets people do work that used to belong to another team, and the brand decisions did not travel with the task. Marketing work is the kind that travels most. Whoever picks it up is working from memory, a blank prompt, or an old deck.

OpenAI measured this. In research published on July 27, it studied more than 800,000 work-related messages from U.S. ChatGPT users and found that 43.5 percent of the job-specific ones concerned tasks historically associated with another occupation. Its own summary: marketing and engineering tasks “travel farthest.” In the sample, 29 percent of salespeople's job-specific messages looked most like marketing work, and so did 28 percent of designers' and 26 percent of customer experience staff's.

Read the limit too. The study counts what people asked for, not what shipped. OpenAI says it does not observe whether the output was used, how good it was, or whether a specialist reviewed it. That gap is exactly where a brand gets into trouble: the work happened, and nobody from marketing saw it.

The prompt box has no respect for your org chart. It has no opinion about whether the launch email sounds like you.

Is it only people crossing team lines?

No. Tools are crossing them too, and they move faster than any approval queue. The agent inside your design tool and the connector inside your ad account can now do brand work in places your brand team may never look.

On October 6, Figma took its design agent out of beta. TikTok's ads connector for AI assistants lets an AI agent create campaigns, adjust budgets, and refine audiences in TikTok's ad platform. Neither is a problem on its own. Each is one more place where your brand gets made without your brand team in the room.

What should travel with the work?

Four kinds of decision, written down in plain words. Together they are what a good marketer checks without thinking, and what a salesperson or an agent cannot guess.

• What you promise: the positioning and audience decisions already made, so nobody reinvents them in a prompt.
• What you can prove: the claims you can support, and how strongly. A confident sentence is not evidence.
• How you sound and look: voice, visual rules, and examples of work you would ship again.
• Where it stops: what a person or a tool may change on its own, what it must leave alone, and who owns the edge case.

Figma's guidance for its agent shows the shape this takes. It asks teams to write each design rule in four parts: when it applies, what to do, what to avoid, and why. Those files cover “conventions that the agent likely won't be able to figure out from the assets alone.” The same is true of your brand. The avoid line matters most, because it is the one nobody can infer from your best work.

Where should these brand rules live?

In one source your company owns, with every tool reading from it. If your design tool, your chat assistant, and your ad agent each hold their own copy of the rules, you have three brand books drifting apart on three schedules.

Figma now reads guideline files you upload. ChatGPT, Claude, and the rest each have their own way to load instructions. Keep one master and treat every tool's copy as a delivery from it, never as the place you edit. When positioning changes, it changes once, and every tool gets the new version. That master is what we call a customer-owned Brand Brain: your approved decisions in plain text files your company keeps, versioned, and editable without us.

If you want the practical setup for the assistants your team already uses, read how to make ChatGPT and Claude follow your brand guidelines.

What does this look like in a working company?

We run Adaptius this way, so here is ours. Three AI agents from three companies do jobs that would sit in three departments. Claude drafted this article. OpenAI's Codex made its cover from our approved character references. Grok runs our account on X. Each works on its own clock. Codex builds what Claude briefs. Grok's posts wait for Claude, as the CMO, to approve them, and its replies follow a written playbook.

All three read the same Brand Brain before they work, and one claims file decides what any of them may say about us. When we correct a rule, it changes once, in the Brain, and the next run of every agent picks it up. That is the whole trick: the work crosses lines all day, and the decisions travel with it.

It does not make the agents perfect. It makes each mistake traceable to one rule, so the fix lands once instead of three times. The checking half, what each agent leaves behind so a person can review its run, is in every AI agent should leave a receipt.

What can I do this week?

Pick one piece of customer-facing work that another team made with AI last month: a sales deck, a help article, a product page. Ask three questions of it.

Every answer you had to make up on the spot belongs in the shared source. If you are writing rules for an agent, read how to test a brand rule without blocking the work first.