Last fall, I lost most of my ChatGPT account in a glitch during an upgrade.
Two and a half years of chat history. Custom GPTs I’d spent weeks building. Client context I’d fed it conversation by conversation over hundreds of sessions. Lead magnets. Ghost-written content for clients across multiple industries. Gone, or close enough to it that the account I was left with didn’t know me anymore.
I ran a test query after the glitch. The response opened with something like: You are the executive director of So-and-So and you sell real estate. That was not me. Not even close.
What I learned in those two days of rebuilding was something I already knew intellectually but had never been forced to prove under pressure: training an AI tool through conversation history is not the same as having a documented brand foundation. One is reactive. The other is proactive. When one of them disappears, only the other one survives.
What Most People Get Wrong About Training AI on Their Voice
Most people who use ChatGPT or Claude consistently believe they’ve trained their AI on their voice. And in a conversational sense, they’re right. The tool learns patterns from repeated use. It starts to mimic certain rhythms, catch certain preferences, get closer to the output they’re looking for.
The problem is that training through conversation is cumulative and fragile. It lives inside a platform you don’t own. It can’t be exported as a file and dropped into a new account. It can’t be handed to a VA, a content strategist, or a new AI tool without starting the whole process over.
A brand book is different. It’s a standalone document, external to any platform, that captures who you are, how you talk, what you believe, who you serve, and what makes your work different from everyone else in your market. When that document lives inside your Claude project or your ChatGPT custom GPT, the AI isn’t guessing based on past patterns. It’s working from a source of truth.
Those are not the same thing, and the difference shows up clearly the first time something goes wrong.
Why Brand Foundation Comes Before AI Tools
I spent the first six years of my marketing career learning this the slow way.
I joined Guild Mortgage knowing marketing but knowing nothing about mortgage. I sat outside the offices of top producers for months. I listened to how they explained rate locks, what language they used with first-time buyers versus investors, how they described the closing table experience. I built content for them by absorbing everything they said until I could write in a voice that sounded like theirs.
That process was essentially manual brand extraction. What I was doing, call by call and conversation by conversation, was building the context that made the ghostwriting work.
AI doesn’t do this on its own. You can open Claude or ChatGPT and tell it you’re an expert copywriter, a top-producing loan officer, a trusted mortgage advisor with twenty years of experience. The tool will accept the role. What it won’t know is the story behind how you got there. The client call where something shifted. The deal that taught you something you still think about. The reason you get out of bed to do this specific work and not something else.
That is what content is made from. Not the role. The story behind it.
When I do a Brand Builder with a client, the 90-minute interview I run isn’t about their services or their production numbers. It’s about who they are underneath all of that. By the end of it, we have something closer to 150 pages of documented context, and the clients who read that document for the first time tend to say some version of: this is exactly what I’ve been trying to say but couldn’t find the words for.
That document is what goes into the AI. The output on the other side stops sounding generic because the input finally isn’t.
The Specific Problem With Outsourcing Your Sound
There’s a version of AI content use that I see constantly, and it’s the one that’s quietly eroding trust in industries where trust is the entire business model.
A loan officer, a realtor, a financial advisor sits down to write a post. They feel the friction of not knowing what to say or how to say it. They open ChatGPT, type something like write me a LinkedIn post about the current mortgage market, and copy whatever comes out. Maybe they change a word or two. Then they post it.
The content looks professional. The grammar is clean. Nothing about it is technically wrong.
And yet the people who know them can feel that something is off. The content doesn’t sound like the person they got coffee with last Tuesday. It doesn’t have the dry humor or the specific way that person talks about their clients. It has no texture. It could have been written by anyone who does what this person does, which means it’s effectively doing no work for the relationship.
I said something on a podcast recently that I want to say again here because I think it’s easy to dismiss until you really sit with it: information is no longer a unique value proposition. Anyone can get a reasonable answer to a mortgage question from an AI tool in about thirty seconds. The reason someone would choose to work with a specific loan officer in a specific market has almost nothing to do with access to information and almost everything to do with trust in a person.
Your content is the thing that builds that trust before anyone picks up the phone. When the content sounds like it came from no one in particular, the trust doesn’t build. It doesn’t exist. And the person who might have become your next client keeps scrolling.
What “Human in the Loop” Actually Means in Practice
I recently helped a coach named Michelle Berman-Mikel build what she described as a digital version of herself inside ChatGPT, something her coaching clients could interact with when they needed her feedback on a prospecting message but she wasn’t available.
The reason it works is not because the AI is particularly sophisticated. It works because we built it on a deep extraction of how Michelle actually thinks. We went through her existing chat history. We documented her coaching frameworks. We gathered real examples of messages she’d rewritten for clients over the years. By the time the GPT went live, it had enough of Michelle’s actual perspective in it that the output landed the way her clients expected.
That’s the standard. Not “does this sound like a professional” but “does this sound like this specific person.”
Getting there requires the person to be genuinely present in the process. You can’t outsource the extraction and expect the output to hold. Someone has to do the interview, answer the questions honestly, contribute the real stories and the actual opinions. The tool then takes that material and scales it.
That’s using AI to multiply who you already are. The alternative, using AI before you’ve done that work, is just producing content faster without producing anything worth reading.
The Brand Book as Infrastructure
When my ChatGPT account lost its memory last fall, I took my brand book, opened a fresh account, uploaded the document, and told the tool to use it as its source of truth. By the end of the week, I was producing content that sounded like me.
The account I’d lost had two and a half years of accumulated training in it. The brand book had been built in a single intentional process. In terms of output quality, the fresh account with the document outperformed what I’d rebuilt in the glitched one.
That was the proof I needed that documented foundation beats accumulated history every time.
Most people using AI tools right now are doing the equivalent of keeping all their important files saved only in a platform they don’t control. The brand book is the backup. It’s the file that travels. It’s what makes the tool portable across platforms and personnel. If your VA changes, if you switch from ChatGPT to Claude, if the account glitches, the document survives.
Building it is the work that only you can do. Someone can ask you the questions. Someone can write it up. But the answers have to come from you, from the actual story, the actual beliefs, the actual way you work with people. No AI can extract that from nothing. It can only work with what you give it.
What This Means If You’re Using AI Right Now
If you’re already producing content with AI tools and it’s landing well, that’s worth examining honestly. Is it landing because the tool knows you well, or is it landing with a generic professional audience in a way that could belong to anyone?
If you’ve ever read your own AI-assisted content and thought something felt slightly off, or if a client has ever seemed slightly surprised meeting you in person after following your content for a while, that gap is usually a foundation problem, not a tool problem.
The tool is only as good as what you give it.
Comments +