The Documents That Changed How AI Writes for Us
Six documents that give AI the business knowledge it's missing
Was this newsletter forwarded to you? Sign up to get it in your inbox.
You ask AI to help with something for your business. An email, a brief, a bit of copy. What comes back is about as fine as Ross was in that episode of Friends.
What it gave you wasn’t wrong. It just wasn’t you.
So you try to fix it. More context, more back and forth, more explaining what makes you different and who your customers actually are. Three rounds later (if you’re lucky) you’ve got something you can live with—and you’ve spent longer on the prompt than it would’ve taken to write the thing yourself.
The problem isn’t your prompting. AI just doesn’t know your business—not the way you do. And no amount of back and forth is going to change that.
What we built
At Wordnerds, we spent a few months building what I’ve started calling foundational documents—a set of reference documents that give AI the business knowledge it needs before you give it any task. These are part of a broader framework I’m developing for getting genuinely useful output from AI. But this layer—the foundation—is where you’ll notice it most.
Here’s what’s in ours.
Jobs to Be Done—why our customers actually buy. Not because they want “customer feedback analytics” but because they’re drowning in data they can’t make sense of, they’re making decisions on gut feel because they can’t analyse the data quickly enough, and they want to do justice to what their customers are telling them but can’t. Bob Moesta’s JTBD framework gave us the structure for this—it captures the emotional and social reasons people buy, not just the functional ones. That’s where the really interesting insights live. When AI has this, it stops selling features and starts speaking to the reasons people actually buy.
Positioning—who we are, what we do, and what makes us genuinely different from the alternatives. Not “we’re the best at customer feedback analysis” (every competitor says that), but the thing we do that they can’t claim. Shout out to April Dunford’s updated work on differentiated value—we built our positioning around it. When AI has this, everything it writes reflects our actual position in the market rather than defaulting to industry boilerplate.
Personas—detailed profiles of our actual customers, built from real conversations. Not demographics—their goals, their struggles, the language they use. We have three: Anna the overwhelmed analyst drowning in spreadsheets, Mark the strategic manager trying to turn insights into action, Sarah the transformation leader trying to shift from gut-feel to data-driven decisions. Give AI these and it stops writing for “businesses” in the abstract and starts writing for specific people with specific problems.
Mindset Funnel—what’s blocking our customers at each stage between discovering us and buying, what questions they’re asking, and how they’re feeling. Diego de Jodar’s framework gave us the structure here. This means AI can address the exact objection someone has at each point in their journey, rather than writing vague “we’re great” copy.
Messaging House—the actual language our customers use to describe their problems, organised by job and stage of the purchase journey. Real verbatim, real emotive language. When AI has this to lean on, the copy it writes feels like it was written by someone who’s actually listened to your customers—because it was built from exactly that. One thing to watch: AI can latch onto certain phrases from your customer language and either overuse them or drop them into contexts where they don’t quite make sense—where the reader doesn’t have the background of the original conversation to make it land.
Tone of Voice—how we sound. Not how our customers talk (that’s the messaging house), but our personality, our style, the things we’d never say. Without it, AI has the knowledge and the messaging but not the voice.
I can’t list these documents without mentioning the SYSTM growth course, led by Matt Lerner and Nopadon Wongpakdee. Getting these documents right is really important—they’re foundational to everything else—and the course takes you through the whole process of building them. It’s been instrumental to our growth efforts and pointed us to the practitioners behind each framework—Dunford on positioning, Moesta on JTBD. The documents are ours, built from our own customer research, but the course gave us the structure.
Remember, these are our foundational documents for marketing and growth. Depending on your role, yours will look different. A product manager’s might include a product vision, user research synthesis, and a glossary of platform features so AI knows what you mean when you reference specific functionality. The principle is the same.
The difference it makes
We rebuilt the Wordnerds website recently. One of the tasks was writing a problem section—explaining why organisations struggle with customer feedback.
Without foundational documents, AI produces something like this:
Many organisations struggle to analyse customer feedback effectively. Data is siloed across platforms, teams lack the tools to analyse at scale, and valuable insights go unused.
Not wrong. But it could be on any competitor’s website.
With the foundational documents loaded, the output was completely different. It understood which job it was speaking to, and which persona has that job—the analyst drowning in spreadsheets, the leader making decisions on gut feel because she can’t get to the data fast enough. It could consider what stage of the purchase journey this content might be serving, how they’d describe their problem at that point, and what questions they might have. It articulated how we solve that problem differently to anyone else on the market. And it did all of that in our voice.
Your foundational documents don’t just give you this powerful first step. They also raise the bar for anyone on your team. Someone writing content doesn’t need years of customer conversations behind them—the documents carry that knowledge for them.
But remember, they’re one part of a bigger framework—one I’ll continue to build on in future articles. And you won’t get perfect output every time. There’s always an element of zhuzhing—little touches to make it more human, things that spring to mind after you read it—and the critical thinking you apply is a really important layer. But the point is you’re getting 80% of the way there really quickly.
This is a process, not a shortcut
These documents are worth the investment, but some of them take time to get right. Each one is built from other things—JTBD interviews with real customers, call transcripts, analysis reports, methodology frameworks. The positioning document alone drew on our JTBD analysis, April Dunford’s framework, and plenty of internal conversations about what actually makes us different. All in, it took us a few weeks to get them to a point we were happy with—but months to get the processes in place to actually create them.
They also need maintaining. You need to keep gathering the inputs—running customer interviews, recording and reviewing calls, staying current with how your market is shifting. A persona built from conversations eighteen months ago isn’t the same as one built from last quarter’s.
It can feel like a step backwards—spending weeks on documents instead of just getting on with the work. But there’s a reason it’s worth it—research is starting to show just how much structured context improves AI output. One found that feeding AI structured brand guidelines improved creative content quality by 69%. In medical research, giving AI a structured knowledge base to work from eliminated hallucinations entirely—down from a 40% error rate with no context. The discipline is starting to be called “context engineering”—designing what AI knows before it starts working, not just what you ask it to do.
Making them accessible
Building the documents is one thing. Making sure AI actually has them when you need it is another.
If you’re using AI in the browser—Claude, ChatGPT, Gemini—most now have a Projects feature (or equivalent) where you can upload files that persist across conversations. Each foundational document becomes a file in the project. Every time you start a new conversation within that project, AI already has the context. No re-explaining.
I use Claude Code, which works directly with files on my device. The foundational documents sit in a folder that Claude can read whenever it needs them—no uploading, no copy-pasting, always current. It’s the setup that’s made the biggest difference for us.
You’ll still want to tell AI to draw from them—something as simple as referencing the documents in your prompt so it knows to use them. With Claude Code, I can point it to the folder they sit in and it pulls from them as needed. In browser-based tools, everything sits flat in the project, so being a bit more explicit about which documents to lean on helps.
Either way, it’s the same idea: the knowledge needs to be accessible, not provided from scratch every time you start a new conversation.
This is the second in a series on getting better outputs from AI. The first was about voice input. If this has you thinking about what your own foundational documents might look like, I’d love to hear what you come up with—leave a comment or drop me a message. :)
PS — If you want to go deeper on the frameworks beyond the AI angle:
April Dunford — Obviously Awesome (updated edition). The definitive guide to positioning your product around differentiated value.
Bob Moesta — The Jobs to Be Done Handbook. How to understand why your customers actually buy.
Matt Lerner — Growth Levers and How to Find Them. I also can’t recommend the SYSTM growth course enough—it’s where we learned most of these frameworks.
Diego de Jodar — His mindset funnel session is the best introduction to the concept.
This is really interesting Steph, thank you. I feel I’m very much in catch-up mode: I tried earlier models of GenAI and thought, meh…Then along came Claude and I went into sprint mode. I’m genuinely excited about what this means means for education.