To get non-generic marketing output from AI, give it structured knowledge to draw on, built as separate, single-purpose documents rather than one giant brief. The core set is positioning, point of view, customer pain, brand voice and product knowledge. Build them in priority order, keep each one short and current, and feed the AI only the relevant pieces for each task. A sales-call summary loop keeps your customer insight fresh instead of frozen.
Key takeaways
Build separate, single-purpose documents, not one master brief. It's how you dodge context rot and reuse pieces cleanly.
The core set for marketing: positioning, point of view, customer pain, brand voice, product knowledge.
Build in priority order. Positioning and customer pain move output the most, so do them first.
Feed the AI only the pieces a given task needs, not the whole library every time.
Use AI to summarise sales calls and funnel real customer pain back in, so the knowledge stays alive.
What documents does my AI need?
Five, for marketing, and each does one job. Positioning: what you do, who for, and why you're different, in plain language. Point of view: the things you believe about your market that not everyone agrees with, the opinions that make your content yours and not a summary of the category. Customer pain: what your buyers are actually struggling with, in their words, not your feature list. Brand voice: how you sound, with real examples of on-voice and off-voice writing so the difference is unmistakable. Product knowledge: what you sell and how it works, framed around the problem it solves rather than the spec sheet.
The principle underneath all five is the one this series started with. You're building structured, separable knowledge, not one master document. Five focused pages you can mix and match will always beat a single sprawling brief, because the model does better with a clean, relevant set than with everything you own dumped in at once.
Why separate documents instead of one big brief?
Two reasons. Practically, separate documents are reusable. A LinkedIn post needs your voice and your point of view. A sales one-pager needs positioning and product knowledge. A customer email needs pain and voice. If it's all fused into a single file, you can't hand over just the right slice, so you end up dumping the lot. Which leads straight to the second reason. Load everything every time and you invite the context rot from the first post in this series: the model's accuracy slips as the irrelevant material piles up around the bit that mattered. Separate, single-purpose documents let you give each task exactly what it needs and nothing it doesn't. That restraint is most of the quality.
What order should I build them in?
Priority order, judged by what changes the output most. Positioning first. It's the foundation everything else leans on. Customer pain second, because it's what makes content land and it's usually the least documented thing you own. Point of view third, because it's what stops your content reading like everyone else's. Then brand voice, then product knowledge. If you only ever manage the first two, you'll still capture most of the gain. Don't let "finish the whole set" stop you shipping the two that matter most this week.
How do I keep this from going stale?
Wire it to where new knowledge shows up: your customer conversations. This is the highest-leverage habit in the whole approach, so it's worth setting up properly rather than leaving to good intentions. Your sales and discovery calls are the richest source of real customer pain you have, and most of it evaporates the second the call ends. Record them, run the transcript through AI to pull out the recurring pains, the exact language buyers use, the objections that keep coming back, and feed that into your customer-pain document.
Do that consistently and something quietly useful happens. Your knowledge base stops being a thing you wrote once and becomes a living read on your market. The marketing you generate off the back of it gets sharper, because it's grounded in what buyers said this month, not what you assumed last year. This is the loop that turns documentation from a one-off chore into something that compounds, and it closes the gap between what customers actually feel and what your strategy bothers to respond to.
How do I use these once they exist?
Keep them where you work, and pull the relevant pieces into each task. In practice: store them as separate documents, and for any given job, give the AI only the ones that job needs. Writing a thought-leadership post? Voice plus point of view. Drafting sales copy? Positioning plus product plus pain. You're curating a clean brief every time, which is exactly the context-engineering discipline this series opened with, just done by hand.
How to get started (the tools I actually use)
Here's how this runs in my own one-person consultancy, named honestly as what I use rather than a setup you have to copy.
I keep structured knowledge in Claude Projects, so a given project already carries the relevant positioning, voice and point of view without me re-pasting it every time. Where I need a repeatable process rather than just reference material, I write it as a skill file, a short structured document the AI follows step by step, for example the exact way I run a blog post from idea to live. For the sales-call loop, I capture calls with Granola, then pull the transcript into AI to extract the recurring pains and the language prospects use, which feeds back into the knowledge base. Notion holds the longer-term operational knowledge. The specific tools matter less than the shape: structured documents, separated by job, kept current, fed in selectively. Any decent AI tool with somewhere to store reusable context will do the same job.
Start small. Write your positioning and your customer-pain document this week. Use them on your next three pieces of marketing. You'll feel the difference in the output immediately, and that's usually enough to make building the rest feel worth it.
FAQ
What's the minimum set of documents to start with?
Positioning and customer pain. They move output quality the most and are usually the least documented. You'll get most of the benefit from those two alone, then add point of view, brand voice and product knowledge over time.
Should I put everything in one document or keep them separate?
Separate, single-purpose documents. They're reusable across different tasks, and they let you feed the AI only what each job needs, which produces better output than loading everything at once.
How do I use AI to summarise sales calls?
Record the call (with consent), then run the transcript through an AI tool and ask it to pull out recurring customer pains, the exact words buyers use, and common objections. Feed that into your customer-pain document so your insight stays current.
Which tools do I need for this?
Nothing specialist. Somewhere to store reusable context (Claude Projects, a Notion space, or similar) and an AI subscription. The structure matters far more than the brand of tool.
How often should I update these documents?
Whenever reality changes, and at minimum whenever you notice output drifting or your market shifting. The sales-call loop keeps the customer-pain document fresh almost on its own.
