There's a predictable progression everyone goes through with AI context management. Most people get stuck halfway through and wonder why their outputs feel generic.
Level 1: No context - Raw prompts, generic outputs
Level 2: Basic context - Add some brand voice, slight improvement
Level 3: Context overload - Dump everything, AI gets confused
Level 4: Selective context - Right information at the right time
Level 3 is the trap. You've figured out that context matters, so you start cramming everything into every prompt. Company wiki, brand guidelines, 47 example posts, competitive analysis—the works.
When I work with clients on adapting AI into their business, this is a key thing we focus on. A lot of business can see significant value increase from AI by
find three or four highly specific tasks
engineer solid, reusable prompts for each
add the right context
wrap everything in projects in ChatGPT or Cluade
Short intermission – in case you missed it:
A few weeks ago I published the AI-Enabled Agency Playbook. It contains models, frameworks and examples from my own journey building ai-first agencies.
You can get the playbook for free here
AND:
I heard you loud and clear in last week’s poll:
The “Basics of Automations and Agents” guide is coming soon. I already started working on it, and it’s going to a be a big resource. I’m still considering how to best package it, but I guarantee that at least parts of it will be free for subscribers.
Also, to make sure the content hits right I’m going to test drive it with a few people. Live one-on-one automation coaching based on the guide.
It’s a package with 2×45 min sessions covering process mapping, tools, data, context, and building out a first automation together. It’s perfect for someone like a small business operator/owner looking to kickstart this skill.
Spots are extremely limited and priced at $350.
Reach out to me at [email protected] if you’re interested.
Ok, back to context management:
Think of your business context in two buckets:
Static context = information that's always relevant. Your brand voice, core frameworks, audience details, company values. This goes in every prompt because it never changes.
Dynamic context = information that depends on what you're trying to accomplish right now. Specific examples, relevant case studies, topic-focused data.
Most people nail the static part but struggle with dynamic context because it requires thinking through what's actually relevant for each specific task.
Here's a practical example: say you're using AI to write marketing content. Your static context includes brand voice ("professional but approachable") and content structure ("problem-solution-action"). But if you're writing about pricing, you want pricing-focused examples and objection handling, not general marketing examples.
You need to consciously pull different context based on what you're trying to create.
How to Implement This in Your Business
1: Set up a basic context library structure in Google Drive or similar:
/core
/marketing
/sales
etc.
2: Add context docs in each category Bonus: if you use Google Docs integration the context docs are live. If you upload as static files you have to remember to re-upload new versions if you make edits to your context docs.
3: Set up projects in Claude or ChatGPT and add relevant context into each project.
Example:
In a marketing project you would add the files from /core (Business context profile and other foundation docs) and files from /marketing (Brand voice, marketing strategy, examples).
Congrats! You’ve now set up your first retrieval augmented generation (RAG) system. RAG means that the LLM actively augments it’s knowledge by fetching relevant information (based on the prompt) from additional context (our docs).
Over time you can build out your context library as you use AI in more parts of your business.
When you start looking at agents and automations, getting examples and context right is half the battle. If you already have this in place it’s much, much easier to get started.
Setting up a project in ChatGPT is a simple RAG-system, but it’s the same foundation we use to build more complex autonomous systems.
Here are some examples:
A marketing orhestrator agent will have access to some context. Based on the task it’s working on it will call on specialised marketing agents (e.g. copywriter). These agents have deeper, more niche context available.
Instead of using static docs, the data can be dynamic and live. A social media performance agent may pull in the last 10 posts to analyse.
Don’t over-complicate this in the beginning. Maintaining and improving the quality of the context is as important as setting the initial structure.
Until next week,
Martin
PS: Go connect with me on LinkedIn and say hi! It’s always fun to chat with readers
PS2: Want to kickstart you business’ AI transformation? Fill out this interest form and my consultancy will be in touch asap for a free discovery call.