Most founders use ChatGPT or Claude the same way they Google something. Type a question, get a generic answer, move on. That works fine for looking up syntax or drafting a template. It fails completely the moment you want the AI to help you run your actual business.
The problem isn't the model. The models are good. The problem is context. You're asking an AI to help you write a follow-up email, and it doesn't know your product name, your pricing, your tone, or who you're emailing. So it gives you a reply that sounds like it was written for a completely different company. Because it was.
A personal AI assistant that actually knows your business isn't a different tool. It's the same model, set up differently.
What Does "Knowing Your Business" Actually Mean for an AI Assistant?
When people say they want an AI that knows their business, they mean three things: using real product names and prices, remembering past decisions, and matching their voice. All three are achievable with context files loaded at session start, not a custom model or enterprise tooling.
The structure is a small set of documents that get loaded before the AI does anything. The model reads them, and suddenly it knows who you are.
Which Files Do You Actually Need to Give an AI Business Context?
Three files: an identity file with your products, prices, and audience; a voice guide with actual writing examples; and a memory file that captures past decisions and lessons. Together they transform a generic model into one that works from your real context instead of inventing plausible-sounding answers.
The identity file
This is a short document that describes your business the way you'd describe it to a new contractor on day one. Product names. What each product does. Who buys it and why. Current prices. What's live vs. in development. Your positioning in one sentence.
Keep it under 600 words. Dense, not sprawling. The goal is for the AI to read it once and immediately have the working facts it needs.
Mine covers Xero AI (the agency and tools), the newsletter, the $7 beginner guide, the Build Lab, and a few lines on what Evo is and how it operates. When any tool loads that file first, it stops giving me generic AI advice and starts working with my actual setup.
There's a full template and walkthrough in How to Write an Identity File for Your AI Agent if you want the structure.
The voice guide
A separate document that captures how you write. Not instructions like "be professional" or "be casual." Actual examples. Sentences pulled from your best writing, labeled with what makes them work. Patterns you use. Phrases you never use. Words you avoid.
Without this, the AI defaults to its training distribution, which sounds like a capable but characterless consultant. With it, the output lands much closer to your actual voice on the first pass.
The voice guide doesn't need to be long. A few pages. But it has to be specific. "Conversational" tells the model nothing. "Short paragraphs, no em dashes, no filler intros, start with the problem" tells it something.
The memory file
A living document that captures what's happened, what's been decided, and what lessons shouldn't be lost. This one gets updated over time, either by you or by the agent itself.
The memory file is what turns a stateless session into something with continuity. Without it, the AI has no idea that you tried a particular pricing strategy and it flopped, or that you've already vetted three newsletter tools and settled on MailerLite. It starts fresh every time.
With it, you skip re-explaining and get straight to work.
If you want the full architecture for how this works across sessions, How to Give an AI Agent Persistent Memory covers the daily log plus long-term memory setup I've been running for months.
How Do You Load Business Context Into an AI Assistant?
In plain ChatGPT or Claude.ai, paste the three files at the top of each new conversation. Takes 30 seconds. More advanced setups auto-inject them so context is always loaded without manual work. Either approach produces dramatically better output than starting a session with no business context at all.
In more powerful setups, an AI assistant configured to load these files automatically at session start removes the copy-paste entirely. OpenClaw does this natively. According to OpenAI's documentation on persistent context, structuring input well is one of the highest-leverage improvements you can make to model output.
If you're not running an agent platform yet, the copy-paste method still beats starting from scratch. Get the files written first. The automation layer can come later.
What Should Go in the Identity File?
Product names and one-line descriptions, current prices, who each product is for, your audience described specifically enough that a stranger would recognize them, what you're building toward, what you don't do, and all active URLs. Under 600 words, formatted as bullet points. Not narrative paragraphs and not your origin story.
What to include:
Products/services: Name, one-line description, price, who it's for, what problem it solves. One row per product.
Audience: The actual person you're selling to. Not "entrepreneurs" but something specific enough that a stranger would recognize them. For me: solo founders or people with full-time jobs who want to run a side business with AI tools and can't afford or don't want a team.
What you're building toward: A sentence or two on where you're going. This helps the AI calibrate what advice is relevant vs. distracting.
What you don't do: Equally important. Things that are out of scope, partnerships you don't take, content you don't make. Guardrails on the business context.
Active URLs: Live products, signup pages, blog. So the AI never invents a URL or links to something that doesn't exist.
That's it. Keep it scannable. The AI doesn't need your origin story.
What Are the Mistakes That Kill AI Context Quality?
Overloading the files with backstory the AI doesn't need, letting them go stale after pricing changes, packing everything into one massive dump, and skipping the voice guide. Each mistake produces a specific type of wrong output. Fix them and the gap between a generic session and a context-loaded one becomes impossible to ignore.
Overloading it with backstory. The identity file is not the place for your founder origin story, your values manifesto, or a full brand strategy document. Those might matter for other things. For session context, they're noise. The model needs facts, not narrative.
Letting the files go stale. If you changed your pricing in March and your identity file still says the old number, you've introduced a conflict. The AI will use the wrong price. Update the files when things change. Treat them like product documentation.
One massive context dump. Some people try to pack everything into one document and inject thousands of words at the start of every session. This eats context window, increases cost, and buries the important stuff in filler. Three focused files, each doing one job, works better than one bloated megadoc.
Skipping the voice guide. The identity file fixes factual errors. Without the voice guide, you'll get factually accurate output that still doesn't sound like you. Both matter.
What Does Context-Loaded AI Output Actually Look Like?
Before context files, AI would produce generic tweets like "Here's how you can leverage AI to grow your business." After loading the identity and voice files: short, specific, founder-to-founder, linking to something real. Same model, same capability. The difference shows up across every task because the model knows who it's writing for.
Same model. Same capability. Completely different output because it knows who it's writing for and how the writing is supposed to sound.
The same improvement shows up across every task: email drafts, blog post outlines, customer response templates, decision frameworks. Once the context is loaded, the AI stops being a general-purpose text generator and starts being a tool that's calibrated to your actual situation.
Research from Anthropic on effective prompting consistently shows that providing structured reference context is more effective than detailed instructions alone. The model performs better when it can look up facts rather than infer them.
What Is the Next Step After Building Context Files?
Make the setup permanent so you're not managing context manually. That means an agent system built on top of context files, with scheduled tasks, self-updating memory, and tools connected to your actual stack. The $7 beginner guide covers the full architecture from identity through automation in one sitting.
But the context files alone will change your day-to-day in ways you'll notice immediately. Write the three files. Start loading them. The rest builds from there.
The full architecture, identity files through memory systems through automation, is covered in the Your First AI Agent guide at Xero AI. Start there if you want to move past context files and into an actual operating system for your business.
Related:
- How to Write an Identity File for Your AI Agent
- How to Give an AI Agent Persistent Memory
- What Is a Source of Truth Document for AI Systems
- What Is an AI Co-Founder and Do You Need One
- How to Write an AI Agent Prompt That Works
Start Building Your Own AI System
- Your First AI Agent - $1 launch-test guide, instant download. The fastest way to get started.
- Build an AI Co-Founder - the full architecture ($19).
- AI for the Rest of Us newsletter - practical AI 3x/week for people with day jobs.
Want to build your own AI co-founder?
I'm building Xero in public — an AI system that runs distribution, content, and ops while I work a full-time job.
- Start here: Your First AI Agent — $7 guide, instant download
- Go deeper: Build an AI Co-Founder — the full architecture ($19)
- Newsletter: AI for the Rest of Us — practical AI 3x/week for people with day jobs
- Site: xeroaiagency.com
Originally published at xeroaiagency.com
Top comments (0)