How to write your business down so AI can actually use it

The advice is always the same: write your business down so AI knows what you do. So you open a blank document, type a heading like Company Overview, look at it for a while and decide to do this properly when things are quieter. Things are never quieter.

That blank page is where most businesses stop climbing, and it isn't laziness. Nobody tells you what goes on the page. Written down like that, level two sounds like a writing project with no brief and no finish line, so it loses to every email in the inbox.

It isn't a writing project. You are answering questions you already answer out loud every week, in the words you already use, and the whole job fits in an afternoon or two.

Why writing your own business down is harder than it sounds

Everything about your business is obvious to you, and obvious things are the hardest to describe. You know the discount only starts above 20 units and that the Tuesday delivery run covers one city and nowhere else. You know which supplier needs chasing twice before anything moves. None of that feels like information. It feels like Tuesday.

Then there's the audience in your head. The moment you type a heading, some part of you starts writing for a customer or a banker, and that version comes out polished, careful and worth nothing to a machine. Nobody has ever put "we usually forget to invoice the small jobs until month end" in a company overview, and that is exactly the sort of line that makes AI useful to you.

So change who you are writing for. Write for a capable new person who starts on Monday and has to answer a customer correctly on their first morning without asking anyone. That instruction does most of the work on its own, because it drags the exceptions, the awkward cases and the real prices onto the page. Those are the things a new hire gets wrong, and they are the same things AI gets wrong.

If the blank document still wins, stop fighting it and let the AI ask you instead. Answering questions is easy, and 20 minutes of being interviewed produces more usable material than an hour of staring at a heading.

What your AI is actually missing

Everything AI knows about your company comes from two places, which gives us a short equation worth keeping.

Context = Knowledge + Data.

Knowledge is what your business knows, written down. Your prices, your process, your service promises, the way your team talks to a customer who is unhappy. Most of it currently lives in the heads of three or four people and has never been typed anywhere.

Data is what your business records, connected. Your sales, invoices, stock, ad spend and enquiries. It exists already, but it sits in separate systems and nothing reads across all of it.

If a thing is neither written down nor connected, your AI does not know it exists, and it will fill the gap with an average of every business on the internet. That is where the generic answer comes from.

This article is about the knowledge half, and you should take that half on its own. It is the part you can finish this month with no budget, no software and nobody technical. Connecting your data comes later, needs someone with access to your systems and is worth much less while the knowledge is still missing, because numbers with no explanation attached produce confident nonsense.

What to write first, and why the order matters

There are ten documents. You will not write all of them in one sitting, which is why the order is the useful part. The first three fix more wrong answers per hour of writing than the other seven combined, and a wrong fact contaminates everything built on top of it. Correct tone applied to last year's price is still a bad quote sent to a customer.

Start with the facts that make answers wrong. What you sell, every product or service by name with what is genuinely included in each. Your prices, the current ones, with the discounts and who is allowed to give them. How you quote, meaning the path from an enquiry to a number, and what makes that number go up or down. Nearly every embarrassing AI answer traces back to one of these three.

Then write who it is for. Who buys from you, the main types of customer and what each one actually cares about, because copy aimed at everyone reads as copy aimed at nobody. Who you turn away, the enquiries you say no to and why. That second one gets skipped constantly and it saves your team and your AI from chasing work you don't want.

Then how the work happens. How you sound, the words you use, the words you never use and one reply you were happy with. Your process from start to finish, what happens between the customer saying yes and the job being done and paid for. The questions you answer every week, your real FAQ, in the answer you would give on the phone rather than the one on your website.

Finish with the guardrails. Your numbers, which figures you track, where they live and what a good month looks like next to a bad one. Your rules and limits, what nobody in the business may promise a customer, what needs your approval and what should never leave the company. Write that last one before you hand AI anything sensitive rather than after.

Each of those is a separate document with a plain name, saved in one folder you can find again. They are ordinary text files in your own account, which is worth saying because it is the thing that makes this permanent. When a better AI model arrives next year, you point it at the same folder and carry on. Nothing here ties you to one company's software.

How much detail is enough

A page each is plenty. Some come out at half a page, and the pricing document might run to three if yours is complicated. That is fine.

The bar is not polish. It is true, current and specific. Bullet points are allowed. Half-finished sentences are allowed. Nobody is marking this and the only reader that matters can handle rough writing far better than it can handle a smooth guess. A line like "3 working days, except custom sizes, which take 3 weeks and always get quoted separately" is worth more than a paragraph about your commitment to quality.

Two habits make these documents far more useful than they look. Write the exceptions, because the exceptions are where every wrong answer lives. And write the reasons, not only the rules, because "we don't take jobs under RM2,000" tells AI what to do while "we don't take jobs under RM2,000 because setup costs the same whatever the size" lets it handle the case you didn't think of.

Then apply one test to each document before you close it. Could a capable new person answer a customer correctly using only this, on their first morning, without asking anyone? If yes, the document is finished. If no, the thing they would still have to ask about is the thing you left out.

The four mistakes that make written context useless

Out-of-date prices are the worst of them, and they are worse than having nothing written at all. With no price on file, AI hedges or asks. With last year's price on file, it answers confidently and somebody forwards that answer to a customer. Put the date you last checked at the top of the pricing document, and make the rule that whoever changes a price changes the document the same day.

The second is describing the business you meant to build. Process documents drift towards the version that runs properly, with the approval step everybody skips and the handover nobody does. Write the real one, including the part where someone checks something manually on a Friday because the system doesn't do it. Everything you build later assumes what this document says, so an aspirational process quietly produces work that assumes a step that never happens.

The third is writing the website version. You will feel the pull towards marketing language the moment you describe what you sell, and what comes out is "end-to-end solutions tailored to your needs" instead of "we install and service commercial air conditioning, and we do not do residential". The first sentence could belong to 10,000 companies. The second one is yours, and it is the sentence that makes the answers change.

The fourth is leaving the documents where AI never sees them. A context folder nobody opens does nothing. Paste the relevant document in at the start of a conversation, or upload the folder into a project or custom GPT once so it stays there. Written down and unused looks identical to never written.

What changes once it exists

The first thing you notice is that you stop re-explaining yourself. Every conversation used to start with two paragraphs of background about who you are and how you charge. Now it starts with the actual question.

Quotes come back with your prices and your payment terms in them. Replies to customers sound like your company rather than a friendly stranger. Ask for a follow-up to a quote that has gone quiet and you get the version that mentions the specific thing that customer was worried about, because you wrote down what your customers worry about.

The surprise benefit is that you have accidentally written your induction pack. The same ten documents that make AI useful are what a new hire needs in week one, and most businesses have never had them.

The bigger change is that level three becomes possible. A skill is a repeatable job written down once and run whenever you need it: the Monday report, the reply to a late payer, the product description. Built with no context underneath, a skill produces plausible rubbish very quickly. Built on documents like these, it produces work you would put your name on. More than 15 routines run our own businesses every day, including the reporting, the meeting digests and the finance checks, and every one of them sits on top of documents that started exactly this way.

The test for level two is short. Ask the AI something only someone who works there would know, like what you would charge for an awkward job or how you would answer a complaint about a late delivery. When the answer sounds like a colleague rather than a stranger, you are done.

Where to start today

Set aside 20 minutes this week and use the Context Interview Prompt. It's free, you paste it into whichever AI chat you already have open, and it interviews you one question at a time about what you sell, what you charge, who buys from you and how the work gets done. At the end it writes the whole thing up as a document you keep. That covers most of the first sitting for you, and it beats a blank page every time.

If you'd rather see where your business currently stands before you write anything, the AIOS Score takes a few minutes and tells you which of the five levels you're on and what the next move is.

And if you want the full picture, with every department mapped and a ranked list of what each move is worth in ringgit, that's what the AIOS Audit is for.