If your AI keeps handing you generic answers, here is the actual cause: it knows little about you. You are asking a brilliant stranger for advice about a business it has never seen. No wonder what comes back sounds like it could apply to anyone.

In a recent piece I wrote that AI comes in two levels, the one where it knows you and the one where it works for you. This is about reaching the first, because almost everything good downstream depends on it, and it is simpler than it sounds.

For a while I thought the skill to build was prompting, getting the wording just right. In my own use, that stopped mattering. What changed my results was giving the model the right information to work with, not better phrasing. That is closer to handing over the file than asking a clever question.

You might be thinking the tools already remember you. They do, a little, and that is not the same thing. What a chatbot picks up on its own is whatever it happens to catch as you talk, scraps you cannot see and did not choose. A context document is the opposite: it is what you decide it should always know, written down once, in your words. One is what it overheard. The other is the file you handed it.

Here is what that looks like in practice. Skip the blank page. Let the AI interview you instead. Open a chat, tell it you want to build context about yourself and your business, and answer one question at a time. The sharpest set of interview prompts I have found for this is Allie K. Miller's, which I adapted to my own business. Ten minutes in, it turns your answers into a first page. That page might read something like this:

We are a [what you do] serving [who you serve]. This year we are trying to [top goal], and the thing in the way is [main constraint]. When I make a call, I weigh [how you decide]. We do not compromise on [your line]. Write to me plainly and directly, and always give me the honest read, not what I want to hear.

Start brief. That page can be incomplete. It just needs to exist. But do not stop there, because one page only answers what you do today. It says nothing about who you are or where you are headed, and both of those shape every answer the AI gives you just as much as the business itself does.

I keep three files running, and if you build nothing else, build these. The first is who I am: my values, how I decide when the easy answer and the right one are not the same thing, the line I will not cross no matter what a deal looks like. The second is what I am building: the business itself, the page above, this year's goal and the constraint in the way. The third is where I am headed, not just this year's target but the ones past it, so the AI understands what today's work is actually building toward, not only what is due this quarter.

The first page takes ten minutes. Getting it to actually sound like you, and to catch the decisions that matter, takes a few weeks of correcting it. That is the honest shape of the payoff.

None of the three are ever finished. I correct the first file when I catch myself making a call that does not match what I said I believe. I update the third whenever a target moves. Think of it less as a document and more as an ongoing conversation with yourself that the AI happens to sit in on.

Two more files earn their keep once those three exist. One on your voice, so what comes back sounds like you and not a brochure, the one I keep refining most because the closer it gets the more useful everything else becomes. One on your ideal customer, so its help on your marketing and your sales is grounded in your real buyer instead of a generic one.

Here is the part that makes it close to effortless: I do not type any of this, I talk. I dictate most of what I write now, including my answers when the AI interviews me. By my own count it saves me three to four hours a week. Ten minutes of talking is nothing, and you end up with a better document than you would have typed, because you say more out loud than you do staring at a cursor.

Now, where does this document live so the AI actually uses it every time? You do not paste it into the chat and hope. In both ChatGPT and Claude, you create something called a Project, a named workspace you set up once. You add your document and a few standing instructions, and from then on every conversation you start inside that Project already knows your business. It is the same simple move in either tool, built into the AI you are already using. Nothing to install, and nothing technical about it.

The exact setup differs a little by tool. One keeps your standing instructions in one spot with your files beside them; the other splits them up a bit differently. You do not need to memorize either. If you get stuck while setting it up, ask the AI directly: "How do I best set up context files in you?" It will walk you through its own version. That is the whole theme of this piece anyway. Stop guessing, and let it tell you.

The part that surprises people is how much these files improve by being corrected. Two habits keep mine sharp.

The first is simple. When something it writes does not sound like me, I say so, and I tell it to fix the document so it does not happen again. Early on it handed me a line, "how does that land with you," and I knew in a second that is not how I talk. So I told it, and we added that phrasing to a running list of words it should never use in my voice. It does not use them now. Every correction like that makes the next thing it writes a little more me.

It works for more than wording. If it keeps using em dashes, or lays out a note in a way you dislike, say so and tell it to write that rule into the file. You do not need the technical terms for this. Plain language does it: "I keep seeing this, fix it, and put the rule in the file so it stops." It holds because the fix lives in the document, so every new chat inherits it.

The second is about not losing good work. One day I worked a real decision through with it, start to finish, and reached a clear answer. The next morning it was gone. It had never saved the conclusion. My first instinct was that the tool had failed me. The truth was the opposite, I had not told it to remember. So I changed how we work. Now when we settle something that matters, I have it write the outcome into its memory on the spot, the way you would put a decision in a file instead of trusting you will recall it next week. The lesson stuck with me. You are the one who decides what it keeps.

You do not need new software for any of this. Whatever AI you already use will do it, and you can talk your answers instead of typing them, straight from your phone or the app you already have open. So build the first page this week. Open ChatGPT or Claude, start a Project, tell it to interview you about who you are and your business, talk for ten minutes, and drop the page in. It will not sound like you yet. That is the next few weeks of correcting it, and it is where the real payoff starts.

What is the part of your business you are most tired of explaining from scratch?

The time-savings figure is my own, from my use of voice dictation.

FAQ

Isn't this just ChatGPT's memory?
No. Memory is what the AI picks up on its own as you chat, scraps it decides to keep that you cannot fully see or control. A context document is what you deliberately decide it should always know, in your words, loaded every time. Memory is a useful bonus. The document is the part you control, and it is the part that does the work.

Do I need to be technical to do this?
No. A context document is plain writing about your business. If you can describe your company to a new hire, you can build one, and setting up a Project in ChatGPT or Claude is a few clicks, built right into the tool you already use.

Isn't this just a longer prompt?
No. A prompt is the question you ask. Context is what the AI already knows before you ask it anything. You redo a prompt every time. You build context once and reuse it.

What goes in it first?
Three files. Who you are: your values, how you decide, the line you will not cross. What you are building: the business, who you serve, this year's goal and what is in the way. Where you are headed: your goals past this year, not just this one. Add your voice and your ideal customer once those three exist.

Is my information safe?
Keep client-identifying details out of public AI tools, or anonymize them first. A context document can describe how you work without naming the people you work with.

Cole Dolny is the founder of 6S Advisory Inc. and a Chair with TEC Canada, a Vistage Worldwide subsidiary, serving growth-minded business leaders. He works with CEOs and owners on leadership effectiveness, talent systems, and decision-making, helping them build healthier, more profitable businesses.