Set up your AI · the Bearing File worksheet

why this works

An AI with no context about you can only hand you the average answer. That’s not the model being short on intelligence, it’s short on yours: what you actually weigh when two good things pull against each other, what you’re really trying to do, what you’d refuse even if it worked. Write that down once, hand it to the machine, and the average answer stops being the only answer available.

the questions

A tip before you start: talk, don’t type. Use your tool’s voice mode - or a dictation app like Wispr Flow if you want talking as your default everywhere - and answer out loud, the way you’d tell a friend. More of the real context comes out of you speaking than typing.

Open a notes app. Answer these about real situations, not how you’d rate yourself. Nobody sees this but you, so don’t perform it.

The last decision you put off, and why.

Not a hypothetical. An actual decision you’ve been avoiding. What made it hard to just make?

What you’d refuse even if it worked.

Something effective that you still wouldn’t do, because of what it would cost you to do it.

What people consistently come to you for.

Not your job title. The actual thing people seek you out for, again and again.

Where you’re certain, and where you’re honestly not.

Name one of each. Most people can only find the certain half without prompting.

What this season is actually about.

Not your to-do list. The two or three things underneath it that actually matter right now.

assembly guide

You now have five honest answers. Turn them into whichever of these your tools support - use more than one if you’ve got the tools for it.

Memory entries

Paste into ChatGPT’s memory (Settings › Personalization › Memory) or a Claude Project’s memory setting. Replace the bracketed lines with your five answers, kept short.

Context worth remembering about me, across every conversation: - [What I'm actually trying to do right now - the real goal, not the task list] - [What I'd refuse even if it worked, and why] - [What people consistently come to me for] - [Where I'm certain, and where I'm honestly not] - Don't flatter me. Challenge me, and ask before you assume. - Tell me what you're inferring versus what you actually know.

Custom-instructions block

Paste into ChatGPT’s custom instructions or a Claude Project’s instructions field. This is the standing behavior, not the facts.

Use the context you have on me for everything I ask, not just when it's obviously relevant. If something I ask doesn't fit who I am or what I'm building, say so before you answer it. Don't agree with me by default - if you actually disagree, disagree. Ask a clarifying question before you assume what I mean. When I correct you, treat that correction as permanent, not a one-time fix.

A CLAUDE.md-style bearing file

For Level 4 and up: save this as CLAUDE.md at the root of a working folder, or the equivalent for your coding agent.

# Bearing File ## Who I am [One paragraph. The situation you're actually in.] ## What I'm building [The real goal this season, not the task list.] ## What I value when two goods conflict [The tradeoffs you actually face, and which way you lean.] ## What this season is about [2-3 things that matter now. What you're deliberately not doing.] ## How to work with me - Challenge me. Don't agree by default. - Tell me what you're inferring versus what you know. - Ask before you assume; don't fill gaps with a guess. - Optimize for what's actually right, not for what looks fast. - When I correct you, the correction is permanent. Update this file.

what you just did

It has a shape. A question that didn’t have an answer yet. A messy middle where the words didn’t sound true. Then, somewhere in the honest answers above, a sentence that finally sounded like you.

That shape has a name, and there’s more to it than a worksheet. Read the full argument →

And if you want a measured read of where you stand, the WILDiQ self-assessment is free ↗.