Free resources

Set Up Claude So It Already Knows You

A how-to. Twenty minutes once, and you stop re-explaining yourself forever.

July 2026Twenty minutes, once

Written July 2026. Menus move, so if a button isn't where I say it is, look nearby. The idea doesn't change.

If you use Claude in a fresh chat every time, you're starting from zero every time. You paste in the same background, re-explain who your clients are, re-describe your business, and then get something generic anyway because it doesn't know anything about you. That's why it feels beige.

A Project fixes it. You load your context once and it's there permanently, in every conversation inside that project.

This takes about twenty minutes and it's the highest-value twenty minutes you'll spend on any of this.


What a Project actually is

A folder for one area of your life, holding two things: standing instructions it reads before every reply, and files it can look at whenever it needs them.

That's it. Instructions and files.

You want a separate one per area, not one giant one. My clients, my content, my household, whatever your areas actually are. Mixing them makes every answer slightly worse, because it's holding context it doesn't need.


Step by step

1. In the sidebar, find Projects and create a new one.

2. Name it for the area of your life, not the tool. "Client work" rather than "Writing help."

3. Open the project's instructions. This is the part that matters, and there's a fill-in version below.

4. Add your files. Drag them in.

5. Start every conversation about that area inside the project from now on. This is the step people skip, and then wonder why nothing improved.


What to put in the instructions

Copy this, fill it in, paste it. Delete anything that doesn't apply to you rather than leaving it blank.

WHO I AM
I'm [name]. I [what you actually do, in one sentence].
I [worked in / still work in] [field] for [how long], so assume I know
[the things you don't need explained] and don't assume I know
[the things you'd rather it explained].

WHO I SERVE
My people are [who they actually are, specifically].
They come to me when [the moment they show up].
The sentence they say to me most often is "[an actual sentence one of them said]".

HOW I WANT YOU TO WORK
Ask me questions before you write anything long.
Give me the work itself, not a description of it.
If my request is vague, say so and ask, instead of guessing and producing
something I have to throw away.
If you don't know, say you don't know.
When you suggest something, tell me why that and not the simpler option.

HOW I WANT YOU TO SOUND
Write the way I write, not the way the internet writes.
No em dashes. No emoji.
Never use these words: [your own banned list].
Short paragraphs. Contractions. Say the whole thought instead of
making it a clever fragment.

WHAT NOT TO DO
Never [what you'd hate].
Don't [the pattern that annoys you].

The two lines that do the most work are the actual sentence one of your people said, and your banned word list. The first gives it a real voice to write toward instead of an imagined audience. The second is the fastest way to stop output sounding like a LinkedIn post.


What files to put in

Good candidates:

  • Things you wrote yourself. Five to ten pieces. This is how it learns your voice, and it works far better than describing your voice.
  • Your offer or service description, as it actually is, not the polished version.
  • Real examples of the work. Past plans, past emails, past reports, with names taken out.
  • Your rules. Pricing, what you don't take on, how you word difficult things.

What not to put in:

  • Anything with a client's identifying details in it. Strip names, dates of birth, account numbers, medical details. This is a real obligation if you came out of healthcare, finance, or law, and "it's just for me" isn't a defense.
  • Forty files. Fewer and better beats more. Ten strong examples of your writing beat a hundred documents you never look at.
  • Anything you'd be uncomfortable seeing quoted back at you.

The thirty-second habit that keeps it good

When it produces something wrong in a way that annoys you, don't just fix that reply. Open the instructions and add one line so it never does that again.

That's the maintenance routine. Mine has grown a line at a time for months, and almost every line exists because something irritated me once.


What it looks like when it's working

You open a conversation and type "draft the follow-up for the Thursday call" and it comes back sounding like you, aimed at your actual person, in your format, without you explaining anything.

The first time that happens you'll feel slightly robbed of the forty minutes you used to spend, and then you'll get over it.


If it's still generic after all that

You're chatting outside the project. Easiest mistake to make, and the most common. Check where you are.

Your instructions describe your voice instead of showing it. "Warm and professional" means nothing. Put in three things you wrote and say "like this."

You didn't say what not to do. Guardrails do more than instructions. Spell out what to avoid, not just what to include.

Your ask was vague and the project can't save you from that. Context makes a good ask better. It doesn't turn a bad one into a good one.

Everything here is free

There's more where this came from

I write every week about AI, momhood, and entrepreneurship.

Subscribe on Substack