The short version
If you paste the same background into ChatGPT or Claude every time you start a chat, a project fixes that.
A Claude project is one workspace with three things in it: your files, your instructions and your chats. Set it up once, and every new chat inside it starts out knowing your business.
This is the written version of the video. It was recorded October 26, 2026. Menus and plans change, so check what your account shows.
Why agents should use projects
- Consistency. Every new chat starts with the same knowledge about you, because you chose what it knows. That's when it starts to feel like it knows you.
- Files. Your business brief, example emails, the way you like a listing described. Upload them once instead of pasting them into every chat.
- Instructions. The rules it follows every time, like "always ask me before anything goes to a client." When it does something you don't like, you fix the instruction once and it stops. The more you use it, the better it gets.
- Organization. Every chat for that job stays under the project instead of in one long list of past chats.
Every Claude plan has projects, including the free one. Free accounts can create up to five.
Step 1: one project per job
Click Projects in the left sidebar, then New Project in the upper right. Give it a name and, if you like, a description. Those are for you; Claude can't see them.
The rule I use: one project per job you do over and over. Not one per topic. One per job. Writing your newsletter is a job. Answering new leads is a job. Writing listing copy is a job. Each one gets its own project.
Step 2: add project knowledge
This is the part that makes it all work. Click the + in the project knowledge area on the right, and upload the background you would otherwise paste into every chat.
Good things to put in:
- a short brief about your business;
- your style notes and the words you never use;
- examples of copy you love.
Write your business brief once, as one file, and add it to every project. Then add a file or two that only that job needs.
Let Claude write the brief with you. Open a regular chat and paste this:
I want to build a Claude
project for [the job].
What should go in its knowledge?
Interview me one question at a
time, then write everything as
one file I can upload to the
project.Answer its questions, and it hands you a file you can upload to the project. Use AI to make the things the other AI needs.
The limits, plainly: each file can be up to 30MB, and you can add as many as fit in Claude's context window. On paid plans, Claude can stretch project knowledge further when it gets close to that limit (Anthropic says by up to 10x). In private projects you can also add Google Docs from Google Drive, and they stay synced with Drive.
Step 3: write the instructions
Click Set project instructions, paste yours in, and click Save instructions. These rules apply to every chat in this project and only this project.
Keep them to three things: the job, where the facts come from, and what to do when a fact is missing. Then add two more lines. Here's the set I use:
You help me with
[the job].
Use the files in this project
for facts, voice and details.
If a fact isn't in the files or
my message, ask me instead of
guessing.
Name the file each fact came
from.
Nothing goes to a client until I
approve it.The line that matters most: ask me instead of guessing.
The three lines I use for a listings project, ready to copy, are on the Claude project instructions page.
Anthropic's help pages disagree on whether free accounts get project instructions, so check whether your plan shows the button. If you don't see it, that's your plan, not you.
Your whole Claude account also has its own instructions, apart from any project. Click your initials in the lower left, then Settings, then Account and Profile (some newer accounts show Your Claude here instead), and find Instructions for Claude. Put the basics there that every chat should know: your name and your brokerage. Each project's instructions add to that, for one job.
Step 4: give it a real job
Start a new chat inside the project and give it a real task. It uses your files without being told.
The agent example: Dana's business brief
Here's the first project I'd build as an agent, with a made-up agent so you can see it work. Dana Brooks and Brookside Realty are fictional, and so is the listing.
Create a project called Agent business brief and upload this as a file:
Dana Brooks, agent at
Brookside Realty.
Market: Millbrook. Known for:
first-time buyers and townhomes.
Voice: warm, direct, no hype
words ("stunning", "must-see").
Always: describe the property,
not who should live there. No
school or safety claims.
Sign-off: "Dana Brooks,
Brookside Realty."Then set the instructions:
Write as Dana. Use only
the facts from the brief or my
message. If something is
missing, ask.
Name the file each fact came
from.
Nothing goes to a client until I
approve it.Now open a chat in the project and paste prompt 1 from the free prompts guide, the listing description, with a sample listing:
88 Willow Bend Drive
(sample listing): 3 bed, 2.5
bath townhome, 1,620 sq ft,
kitchen updated 2023, attached
garage, walking path behind the
unit.You get three versions in Dana's voice, and it asks about what's missing instead of making it up.
Swap in your own details and you have a project that already knows your business.
A second project: listings
The next one I'd build is a listings project. Its instructions say what to do when you hand it a new listing: take everything you paste in about the house and draft the MLS description and the rest of the listing copy you need. Then you go back and forth to fine-tune.
Read every word before it goes out. You're the one who sent it.
Memory: what carries over
Memory is a separate layer, and each project gets its own. Inside a project, Claude uses that project's memory plus your memory from chats outside projects. It never uses another project's memory.
Memory is on by default on the Free, Pro and Max plans. I turn it off and put what I want remembered into my instructions, so I decide what it knows.
If a project holds client details, turn off Use account memory under Memory on the project's page. That keeps the project separate from your regular chats.
Chats inside a project don't share their context with each other; files do. With memory on, something you said in one chat can still come up in another through project memory. So anything every chat should know belongs in project knowledge, not in a chat.
Keep it current
A project is only as good as the files you give it. If the knowledge is old, the answers will be wrong.
- When your facts change, replace the file. Don't keep old versions side by side.
- Ask AI to help: paste the file and say what changed, then upload the new version.
- If a teammate's version works better, ask them for the file.
Today you swap the files by hand. It's a little annoying, and it's worth it.
Get the prompts
The listing prompt above is one of 10 free ChatGPT prompts for real estate agents. They work in Claude too, and you don't need to sign up to read them.
Source: Anthropic Help Center, "What are projects?", "How can I create and manage projects?", "Uploading files to Claude" and "Use Claude's chat search and memory," checked October 2026.
