You know the moment.
Three weeks ago you had a genuinely useful conversation with your AI tool. It helped you untangle a proposal, sharpen a board paper, or finally find the words for something you’d been circling for days. Today you need it again.
So you scroll. And scroll. You search for a word you think you used. You find four chats called “New conversation” and none of them is the one you want.
If that’s you, you’re in good company, because it was me too. Then I started using the feature I now open every single working day: Projects.
In this edition of Everyday AI I’m walking you through what a project is, how I actually use mine, and why I think it matters far more for leaders than the slightly boring name suggests. The video demo is above if you’d rather watch than read.
Think of it as a file folder
A project is a file folder for your AI. Every chat on the same subject lives together, so when you need that conversation from three weeks ago, you know exactly where to look.
That alone would earn its place. But a project does two more things that change how you work with AI entirely. You can store files in it, and you can give it standing instructions.
Put those three together and you stop having one-off conversations with a clever stranger who needs the whole backstory every time. You start working with something that already knows the brief.
What it looks like in practice
In the video I build a test project from scratch so you can see how simple it is. In Claude, go to Projects in the upper left-hand corner, click new project, give it a name and a short description of what it’s for. Done.
Then you add context. That can be files from your own device, or material from tools you’ve connected, such as Canva, Google Drive or Notion.
Then I open the project I use most: my LinkedIn newsletters and posts. There are a gazillion chats in there, every newsletter and post I’ve worked on. Sitting alongside them are my white papers, my models and frameworks, and my positioning statement. Everything I want it to draw on when it writes with me.
And then there are the instructions, which is where the real power sits. [KIM: the video says 8,000 characters. I couldn’t verify the current limit for Claude Project instructions, so please confirm or we keep it general as written here.] You get room to write a detailed, comprehensive brief on how the work should be done: your voice, your audience, your standards, what to avoid, what good looks like.
Once that’s in place, the ask becomes almost embarrassingly short. In the demo I typed one request: write a 300-word LinkedIn post on how AI is affecting the workforce in Australia, which jobs are being hit most and fastest, and the five-year outlook.
Back came a draft of roughly 300 words (it never counts accurately, and I’ve stopped expecting it to), complete with hashtags, a suggested posting time and alternative hooks. I didn’t ask for any of those extras in the prompt. They come every time because my instructions ask for them every time.
Do I post it as written? Almost never. I always go in, tweak it and put my own spin on it. But after this many newsletters and posts, I don’t have to tweak much, because the project knows how I write.
The part that catches people out
A project is designed to ring-fence its information. It works from what you’ve given it inside that project, and it is built not to wander into another project and pull in something irrelevant.
For anyone working across clients, that’s exactly what you want. If you’ve set up a project for ABC Company, you don’t want it borrowing from what you told it about XYZ Company. You want it to know ABC, and only ABC.
So don’t expect a project to know what you told a different one. Give each project what it needs. [KIM: Claude Projects now have a memory setting that keeps a project either separate from or connected to your general memory, so “it won’t remember anything outside the project” is only true on the separate setting. I’ve worded this as the design intent. Happy with that?]
Yes, uploading the same files into several projects is a bit of a pain. You can connect a Google Drive instead and update it in one place, which is handy, but in my experience it burns more tokens, and tokens are getting expensive. So I suck it up and upload.
If your organization runs on Microsoft and your AI tool is already connected to OneDrive or SharePoint, that step is taken care of for you. [KIM: the transcript says “Microsoft co-work”. Did you mean Microsoft Copilot, or Claude Cowork connected to OneDrive/SharePoint? I’ve kept it generic until you confirm.]
ChatGPT has its own version of projects too. I’ve used both, and for this kind of work I’ve found it more expensive and not quite as reliable. [KIM: the transcript reads “Clouder Chat GPT”. I’ve read this as ChatGPT being the comparison; please confirm.]
Why this matters beyond your own content
I’m showing you a content project because that’s mine. But step back and look at what a project really is. It’s a place where knowledge, standards and judgment get written down once and applied consistently every time someone asks for help.
Now think about your own team. The proposal approach your best bid manager carries around in her head. The way your organization talks to clients. The compliance rules everyone half-remembers. The tone your CEO expects in every board paper. What happens when all of that sits inside a project the whole team can use? And what happens when it doesn’t, and fifty people each prompt from a blank page and get fifty different standards of output?
This is where I see the human side of AI adoption play out again and again. The organizations getting real value aren’t simply the ones with the most seats. They’re the ones where people understand the tool well enough to set it up properly, and where hard-won expertise lives in the instructions rather than in one person’s memory.
And notice what the tool doesn’t do. It doesn’t post for me. The final judgment, the tweak, the bit that makes it sound like me, still comes from me. A project doesn’t replace your expertise. It holds onto it, so you’re not re-explaining yourself every time you sit down to work.
Try it this week
Pick one task you do over and over, like a weekly report, client follow-ups or meeting summaries. Create a project for it and upload the documents you’d hand a new team member on their first day. Write the instructions as if you were briefing a sharp new starter who knows nothing about how you work. Then run it a few times, and every time the output misses, fix the instructions rather than the output.
So I’m curious. What’s the one task you’d build your first project around? And if you’ve already got one running, what’s the line in your instructions that made the biggest difference?
