This is the full transcript of How to ACTUALLY Use Claude Projects (Do This), published on YouTube by Parker Prompts. Every paragraph carries the moment it was spoken, so you can click any line to jump straight to that point in the video, search the whole thing for a word, or copy it out.
0:00If you've made a Claude project and just dumped your files into it, you're getting maybe a tenth of what it can do. Set it up right though, and a project does a whole job for you, in your voice and from your own knowledge, every single time you open it. So, today we're building three projects that'll automate 90% of the work you do manually, and I'll give you the exact prompts so you can build your own. A project is a workspace in Claude, over on the left of claude.ai, that holds the files you give it and the instructions you set, and keeps every chat about that job in one
0:26place. Any project really comes down to three things, which I like to call KIP. Knowledge, [music] which is just the files you give it, instructions, which tell it how to behave, and prompts, which is how you talk to it once it's running. But, instead of talking you through all that in theory, I'll just build the first one, and you'll see all three click into place as we go. So, project one is the ghost writer, a project that writes like you. So, what it gives back is pretty much ready to post, and every build today starts from the same file, [music] one document that lays out who you are, what you're
0:54working on, and how you sound, and it's what stops Claude guessing and handing you the same generic answer it would give anyone. Now, you don't write this one yourself, you get Claude to interview you and write it for you. So, I open a normal chat, paste in the interview prompt below, and it asks me one thing at a time, then turns the
1:10whole conversation into a clean profile. One quick tip that makes the profile way better is don't type your answers, say them out loud with a voice-to-text tool like WhisperFlow or Hacks. When you type, you tie yourself up and sound a bit stiff, [music] but when you talk, you fall into your natural voice, and that's what the profile picks up. So, everything the ghost writer writes later sounds like you. About 10 minutes of talking and the profile's done, so I save it. And now I build the project itself. So, I make a new project, call it ghost writer, and drop that profile into the knowledge, plus two or three of
1:40the best things I've published. Those two do different jobs, the profile teaches it how I sound, and my best pieces teach it how I build one, my structure and the way I land an ending. Then I set the instructions, which is just a short brief where I tell it its role, who it's writing for, what I'll use it for, and a do not list of the stuff that drives me mad, [music] like m dashes or invented numbers. I also pointed at the profile I just added, so it works off that every time instead of writing blind. [music] And now I test it. I ask for a short post on something
2:05I'd put out this week, and as you can see, what comes back leans on my examples and reads like I wrote it. And when I don't know where to start, I just give it three rough bullet points, and it turns them into a full first draft in my voice. So I'm tidying something up instead of writing from scratch. And when I'm not sure about a piece, I paste it in and ask where it's weak. And because it knows my standards, it tells me straight instead of just hyping me up. Once a piece is done, it repurposes [music] it, too. The same idea turned into a thread or a newsletter intro,
2:31still sounding like me. So one draft becomes a week of content. Best part is it gets sharper the more you use it. So when a line lands wrong, you don't fix that one reply, you add the rule to the instructions, and it never slips there again. [music] So that's the ghost writer, about half an hour of setup for something that writes like you from now on. And it runs off a profile we're about to reuse twice more. By the way, I've included all three of these projects, plus every prompt I use in this video and one free pack linked below. Project two is the cabinet, a panel of the sharpest people
2:59you trust on call for any decision you'd normally sit on alone. And honestly, it's the one that surprised me the most. You pick two or three people whose thinking you [music] rate, and the move that makes or breaks it is picking ones with plenty of public material out there. For mine, I went with Elon Musk, Andre Karpathy, and Dario Amodei, and all three have hours of interviews out there for Claude to learn from. Now, with the ghost writer, the knowledge was all about you, but with the cabinet, it's other people. So you build this one a bit differently. And you don't just dump their raw transcripts in, that's
3:27[music] the mistake. You run each person through the ingest prompt below first. And what that does is take hours of their material and boil it down to a tight profile of how they think and how they argue. That short profile is what the cabinet actually uses, so it's working from how each person thinks, not
3:42from a thousand pages of raw transcript. So I do that for each person, drop my own profile in alongside them, and now the panel weighs my situation, not some textbook case. Then I'll write the instructions, and the cabinet prompt below tells it to answer as that panel. Each one giving their own take in their own voice, then where they'd agree and where they'd fight. So now I bring it a call I'm stuck on, whether to go all in on one thing or keep a few bets running at once. Instead of one flat answer, each advisor steps in. One says protect the cash, another says it's a
4:10distraction from the bigger money, and because the panel has my numbers, they're arguing about my business. The bit that gets me is when they disagree, because that trade-off laid out by two people who take opposite sides is the thing you never get from a single answer. You can also run it as one person, a single mentor on call. Load only their material and put anything to them, which is honestly the cheapest coaching you'll ever get. And it's not only for making the call, I'll hand the cabinet a plan I've pretty much settled on and ask them to tear it apart, so they catch the weak spot while it's
4:38still cheap to fix instead of after I've committed to it. And it only grows from there. Every time I find someone worth listening to, I add their material and the panel gets deeper. So that's the cabinet, the people you'd never get a meeting with, weighing in on every call you used to make on your own. [music] Project three is the autopilot, and this is the one that does the work for you, because it runs on your own computer while you're away. This one runs in the Claude desktop app in Coda, where the project is a folder on your computer, and there's no fresh setup here because
5:03you can import a project you've already built straight in. So I bring the ghost writer across with its knowledge and instructions intact. [music] Now what makes this one different is connectors, which let it reach into your other apps. So I plug in my Google Drive and my Gmail, and now the project works off my actual files and mail on its own. And if you skip that, you're back to copy-pasting everything in by hand. Then I hand it a job I do every week, turning my latest video into a week of posts. I tell it to take the transcript from this folder, pull out the best ideas, write
5:31me seven posts in my voice, and save them back ready to go, and the exact [music] prompt is below. And this time I can watch the whole thing happen. It reads the transcript, drafts the first post, checks it against my do not list, tidies what it doesn't like, and moves on to the next. A couple of minutes later, there are seven finished posts sitting as files on my computer, each one sounding like me and ready to schedule. That is normally a good 2 hours of my week, and it did all of it
5:54in the time it took me to grab a coffee. Then comes the step that makes it a hire instead of a tool, the schedule. I set it to run every Monday morning, and as long as my computer's awake and the app's open, the posts are written and waiting before I've sat down. So, I'm not starting this job anymore, it just gets done, the way a personal assistant would have it ready before I'm even [music] in. And once one is running, you can point the same setup at your inbox calendar. So, it works through everything overnight and leaves you a
6:17short brief of what needs you that day. So, that's the autopilot, about as close as a tool gets to hiring someone, working your apps, and handing you finished files without being asked. And with all three built, the last thing worth knowing is what makes a project go wrong, so yours doesn't. The big one is the same trap from the start, you load in files, but never set any instructions, so the project has nothing telling it what to do. The rest are smaller, old files left sitting next to new one, so it picks the wrong one, one project trying to do five unrelated jobs at once, and instructions you set on day
6:45one and never looked at again. The fix for all of it is the same, keep one project per job, clear out the old files, and when it keeps getting something wrong, fix the instructions instead of correcting it in the chat, and keep anything private out of the knowledge, or take it out before it goes in. The other thing worth getting right is what belongs in the knowledge and what belongs in the instructions, because I had that muddled for ages, and it's what makes a project feel dumber than the tool it's built on. Knowledge is anything that stays true whether or not you've asked a question, so your
7:11profile, your numbers, your best work, the transcripts, or the documents you'd hand a new hire on their first day. [music] Instructions are how you want it to behave every time it opens, so it's role, who it's talking to, the format you want back, and the things you never want to see. The test I use is whether the sentence would still make sense if nobody had asked anything yet. [music] If it would, it's a fact, so it goes in the knowledge, and if it only makes sense as an order, it goes in the instructions. And once that's straight, picking your own three is easier than it sounds, because the jobs worth a project
7:38all look the same. It's any job you do more than once a month where the output depends on things Claude [music] has no way of guessing and where you'd be re-explaining yourself from scratch in a new chat every time. Writing in your voice qualifies, weighing a decision against your own numbers qualifies, and answering the same questions off the same set of documents qualifies. Booking a flight doesn't because there's nothing
7:57about you it needs to know first. [music] So, take the three jobs you re-explain the most and those are your first three projects. Build one of them properly before you build all three though because one project you actually trust will do more for you than five [music] half-finished ones sitting in the sidebar. And all of it is still just KIP, the same three things from the start, [music] doing the whole job for you once you've set them properly. And that setup, your knowledge plus the right instructions, is what turns Claude from a chatbot into
8:21something that does the work for you. Getting good at this across every AI tool is the whole point of my school community, AI Fluency. So, when you're ready to build the rest, come join us with the link in the description. Thanks for watching and I'll see you in the next one.
No line in this video contains that word.
Where these words come from. This is the caption track YouTube holds for this video, written automatically by YouTube rather than by the creator. We read it, tidied the line breaks and laid it out so it can be read. The plain text version is at https://viewrankai.com/tools/youtube-transcript/lAUvDCQEFMk.txt.
All rights in this video belong to Parker Prompts. Watch it on YouTube. If this is your video and you would rather this page did not exist, tell us and we will remove it.