How To Use Claude(Chat Is Just The Smallest Form Of It)
You don't need Claude for ideas. You already have enough ideas.
You typed "how to use claude" into Google, which means you already did the hard part: you picked the work tool on purpose.
I've put 700 hours into this since June (I track my time), and most of what follows is what I'd tell you if you were sitting next to me.
So let's skip the tour of the chat box...
Claude comes in more than one form, and the form you open decides what you get back.
Level 1 is Claude in a web browser or on your phone
You ask questions and get answers.
Level 2 is Claude Desktop
It reaches your own files and services and builds the thing itself (ie the email, the deck, the sorted folder).
Level 3 is Claude Code and Claude Cowork
Now you're cooking. The job runs on a schedule, whether or not you're sitting there.
Most people stay on Level 1 because it's the one a search result drops you into.
How I Actually Use Claude (And You Can Too)
You'd think the fast way to use Claude is a better prompt. It isn't. Watch three things get done at once with no prompt at all, just me talking, and one folder of 55 files getting sorted while I do it.
Take The Skill I Built For You
It's a start-here skill: it gets you dictating instead of typing, asks what you want off your plate, connects that one thing, and gets you a real result in your first session.
To add it in Claude: Settings, then Skills, then upload the file. It works on paid plans, and in Claude Code.
What Claude Is For
Twenty years of Google trained you to type a question and collect an answer. That habit is what's holding your Claude use where it is.
The best way to use Claude is not for ideas. The best way to use Claude is for execution of your ideas.
It's a time compressor and a force multiplier, and neither of those has anything to do with better wording.
Which brings up the prompt libraries. If you're searching for prompts, you're already behind.
It's not about finding better words, it's about the size of what you're asking for.
Here's the same job, asked two ways.
"what are the ten best subject lines for my email marketing?"
"look at my last three months of emails, compare the best and worst open rates, and from those deltas write a new email with a subject line built to raise open rates and a body built to raise click-through, then schedule it to send at 5pm tomorrow"
The first one hands you a list to choose from, and then you go do the work. The second one hands you next week's email, written off your own numbers, sitting in your platform ready to go out.
That second ask needs something the first one doesn't: Claude has to be able to reach your email platform. That's a setup thing, and no amount of better phrasing gets you around it. Paste it into a browser tab today and you'll get a polite explanation of why it can't.
What separates those two asks is which level you're on.
One rule carries across all three: give it your material (ie your last three months of emails, your five best ads, your customer language) instead of adjectives. Your own good work beats any instruction to make it punchier.
There Are Three Levels Of AI Users
Level 1
Typing questions into Claude
You ask, it answers, you copy the answer out and carry it somewhere yourself. Useful, but capped, because you're the one carrying everything out of the chat window to wherever it needs to go.
There's one Level 1 upgrade worth doing today, and it's a Project. Anthropic describes Projects as self-contained workspaces with their own chat histories and knowledge bases: you upload your own documents into the project's knowledge base and set instructions for that project.
What that means in practice is you stop re-explaining your business at the top of every chat, you put it in a Project once.
And when explaining would take longer than showing: I take a lot of screenshots and drop them straight into the chat. That counts as handing over your material too.
Level 2
Voice prompting and handing over full jobs
Claude Desktop runs on macOS and Windows, with Linux in beta, and this is where Claude starts working on your stuff instead of just describing it.
Three things change here.
Connect It To Your Stuff
Your files. On desktop, Claude works directly with folders on your computer. You attach the folders you want it in, and inside those it can read, create and modify files. Reads and writes are limited to the folders you connect, so choose what you attach.
Your services. Anthropic describes connectors as the way to connect Claude directly to the tools and data sources that matter most to your workflows. Inside the permissions you granted, Claude can read data in a connected service, change it, and take actions there on your behalf. Connectors work on Claude web, Claude Desktop, Cowork and mobile. Go look at which of your own tools are in there, because that list is what tells you what Level 2 is worth to you.
Put Your Rules In A Skill
Skills. If you searched for Claude Skills, this is what they are. Anthropic describes Skills as folders that include instructions, scripts, and resources that Claude can load when needed, available in the Claude apps on paid plans, in Claude Code, and through their developer platform.
A Skill is where your rules go so you stop typing them out every time: how your emails are formatted, what your brand never says, the order you want a report in.
How you make one: at the end of a session where you went back and forth (the meal one, say: what you like, what you don't, how you cook), say "turn this into the meal planning skill." A week later you open a new session and say "load the meal planning skill and plan five dinners," and none of that gets re-explained. Same for chores, family planning, finances, any of it. Your knowledge goes in once and keeps going forward.
Talk, Don't Type
And the upgrade that costs you nothing: if you're typing into Claude, you're behind.
Talk to it. I use Spokenly, which is free, or any dictation app you like. A big ask is forty seconds of talking and four minutes of typing, and on a phone it's worse than that.
You say: "read the folder I connected, find the three proposals that closed, tell me what they have in common, then write the one I've got due Friday in that same shape"
That's much easier to input to AI versus trying to think and type in a prompt. It saves you time and brain cells and speeds you up, which is also, I think, part of the whole point of leveraging AI.
Level 3
Having it run without you
Claude Cowork is the one most people should look at first. Anthropic's description: it completes tasks you can steer from anywhere, you give it a goal, and it works across your files and tools. Then the scheduling: close your laptop, it keeps going, schedule a task for any cadence and it runs unattended, in the cloud, with your computer asleep and the desktop app shut.
It works even when you're not at the computer.
Claude Code is the other one. If that's the search that brought you here: it was built for code, and it's very good at code. Anthropic describes it as working directly in your codebase, integrating with GitHub, GitLab and your command line tools, plus Routines you configure once that then run on a schedule, from an API call, or in response to an event.
The reason people who don't write code end up in there anyway is that it reaches whatever your command line reaches, and that means your accounts, not just your documents.
I run my own marketing from there. My ad accounts and my email platform are operated from Claude Code, and I don't open the dashboards to do the work.
The newest piece is Claude Design: I handed it a page I didn't like, asked for three layouts, picked one, and brought the file back to Claude Code to put it live.
Level 2 is a few clicks of setup, and Level 3 is one real piece of it.
Real Jobs To Hand Over
These aren't prompts, they're whole jobs. Read these out loud into whichever Claude you have open.
Say this:
"here are five things I've written that worked. read them first, then write the draft of [the thing] in the same register, and tell me what my own five have in common that I probably can't see"
Say this:
"I'm deciding between [A] and [B]. ask me the questions you'd need answered to have an opinion, one at a time, then give me your opinion and what would change it"
Say this:
"read this contract and give me: what I'm agreeing to, what's unusual compared to a standard one, and the three lines I should push back on"
Say this:
"turn this document into a ten slide deck for [audience], one idea per slide, and put my numbers on the slides where they help the argument"
Say this:
"go through the folder I connected, tell me what's in it, then propose a folder structure based on what the files actually are and rename them so I can find things"
Say this:
"go through my unread mail, group it into needs a reply from me, needs a decision, and can wait, then draft the replies for the first group in my voice"
Say this:
"here's what we ate the last two weeks and what's in the pantry. plan next week's dinners around that and build me the grocery list in order of where things are in the store"
The files one and the email one only finish the job at Level 2 or 3, where Claude can reach your folders and your mail. At Level 1 you'll get the work back as text to carry yourself, which is still worth having.
DELETING Clicks Is The Job
UI (user interface) is your enemy. It's sucking seconds of your life away without you knowing about it.
It's not your calendar or your inbox volume, it's the clicks.
Try this: count the clicks in any ordinary task you're doing on the computer.
Pull last month's numbers out of the platform.
(click) open the platform
(click) pick the date range
(click) change the attribution window
(click) export
(click) open the export
(click) sort it
(click) notice the column you needed isn't there
(click) go back
That's not analysis, that's a thousand clicks standing between you and four sentences somebody could have just told you.
Claude is for deleting clicks.
At Level 3, the report you used to click through arrives as an answer, or never gets requested at all because a scheduled job already read it and only tells you what changed. The inbox you dig through on Monday is sorted before you open it.
If your version of this is marketing specifically, I wrote the long one about that: how to use ChatGPT and Claude for marketing, job by job at all three levels.
Where To Start
By now you can probably tell which level you're on.
On Level 1? Don't jump to automation. Open a Project, put your real material in it, and use dictation for a week. Most of the disappointing output at Level 1 comes from asking for finished work while handing over nothing of your own.
On Level 2? Connect one more thing than you've connected today. Files first, then the one service where you do the most repetitive clicking. Then hand over a whole job instead of a piece of one.
Headed for Level 3? Start with something read-only. An audit of what's live, a ranking of what worked across a year. Get it right while it's only reading, then let it write.
The best way to use Claude is not for ideas. The best way to use Claude is for execution of your ideas.
The real test isn't whether the answer was good, it's whether the job got done while you were off doing something else.
And if you read all this and decide the chat window is plenty for what you're running, that's a real answer. Plenty of good businesses would be fine forever at Level 1.
If You Want Help Getting Your Own Accounts There
The distance between the levels isn't knowledge, it's setup: which accounts Claude can reach, what to hand over first, and what stays yours to do.
That part is easier to talk through than to write up as a tutorial.
So tell me what you're trying to accomplish with AI. I'll tell you what level it needs, what to set up first, and what a session with me looks like.
