To use Claude well, open a chat, describe the job in plain words with the context it needs, and say what the finished output should look like. Claude does its best work on writing, research and code when you hand it the material up front (a file, a brief, a rough draft) and treat the first answer as something to steer. Once a prompt works, keep it in a project so the next chat starts where the last one ended.
What Claude is good at
I use Claude for four kinds of work, and the free guide is built around the same four.
Writing that has to sound like a person. Emails, landing copy, newsletters, video scripts. Give Claude two samples of your own writing and it holds the tone across a long piece instead of drifting into generic copy by paragraph three.
Reading and research. Paste a long report, a transcript or a contract and ask what matters. Ask for the quotes that support each point so you can check them against the source.
Code. Explaining an error, writing a short script, reviewing a function, or building a full feature from the terminal with Claude Code.
Thinking out loud. Argue the other side of a decision, stress-test a plan, turn a page of notes into a structure. This is the use I underrated for the longest time.
The pieces of the interface
Five pieces cover almost all of my Claude use. The names below are the ones I use day to day.
| Piece |
What it does |
When I reach for it |
| Chat |
One conversation. Claude remembers what was said inside it. |
Quick jobs, single questions, one-off drafts. |
| Project |
A folder of chats that share the same instructions and files. |
Anything I do more than twice: a client, a product, a newsletter. |
| Custom instructions |
Text Claude reads before every reply, so you stop repeating your preferences. |
Tone, format rules, the things I always want and never want. |
| Artifacts |
When Claude writes a document, a page or a block of code, it opens in a side panel you can edit and download. |
Landing pages, long docs, anything I will iterate on. |
| Files |
Attach documents, spreadsheets or images and ask questions about them. |
Research, contract review, data questions. |
The habit that changed the most for me: build the project first, then chat. The project holds the brand voice, the product facts and the format rules. Every chat inside it starts already briefed.
Five prompt patterns that get a usable first draft
Each pattern below is short enough to type from memory. The 10+ mega-prompts in the guide are the long versions, with the role, the checklist and the output format written out.
1. The brief. Role, context, task, format, in that order.
You are an editor for a B2B newsletter. Context: the readers run small agencies. Task: rewrite the paragraph below so it opens with the result. Format: one paragraph, under 80 words.
2. Ask me first. Let Claude collect what it needs before it writes anything.
Before you draft, ask me up to five questions you need answered to do this well. Then wait for my answers.
3. Match the sample. Paste two pieces you wrote, then the job.
Here are two emails I wrote. Study the sentence length, the openers and the sign-off. Write a third email in the same voice about the topic below.
4. Extract with evidence. For long material you need to trust.
List the five commitments in this contract. Under each one, quote the exact sentence it comes from.
5. Draft, critique, rewrite. Three turns in one chat.
Write the draft. Then read it as a skeptical customer and list what would make them close the tab. Then rewrite it, fixing only those points.
Pattern five runs on every piece of copy that goes out under my name.
The mistakes that waste the first hour
A one-line prompt for a ten-page job. Claude fills the gaps with averages, and averages read like every other AI page on the internet.
Drip-feeding context. Paste the whole brief, the whole document, the whole thread. Claude reads long input well, and you lose more by holding material back than by giving too much.
Accepting the first draft. The first answer is a proposal. Say what is wrong with it in one sentence and ask for the second.
A fresh chat for every task. If you have typed the same background twice, it belongs in a project.
Asking for pictures without checking. Whether Claude makes images has changed over time, so read can Claude generate images before you spend a prompt on it.
Blaming your prompt for an outage. When replies stall or fail, check whether Claude is down before you rewrite anything.
Claude vs ChatGPT for common jobs
Both products change often, so this table records where I start today rather than a verdict. Check the plan pages before you commit money.
| Job |
Where I start |
Why |
| Long-form writing in a set voice |
Claude |
Holds tone across long pieces when given samples. |
| Reading a long document and pulling evidence |
Claude |
Handles long input and quotes the source back. |
| Code inside an existing repo |
Claude Code |
Works in the terminal against real files. |
| Generating images |
ChatGPT |
Read can Claude generate images for the Claude side. |
| Quick lookups with live web results |
Depends on plan |
Feature parity moves month to month. |
If you already pay for one of them, learn that one properly before adding the other. The five prompt patterns above transfer.
What is inside the free Claude Mastery Guide
- Claude Prompt Engineering Guide
- 10+ Claude Mega-Prompts
- Claude Prompt Engineering Mini-Course
- Lifetime access and updates
Leave your email and the PDF arrives in your inbox. The mega-prompts are the long-form versions of the patterns above, written for the jobs I see the most: newsletters, landing copy, research summaries, code review.
Where to go after it: the Claude prompts in the library for ready-made prompts sorted by task, the Claude Fable 5 guide if you want the newest model specifically, and the Claude Skills mini course when you are ready to turn a prompt you keep reusing into a skill Claude loads on its own.
