Getting Started with Claude
Most people's first conversation with Claude goes the same way. They type a short request, get a generic answer back, and conclude that AI is overhyped for their kind of work. The problem usually isn't Claude. It's that a one-line request is treated like a search query, when it works much better as the start of a conversation.
Tip
The single biggest shift when you start using Claude well: stop asking for a finished answer, and start asking for a first draft you'll refine together.
Claude is an AI assistant made by Anthropic, designed for extended, structured work rather than short back-and-forth chat. That distinction matters more than it sounds. A search engine gives you the same answer no matter how you phrase the question. Claude's answer depends heavily on what you tell it, how you respond to its first attempt, and what you've already discussed earlier in the conversation. Before applying it to any specific job, it helps to build a few habits that make every later conversation more productive.
Give Claude the full picture, not just the task
Claude produces noticeably better output when you describe the situation, not only the task. That means the audience, what already exists, any constraints, and what "good" looks like for this particular piece of work.
The difference this makes is easy to see side by side.
Weak prompt: "Write a follow-up email to a customer."
Strong prompt: "Write a follow-up email to a customer who requested a demo two weeks ago and hasn't responded since. Keep it under 100 words, friendly but not pushy, and end with one specific, low-effort next step."
Both prompts describe the same task. Only one gives Claude enough to produce something you could actually send. The weak version forces Claude to guess at tone, length, and what "follow-up" even means in your context, so it defaults to something generic. The strong version removes the guesswork.
“
Write a follow-up email to a customer who requested a demo two weeks ago and hasn't responded since. Keep it under 100 words, friendly but not pushy, and end with one specific, low-effort next step, like replying with a preferred time.
”Treat the first reply as a draft, not a verdict
A lot of people ask Claude for something once, get a mediocre result, and stop there. That's the equivalent of reading a colleague's first draft and deciding they're bad at the job. First drafts are supposed to need revision, from a person or from Claude.
- 1
Ask for the first version
Give Claude the context above (audience, goal, constraints) and let it produce a full attempt, even if you expect to change it.
- 2
React the way you would to a person
Tell it what's wrong in plain language: "too formal," "cut this paragraph," "the second option is closer, expand on that one instead."
- 3
Ask for one more pass, not a restart
Claude keeps the context of everything you've said so far in the conversation. There's no need to re-explain the brief. Just point at what still needs to change.
Most people abandon the conversation after step one. The people who get real value out of Claude almost always go through all three.
Common mistake
Starting a brand new conversation every time the first answer isn't quite right. This throws away everything Claude already understood about what you wanted, and you end up re-explaining the same context from scratch.
Build a small set of reusable instructions
Once you've had a few good conversations, you'll notice the same context keeps coming up: your tone, your standard format, the kind of detail you always have to add. Instead of retyping it every time, keep a short reference you paste in whenever it's relevant.
This doesn't need to be complicated. A useful reusable instruction set is usually just a few sentences that answer the questions Claude would otherwise have to guess at.
A short description of your tone (2-3 sentences, not a full brand book)
One example of writing you're proud of, that sounds like you
Any format you use repeatedly, like a standard email structure or slide layout
Constraints that always apply, like a length limit or words to avoid
The goal isn't to write a perfect prompt once. It's to stop re-explaining yourself every single time.
Inside Claude Tutorial
This is one habit. There are more.
If keeping a reusable instruction set like this feels useful, that's exactly the kind of habit the app teaches through short lessons and hands-on practice, not just reading about it.
A few things that trip people up early
Three habits account for most of the frustration people report when they're new to Claude:
Assuming Claude remembers something from a previous, separate conversation. Context only carries within one conversation. If you start a new chat, Claude doesn't know what you discussed yesterday unless you tell it again or reuse a saved instruction set.
Asking for everything at once. A request that tries to cover five different goals in one message usually produces something shallow on all five. Breaking it into a sequence, draft the outline first, then the content, then the polish, gets better results than one giant prompt.
Not saying what "done" looks like. If you don't specify a length, a tone, or a format, Claude will guess reasonably, but reasonably isn't the same as correctly. A ten-second sentence about what good output looks like saves several rounds of revision.
Here's a prompt that puts everything above into practice: full context, a clear constraint, and a specific definition of what a good result looks like.
“
I'm drafting a short update for our team about a project delay. Audience: the whole team, including non-technical stakeholders. The delay is two weeks, caused by a vendor issue outside our control. Tone: calm and direct, not apologetic or overly detailed. Keep it under 120 words, and end with the new expected date.
”Breaking a big task into a sequence
Not every job is a single message. When a task has several distinct parts, outline, draft, polish, trying to get all three right in one prompt usually produces something shallow at every stage. Splitting it into a short sequence, where each step builds on the last, almost always works better.
Say you need a full one-page proposal for a client. Instead of asking for the whole thing at once, work through it in stages:
- 1
Ask for a structure first
"Before writing anything, give me a proposed outline for a one-page proposal: sections and roughly how much space each should take." Review the outline before any prose gets written.
- 2
Fill in one section at a time
Once the outline is approved, ask Claude to draft the highest-stakes section first, usually the pricing or the recommendation, so you can react to the part that matters most before investing in the rest.
- 3
Polish as a final pass
Only once the content is right, ask for a tone and length pass across the whole document. Polishing before the structure is settled means redoing the polish every time something moves.
Tip
This matters most for anything long or high-stakes. For a quick email, one message is fine. For a document someone else is going to read closely, the sequence saves real revision time.
Continue reading
- Writing Better Marketing Copy with Claude: see these exact habits, context, iteration, and reusable instructions, applied to a real recurring task.