Claude.ai — Your Starting Point

Let's open claude.ai and actually do something.

If you haven't already, create a free account at claude.ai. Email, Google, or Apple sign-in — any of them work. Takes about a minute.

You'll see a text box. A blinking cursor. This is where most people freeze. "What do I even ask?" Let's skip the existential moment and try something practical.

Think of something you're actually working on right now. Not a test question, not "tell me a joke" — something real. Maybe you're writing an email you've been putting off. Maybe you're trying to understand a concept for work. Maybe you need to summarise a long document.

Here's an example of a weak first message vs a strong one:

Weak: "Help me write an email"

Strong: "I need to write an email to my manager requesting work from home on Fridays. The tone should be professional but not stiff — my team is fairly casual. I want to keep it short, maybe 4-5 lines. The main reason is a long commute, but I also want to mention that my productivity is higher at home. My manager is generally supportive but likes people to be direct."

Same task. But the second message gives Claude everything it needs to actually help. One message, one good response. No back-and-forth, no "can you make it more casual?" three messages later.

This is the mental model from Module 1 in action. Brief your colleague. Be specific about what "good" looks like.

Try it now. Open Claude, think of something real, and write your first message with context. See what comes back.

Every AI guide says "be specific." It's technically correct but practically useless, because nobody tells you what to be specific about.

Here are the four things that actually move the needle. You don't need all four every time — even one or two makes a noticeable difference.

1. Your situation

One line of context about who you are and what you're doing.

"I'm a freelance designer preparing a proposal for a client in the healthcare industry."

Claude immediately adjusts its tone, vocabulary, and assumptions. Without this, it defaults to generic.

2. The real goal

What you actually need, not just the surface-level task. There's usually a goal behind the goal.

"I need to summarise this article" → that's the task.

"I need to summarise this article so I can explain it to my non-technical team in our Monday standup" → that's the real goal.

The second version gets you a summary that's already shaped for your audience.

3. What good looks like

Tone, length, format, style. Anything that helps Claude understand what you'd be happy with.

"Keep it under 200 words, use bullet points, keep the language simple — no jargon."

4. What you've already tried

If relevant. Claude is better at building on your work than starting from scratch.

"Here's my rough draft. The opening feels flat and the conclusion trails off. The middle section is fine."

Claude now knows exactly where to focus.

Once you've had a few conversations, you'll notice a problem: everything is a disconnected chat. Your job application help is mixed in with your recipe ideas and your research notes. There's no structure.

Projects fix this. They're available on Pro and let you:

• Group related conversations in one place. All your "marketing plan" conversations live together.
• Add persistent instructions. "I'm working on a marketing plan for a SaaS startup targeting Indian SMBs. Budget is ₹5 lakhs. Timeline is Q2." You write this once. Every conversation in the project starts with this context already loaded.
• Upload reference documents. Drop in your brand guidelines, competitor research, or any document Claude should know about. It'll draw from these in every conversation.

Think of a Project as giving Claude a desk to sit at. Instead of briefing your colleague from scratch every morning, they walk in and their desk already has the brief, the documents, and the context from yesterday.

When to create a project

Don't create one for every little thing. A quick question doesn't need a project. But if you're going to have more than 2-3 conversations about the same topic, a project saves you from repeating yourself — which saves messages, which saves tokens.

Good candidates: a work deliverable you're iterating on, a course you're studying, a writing project, job applications, a research topic.

How to create one

Click "Projects" in the sidebar → "New Project" → give it a name → add instructions and documents if you have them. That's it. Future conversations can be started inside this project.

Sometimes Claude doesn't just reply with text — it creates something more structured. A document, a piece of code, a diagram, a spreadsheet-like table. These appear in a separate panel on the right side of the screen. Anthropic calls them "artifacts."

You don't need to do anything special to trigger them. When your request calls for a substantial piece of content — a full email draft, a detailed plan, a formatted document — Claude will naturally create an artifact. You can then:

• Copy it directly
• Edit it and ask Claude to refine specific parts
• Download it when relevant
• Iterate on it across multiple messages ("Make the intro more concise," "Add a section about pricing")

The key thing to know: artifacts are your working canvas. They make Claude feel less like a chatbot and more like a collaborator with a shared document in front of you.

This is one of the most underused features, and it's one of the simplest.

Custom instructions let you tell Claude things about yourself that apply to every conversation. Instead of saying "I'm a marketing professional, keep things concise, use British English" at the start of every chat, you set it once and it's always there.

Go to Settings → Custom Instructions. You'll see a text box. Here's a starter template you can adapt:

About me: [Your role/situation in one line]
Communication style: [How you like responses — concise? detailed? casual? formal?]
Preferences: [British/American English, formatting preferences, anything consistent]
What I don't want: [Common AI behaviours that annoy you — e.g., "Don't add disclaimers at the end of every response"]

Example:

About me: I'm a content strategist at a mid-size Indian fintech startup.
Communication style: Concise and direct. Bullet points over paragraphs when possible.
Preferences: British English. Use ₹ for currency, not $.
What I don't want: Don't start responses with "Great question!" or add unnecessary caveats.

That takes two minutes to set up and improves every single conversation going forward. And it saves tokens — Claude doesn't need you to repeat these preferences.

Not fifty tricks. Not a hundred hacks. Five patterns. Each one is a structure you can apply to almost any task. Learn these and you'll handle most of what you need from Claude without ever reading another "prompting guide."

Pattern 1: The Brief

For: getting work done — drafts, plans, analysis.

"Here's what I'm working on: [context]. I need you to [specific task]. The output should be [format/tone/length]. Here's what I have so far: [existing work, if any]."

This is the workhorse. Module 1's colleague metaphor in action.

Pattern 2: The Reviewer

For: improving something you've already created.

"Here's my [draft/plan/email/code]. Review it for [specific criteria — clarity, tone, logical gaps, grammar]. Be direct about what's weak. Suggest specific fixes, don't just point out problems."

The key instruction is "suggest specific fixes." Without it, you get vague feedback like "the introduction could be stronger." With it, you get "Replace the first sentence with X because Y."

Pattern 3: The Explainer

For: understanding something complex.

"Explain [topic] to me. I have [this level of background]. Use analogies from [domain I know well]. I specifically don't understand [the confusing part]."

Telling Claude what you do know is as useful as telling it what you don't. It prevents Claude from starting at a level that's too basic or too advanced.

Pattern 4: The Brainstormer

For: generating options when you're stuck.

"I'm trying to [goal]. Here are the constraints: [list them]. Give me [number] different approaches. For each one, tell me the main trade-off — what's the upside and what's the risk."

The "trade-off" instruction is crucial. Without it, Claude gives you a list of ideas that all sound equally good. With it, you can actually choose.

Pattern 5: The Transformer

For: converting content from one format/style to another.

"Here's [content]. Convert it to [new format/audience/style]. Keep [what to preserve] but change [what to adapt]. The audience is [who will read it]."

Works for: making a formal report casual, turning meeting notes into action items, adapting a presentation for a different audience, translating technical content for non-technical readers.

Going deeper (if you want to)

These five patterns will cover the vast majority of what you'll use Claude for day-to-day. We deliberately kept this practical rather than technical — you shouldn't need to study "prompt engineering" to get good results.

But Anthropic (the company that makes Claude) publishes their own prompt engineering guide. It goes into things like structuring prompts with XML tags, giving Claude examples of your desired output, and asking it to think step by step through complex problems. It's thorough, it's free, and it's written for a more technical audience.

You don't need any of that to use the five patterns above well. But if you find yourself wanting more control over how Claude responds — or you're curious about what's happening under the hood — that's the place to go. Straight from the source.

This happens. And it's confusing, because sometimes Claude says "I can't" when it seems like it should be able to help. Let's clarify what's actually going on.

Safety boundaries. Claude will decline requests it considers harmful — generating malicious code, creating deceptive content, detailed instructions for dangerous activities. These boundaries exist for good reasons and aren't something to work around.

Knowledge limits. "I don't have information about that." Fair enough. Claude's training data has a cutoff date. For very recent events or very niche topics, it genuinely might not know.

Misinterpreted request. This is the most common one and the most fixable. Claude sometimes reads your request as asking for something you didn't intend. If Claude declines something that seems reasonable, try rephrasing. Add context about what you're actually trying to do and why. Often, the refusal was about a misunderstanding, not a hard limit.

Overlong input. If you paste an extremely long document, Claude might struggle to process it. Try breaking it into sections and working through them one at a time.

The general principle: if Claude says no to something reasonable, add more context about your intent. Nine times out of ten, that resolves it.

We covered the basics in Module 1. Here are a few more practical habits, now that you're actively using Claude.

Front-load context, don't drip-feed it. Five short messages cost more tokens than one well-structured message. And they get worse results.

Use Projects for recurring work. The context you set in a project loads once and applies everywhere. Without it, you're spending tokens re-explaining yourself.

Set custom instructions. Same principle — context that loads automatically instead of you typing it each time.

Say "stop" when you have enough. If Claude is generating a long response and you already have what you need, you can stop generation. Tokens are consumed as they're generated — stopping early saves the remainder.

Keep conversations going instead of starting new ones. Claude maintains context within a conversation. Starting fresh means re-establishing everything from zero.

None of these require you to think about tokens while you're working. Set them up once, and they quietly keep your usage efficient in the background.

You can stop here. With Modules 1 and 2, you have a solid foundation for using Claude.ai in your daily work. The five patterns, custom instructions, and Projects are enough to make Claude genuinely useful — not in a theoretical "AI can do amazing things" way, but in a practical "this actually saved me an hour today" way.

Module 3 introduces Claude Code — the tool that lets Claude work directly with your files and projects. It's more powerful, but it's also a bigger step. Take it when you're ready.