Getting Started with Claude

1. Start Fresh Chats Often — Context Is Everything

Every conversation has a "context window" — a limit on how much Claude can remember within a single chat. The longer a conversation goes, the more likely Claude starts to lose track of earlier details or — worse — hallucinate (confidently give you wrong info based on stale context).

The fix: Start a new chat whenever you switch topics or feel like Claude is getting confused. It's free and instant. Think of it like clearing your desk before starting a new task. One focused chat per topic beats one massive thread every time.

2. Screenshots Are Your Best Friend

When Claude gives you setup instructions — installing an app, configuring settings, connecting services — take a screenshot of what you see and paste it right into the chat. Claude can read images and will give you precise next steps based on exactly what's on your screen.

Why this matters: Instead of spending 10 minutes clicking around menus trying to find the right button, just screenshot → paste → ask "what do I click next?" Claude sees your exact UI and tells you exactly what to do. This alone will cut your frustration in half.

3. Use a "Second Opinion" Chat to Audit Your Work

Here's a power move: when Claude helps you build something — a plan, a workflow, a website — open a separate chat, paste the output in, and ask Claude to audit it. Fresh context, fresh eyes. It'll catch things the first chat missed.

4. Tell Claude What NOT to Do

Claude tends to be thorough — sometimes too thorough. If you want a concise answer, say so. If you don't want a history lesson before the advice, say that too. Negative constraints are just as powerful as positive ones.

5. Frame Claude as an LLM, Not a Person

The worst prompting advice floating around is "tell Claude to act as a senior [insert profession]." Don't do it. Claude isn't a person — it's an LLM that has absorbed an enormous amount of information across virtually every domain. When you tell it to "act as a senior tax attorney," you're actually constraining it to roleplay a human caricature instead of leveraging what it actually is.

❌ DO NOT: "Act as a senior CPA with 20 years of experience and advise me on..."

✅ DO INSTEAD: Frame it as an LLM drawing on specific domain knowledge, state your goal, list explicit tasks, and always ask for clarifying questions first.

6. Iterate — Don't Start Over

If Claude gives you something that's 70% right, don't re-prompt from scratch. Say "this is close, but change X and make Y shorter." Claude is built for iteration. Treat it like editing a draft with a colleague, not placing a new order at a restaurant.

7. Upload Files Instead of Explaining Them

Don't spend 10 minutes describing your spreadsheet layout. Just drag the CSV or Excel file into the chat and say "analyze this." Claude reads files directly — it's faster and more accurate than you typing out what's in them. Works with PDFs, images, Word docs, code files, and more.

8. Save Your Best Prompts

When you write a prompt that gets a great result — the morning market brief, the expense categorizer, the betting lines pull — save it somewhere (Notes app, a doc, a Claude Project). You'll reuse your best prompts dozens of times. Build a personal prompt library over time.

9. Use Projects for Recurring Work

Claude Projects let you pin instructions and reference files that apply to every conversation in that project. Set up a "Trading" project with your risk tolerance and watchlist, a "Client Work" project with your firm's memo templates, a "Fantasy" project with your league settings. Claude gets smarter about your context every time.

10. Verify Numbers — Trust but Check

Claude is excellent at analysis, but it can miscalculate or misinterpret data, especially in long conversations. For anything with financial or legal consequences — tax numbers, trade P&L, client-facing figures — spot-check the math. Ask Claude to "show your work" or "break this down step by step" so you can follow the logic.