How Claude Is Different From ChatGPT
Claude and ChatGPT are both large language models, but they're built with different design priorities. Understanding this shapes how you should use each one.
| Capability | Claude | ChatGPT |
|---|
| Context window |
200K tokens (~150K words) |
128K tokens (GPT-4o) |
| Document analysis |
Excellent — full contracts, reports, code |
Good |
| Honesty / refusals |
More likely to say "I don't know" |
Sometimes more confident than it should be |
| Writing quality |
More nuanced, better tone control |
Good, slightly more generic |
| Web browsing |
Limited (claude.ai has some search) |
Yes, with GPT-4o |
| Image generation |
No |
Yes, with DALL-E |
| Plugins / GPTs |
No plugin store |
GPT Store |
💡
When to pick Claude: Long documents, nuanced analysis, careful writing, anything where you need an honest answer more than a confident one. If you're uploading a 100-page contract and need a summary, Claude wins. If you need to generate an image or browse the web, use ChatGPT or Gemini.
Claude is also built by Anthropic, a company whose stated goal is "AI safety." In practice that means Claude is more likely to push back on a task if something seems off, and more likely to say "I'm uncertain" rather than making something up. That's not a bug — it's why you should trust it for analysis work.
The Anatomy of a Good Prompt
Here's why most people get bad results: they give Claude a task without giving it enough to work with. A vague prompt gets a vague response. Claude isn't a mind reader.
A good prompt has four parts. You don't always need all four, but knowing them makes you a better prompter.
The 4-part prompt framework
Role
Who should Claude be for this task? "You are an experienced contract lawyer." "You are a plain-English technical writer." This sets the expertise and tone.
Context
What does Claude need to know? Background info, constraints, who the audience is, what tool/format you're working in.
Task
What exactly do you want done? Be specific. "Summarize" is vague. "Summarize in 3 bullet points, each under 20 words" is precise.
Format
How should the output look? Bullet list? Table? Paragraph? JSON? Numbered steps? If you don't specify, Claude guesses — and often guesses wrong.
Bad prompt vs. good prompt
Here's the same request written two ways. Notice the difference in what you'd actually get back.
Make this email better:
Hi John, I wanted to follow up on our meeting last week.
Let me know if you have any questions. Thanks
You are a senior business communications consultant.
Context: This is a follow-up email to a prospect after
a discovery call. John is a VP of Operations at a
mid-size logistics company. The meeting went well —
he seemed interested but asked for a pricing proposal
that I'm sending in a separate attachment.
Task: Rewrite the email below to be warm but
professional, include a clear next step, and make it
feel like I'm confident and organized.
Format: Keep it under 5 sentences. No bullet points.
Email to rewrite:
"Hi John, I wanted to follow up on our meeting last week.
Let me know if you have any questions. Thanks"
ℹ️
You don't need to write an essay every time. Once you internalize the four parts, prompts just come out better. For a quick task: one sentence is fine. For complex analysis: give Claude everything it needs.
Your First Real Prompt
Let's put this into practice right now. Go to claude.ai — free account is fine for this exercise.
Before we get to the exercise, here are a few things worth knowing about how to interact with Claude:
- Claude remembers the whole conversation. You don't need to repeat context you've already given. Just say "now make it more formal" and it knows what "it" is.
- You can ask Claude to try again. "That was okay but the tone was too casual — try again with more authority." Claude doesn't get offended.
- Longer isn't always better. For simple tasks, a single clear sentence works. Add the four parts when the task is complex or the stakes are high.
- Claude will tell you if something is unclear. If your prompt is ambiguous, Claude often asks a clarifying question rather than guessing wrong. That's a feature.
📝 Day 1 Exercise
Rewrite a Work Email Three Ways
Find a real email you sent recently — a follow-up, a request, an update to your team. It should be at least 3-4 sentences. You're going to ask Claude to rewrite it three different ways.
- Open claude.ai and start a new conversation.
- Use the 4-part framework: set a Role, give Context (who's the recipient, what's the situation), give the Task, and specify the Format.
- Ask for Version 1: formal and concise.
- Without starting a new chat, ask for Version 2: warmer, more conversational.
- Ask for Version 3: bullet-pointed action items instead of prose.
- Compare all three. Notice how different they are — and how each one serves a different situation.
💡
Pro tip: After you get a response you like, ask Claude: "What made this version better than my original?" It will explain exactly what it changed and why. This is how you get better at writing by working with Claude — not just having it write for you.
Common Beginner Mistakes
These are the things that trip people up in the first week:
- Starting over instead of iterating. Don't close the conversation and start fresh when you get a bad response. Claude has full context of everything you've said. Just say "not quite — here's what I actually want" and keep going.
- Asking too many things at once. "Summarize this, then analyze it, then write a response, then give me 5 alternatives" — this usually produces mediocre results for everything. Break it into steps.
- Not specifying the audience. "Explain this to a non-technical person" and "explain this to a senior engineer" produce very different responses. Tell Claude who it's writing for.
- Treating refusals as failures. If Claude pushes back on something, it's usually worth paying attention to. It's not always right, but it's often catching something.
Day 1 Summary
- Claude excels at long documents, careful analysis, and nuanced writing. Use ChatGPT when you need images, web search, or plugins.
- Every good prompt has four parts: Role, Context, Task, Format. You don't always need all four, but knowing them makes every prompt better.
- Iterate in the same conversation — Claude remembers everything you've told it.
- Ask Claude to explain its choices. That's how you actually improve your prompting.
Challenge
Pick one prompt you've been getting bad results from — in Claude, ChatGPT, anywhere. Rewrite it using the four-part framework. If the old prompt was one sentence, make the new one a proper paragraph. Post your before/after to whatever community you're part of. You'll be surprised by the improvement.