Thread by @iam_elias1

CLAUDE JUST LAUNCHED OPUS 4.7.

It's the most powerful model they've ever released.

But 95% of people are using it exactly like ChatGPT — and getting 10% of what it can actually do.

Here's how to unlock the other 90%. 🧵

1. Stop writing prompts. Start writing briefs.

Most people open Claude and type one sentence.

Opus 4.7 isn't a search bar. It's a senior collaborator.
Give it:
— Who you are
— What you're trying to build
— Who it's for
— What "good" looks like
— What to avoid

The quality of your output is a direct function of the quality of your brief. Every time.

2. Use XML tags to structure your prompt.

Claude was literally trained on XML-style structure. It reads it better than plain text.

Instead of one wall of instructions, wrap your context in tags:

your situation

what you want

what good looks like

rules to follow

Output quality jumps noticeably the moment you switch to this format. It's the single biggest lever nobody uses.

3. Give it examples. Always.

Telling Claude "write like this" doesn't work
Showing Claude "write like this" works every time

Paste 2 to 3 real examples of the tone, format, or style you want. Then tell it to match

One example = guessing
Three examples = pattern recognition

This single technique will outperform any prompt engineering trick you've ever learned ......

4. Ask Opus to think before it answers.

Opus 4.7 has extended thinking built in — but it won't always use it unless you ask.

Start complex prompts with:

"Before answering, think step by step about the problem. Consider edge cases. Then write your response."

You'll watch it reason through the problem in a thinking block before producing the final output.

The difference on hard tasks is night and day.

5. Use it as a writing partner, not a writing machine.

The fastest way to get generic output is to ask for the final version on the first try.

Instead, iterate:

First prompt: "Give me 10 angles on this topic."

Second: "Expand angle #3 into a rough draft."

Third: "Rewrite it in my voice — here are 3 samples of how I write."

Fourth: "Now tighten it. Remove anything that doesn't earn its place."

Four prompts. Completely different output than one prompt asking for the final piece.

6. Give it a role it can actually inhabit.

"You are a helpful assistant" does nothing
"You are a senior copy editor at The New Yorker with 20 years of experience cutting bloat from feature writing" — that changes the output dramatically

Be specific. Give it:
— A title
— A company or publicatio
— Years of experienc
— What they're known for

The more defined the role, the more defined the response .

7. Use Projects for anything you'll work on more than once.

This is the feature most people ignore and it's the biggest quality multiplier Claude offers.

Create a Project. Upload your style guide, brand docs, past work, or reference material.

Every conversation inside that Project inherits all of it automatically.

You stop re-explaining yourself every time. Claude stops producing generic output. Everything compounds.

If you use Claude daily and aren't using Projects, you're doing it on hard mode.

8. Ask Claude to critique its own output before you do.

After it gives you a response, send one follow up:

"Now critique that response as if you were reviewing someone else's work. What's weak? What's missing? Rewrite it stronger."

Opus 4.7 is remarkably good at self-evaluation.

You'll get a second version that's noticeably sharper — and you didn't have to know why the first one was weak.

9. Turn off your instinct to prompt politely.

"Could you maybe help me with…" produces worse output than "Do X. Here's the context. Here's the format."

Claude doesn't need please or thank you. It needs clarity.

Direct, specific, structured instructions will outperform polite vague ones every single time.

Be kind to humans. Be clear with model

10. Match the model to the task.

Opus 4.7 is the most powerful model Anthropic has ever shipped.

But it's not always the right one.
— Quick drafts, summaries, short rewrites → Haiku 4.5 (faster, cheaper)
— Most daily work → Sonnet 4.6 (the workhorse)
— Complex reasoning, research, critical writing → Opus 4.7

Using Opus for a one-line task is like using a Ferrari to pick up groceries.

Use the right model for the right job and your workflow gets faster AND better.

The real unlock:

Opus 4.7 is not a better ChatGPT.

It's a reasoning partner that rewards preparation, structure, and iteration.

Treat it like a senior collaborator and it will out-think you on problems you've been stuck on for weeks.

Treat it like a search bar and you'll wonder what the hype is about.

The model is extraordinary. Most people just never learn to use it properly.

RT this so more people stop wasting the smartest tool they've ever had access to.