Prompting Your Brain

Prompting isn't about magic words. It's about giving Claude enough context that there's only one reasonable way to interpret what you're asking. With a brain behind it, most of that context is already loaded.

You've built the brain. This is how you talk to it so the output actually lands.

The honest truth about prompting

Prompting is overhyped. The idea that there are secret phrases that unlock dramatically better output is mostly marketing.

What actually matters is simpler: give Claude enough context that it can only interpret your request one way. If there's ambiguity in what you asked, Claude fills it with assumptions · usually reasonable, rarely exactly what you wanted.

The skill isn't prompting. It's clarity. And a brain does most of the heavy lifting, because the standing context · who you are, your clients, your voice, your processes · is already there. Your prompt only has to supply what's specific to the task.

The context-first principle

Before writing a single instruction, ask: does Claude know enough to do this well? Context means:

• Who you are and what you're trying to achieve
• What the output is for and who it's for
• What format the output should be in
• What constraints exist · length, tone, things to avoid
• Any background that changes what "good" looks like

Most prompts fail not because the instruction was wrong, but because this context was missing.

Bad prompt:

Write a cold email to this prospect.

Better prompt:

Write a cold email to [name], VP of Sales at [company].
They run a 15-person sales team doing manual outbound.
Our angle: we build outbound automation that replaces the manual work.
Keep it under 5 sentences. One CTA: reply to this email.
Tone: direct, peer-to-peer, no buzzwords.

Same task, completely different output. The difference is context, not structure. With a brain connected, the parts in brackets and your tone often come from your knowledge base automatically · you just point Claude at the page.

Anticipate where it gets confused

The second layer is thinking like Claude. Where are the ambiguities? Where might it reasonably go a different direction than you want? Address those up front:

• If tone matters, specify it. "Professional" means different things to different people. "Short, direct, like a text from a founder" is unambiguous.
• If format matters, describe it. Bullets vs prose vs table vs numbered list.
• If length matters, say so. "Under 100 words." "Two to three paragraphs."
• If there are things to avoid, name them. "No buzzwords like leverage or synergy. No exclamation marks."

Give it the right frame

Claude responds well to being told who it is in a specific context. Not roleplay · loading the right frame of reference.

"You are a senior AI systems consultant reviewing this proposal" produces different output than no frame. It loads the relevant expertise. Useful frames:

• "Review this as a sceptical client who's been burned by vendors before"
• "Write this as a founder talking to a peer, not a salesperson"
• "Analyse this like a CFO looking for ROI"
• "Edit this to match the tone of the example below"

Show it an example

Examples beat instructions. If you want a specific style, tone, or format, paste something that nails it and say "write in this style." This is where the brain pays off again · your vault is full of examples. Point Claude at the email that landed, the post that performed, the proposal the client loved. It pattern-matches from the real thing far more accurately than from a description.

Useful prompt patterns

These aren't tricks. They're just ways of giving Claude what it needs.

Before/after:

Here's what I currently have: [paste]
Here's what I want it to become: [describe]
Make that change, keeping everything else the same.

Constraint:

Do X.
Constraints: [list them]
Don't: [list what to avoid]

Example:

Here's an example of the output I want: [paste example]
Now do the same thing for: [your actual input]

Role:

You are [specific role/perspective].
[Context about the situation]
[What you want]

Iteration · when the output is close but not right, don't say "make it better." Say what's wrong:

Good start. Three changes:
1. [specific change]
2. [specific change]
3. [specific change]
Keep everything else the same.

When the problem is reasoning, not the prompt

Sometimes the prompt is fine and Claude just isn't thinking hard enough. Before spending twenty minutes refining, try:

/effort high

Or add "think carefully" or "ultrathink." If the output jumps, the issue was reasoning depth, not prompt quality.

The one rule that covers everything

Read your prompt back before you send it. If someone intelligent read it with zero context, could they interpret it differently than you intend? If yes, add the context that removes the ambiguity.

That's it. That's prompting. The Brain is what makes "zero context" stop being true.