Talking to AI

Why Prompting Matters

Here is a scenario you have probably already experienced: you ask an AI a question, and the response is... fine. It is not wrong, but it is not really what you wanted. Too generic, too long, misses the point, or goes in a completely different direction.

The problem is almost never the AI. The problem is the prompt.

If you walked up to a brilliant consultant and said "Help me with my business," they would have no idea what to do. But if you said "I run a 10-person design agency and I am losing clients to cheaper competitors — help me write a pricing page that communicates our value," now they can actually help.

AI works the same way. The quality of what you get out is directly proportional to the quality of what you put in. Not about being clever or using magic words. It is about being structured and specific.

The 8-Part Prompt Framework

We use an 8-part framework for structuring prompts, based on the work of Ruben Hassid. You do not need all eight parts every time — sometimes a quick question is fine. But when you need a great result on something that matters, these eight parts give you a reliable system.

1. Task

Define what you want and what success looks like.

Start with a clear statement: "I want to [TASK] so that [SUCCESS CRITERIA]."

That is the whole formula. The task tells AI what to do. The success criteria tells it how you will judge the result. Without success criteria, the AI is guessing at what "good" means to you.

Weak: "Help me with onboarding."

Strong: "I want to create a 5-step onboarding checklist for new employees at a marketing agency so that every new hire completes their first week with access to all tools, introductions to their team, and a clear understanding of their first project."

2. Context Files

Upload files with your expertise and rules. Stop explaining yourself in the prompt.

This is the single biggest shift in modern prompting. AI went from reading a sticky note to reading an entire book. Claude can handle 200,000 tokens of context — that is roughly 500 pages.

Instead of trying to explain your company, your standards, and your audience inside the prompt, put all of that into files and upload them.

Start your prompt with: "First, read these files completely before responding:"

Then list what you are giving it:

• brand-voice.md — Our tone, vocabulary, and writing rules
• product-overview.md — What we sell and who we sell it to
• audience-research.md — Customer interviews and pain points

Context files are the single biggest lever for improving AI output. A simple prompt with great context files beats a "perfect" prompt with no context almost every time.

3. Reference

Show AI exactly what you want. Upload an example.

Context is background information. Reference is "this is what good looks like." You are showing the AI the quality bar, not just describing it.

Upload an example of what you want, then extract the patterns:

• What makes this reference work?
• What is the tone, structure, and rhythm?
• What rules can you pull from it?

Format each rule starting with "Always" or "Never":

• "Always open with a one-sentence hook before the main content"
• "Never use more than two sentences per paragraph"
• "Always end sections with a concrete action the reader can take"

Do not just say "give me something like this" and hope the AI figures it out. Reverse-engineer the blueprint and hand it over explicitly.

4. Brief

This is the only part you actually type from scratch. Everything else is files.

The brief is your success specification:

• Type of output + length: "A 500-word product announcement email"
• Recipient's reaction: "After reading, the customer should feel excited to try the new feature and understand exactly how to access it"
• Does NOT sound like: "Generic AI, too casual, too formal, jargon-heavy"
• Success means: "The recipient clicks the CTA button. Open rate above 40%"

The brief is short, sharp, and specific. If your context files are doing their job, the brief itself should only be a few lines.

5. Rules

Your context file holds your standards, taste, and audience.

Rules are the non-negotiable constraints. The difference between the brief and rules: the brief is what the client wants, rules are the legal requirements. You might adjust the brief based on feedback, but rules do not bend.

The best practice is to put your rules in a context file rather than the prompt itself. This way they persist across every conversation.

Then in your prompt, add: "Read my rules file fully before starting. If you are about to break one of my rules, stop and tell me."

Example rules:

• "Never use the word 'synergy' or 'leverage' as a verb"
• "Always use metric units"
• "All statistics must include a source"
• "Do not make assumptions about the reader's gender"
• "If you are unsure about a fact, say so rather than guessing"

6. Conversation

You spent years prompting AI. Now it prompts you.

This is the biggest mindset shift in the framework. Instead of firing off a prompt and accepting whatever comes back, you tell the AI: "DO NOT start executing yet. Ask me clarifying questions so we can refine the approach together step by step."

Now the AI interviews you. It asks the questions you forgot to answer. It surfaces assumptions you did not realize you were making. It finds gaps in your brief.

This flips the old model on its head. You are no longer the sole prompt engineer. You and the AI are collaborating to build the best possible prompt together before any work begins.

Key principles:

• Let the AI ask first. Its questions often reveal what your prompt is missing.
• Be specific in your answers. The AI's follow-up questions are only as useful as your responses.
• Build incrementally. Each round of Q&A sharpens the final output.
• Do not start over unnecessarily. If the conversation is 80% there, refine rather than restart.

7. Plan

Claude reads your files before writing a single word.

Before any execution happens, ask the AI to prove it understood your context:

"Before you write anything, list the 3 rules from my context file that matter most for this task. Then give me your execution plan."

This does two things. First, it forces the AI to actually process your context files rather than skimming them. Second, it gives you a checkpoint — if the AI cites the wrong rules or proposes a plan that misses the point, you catch it before it writes 2,000 words in the wrong direction.

A two-minute review of a plan saves twenty minutes of revisions on a full draft.

8. Alignment

Nothing happens until you both see the same aim.

This is the final gate. After the conversation and the plan, confirm alignment before the AI starts producing the real output:

"Only begin work once we have aligned."

Alignment replaces the old prompting era of "send prompt, hope for the best, revise five times." Instead, you invest the time upfront to make sure both you and the AI understand:

• What the deliverable is
• What success looks like
• What the constraints are
• What approach will be taken

Once aligned, the first draft is dramatically closer to the final version. Often it is the final version.

The Complete Template

Here is the full prompt template you can copy and adapt for any task:

You Do Not Always Need All Eight

For a quick question, you obviously do not need the full framework. Here is a rough guide:

Quick questions (1-2 parts): Task only, maybe some context.

• "Summarize this article in 3 bullet points: [paste article]"

Standard tasks (3-4 parts): Task, context files, brief.

• "I want to write a follow-up email to a client who missed a deadline so that the relationship stays strong and we get a new date confirmed. Here is the project context: [upload file]. Keep it under 150 words, friendly but direct."

Important work (5-8 parts): Use the full framework.

• A product requirements document, a business proposal, a complex piece of code, a hiring process — anything where quality really matters and getting it wrong costs real time or money.

The framework scales with the importance of the task. That is the whole point.

Common Prompting Mistakes

Before we wrap up, here are the mistakes we see most often:

• Role prompting. "Act as a senior marketing expert" does nothing useful with modern models. Define the task and success criteria instead.
• Pasting everything inline. Stop writing 500-word prompts. Put your knowledge in files and upload them.
• No success criteria. "Write a blog post" vs. "Write a blog post so that readers understand our pricing change and feel confident it benefits them." The difference is night and day.
• Skipping the conversation step. Letting the AI ask clarifying questions before executing catches problems you did not know existed.
• Accepting the first response. Always review critically and refine. The first draft is a starting point, not an endpoint.
• Starting over instead of refining. If the AI is heading in the right direction, steer it with follow-ups. Do not start a new conversation.
• Skipping the plan. For complex tasks, asking for a plan first saves enormous amounts of time.