← AI 101
Intermediate training

Prompting Fundamentals

Move from "asking Claude things" to writing prompts that consistently produce work-ready output. Builds directly on Module 1.1's five rules and goes deeper on each.

20 minutes Builds on Level 1 Includes prompt builder

What you'll be able to do after this lesson

01

Write a first-try prompt

Get a useful first draft on the first try — most of the time. Not perfect every time. Most of the time.

02

Diagnose a weak answer

When a response comes back generic or off-target, identify the missing piece of the brief and fix it in one round — without starting over from scratch.

03

Keep a personal prompt library

Save the prompts that worked. Three to five reusable templates beat a hundred one-off attempts.

A prompt is a brief, not a question

C

Context

Who is this for, what's the situation, what does the reader already know. The "who, what, why" that the answer has to live inside. Without context, Claude guesses — and the guess is generic.

Constraints

Length, tone, format, audience, what to avoid. The boundaries the output has to fit. Constraints are how you turn "a draft" into "the right draft for this audience, this long, in this voice."

Target

What good output looks like. A sample, a structure, a clear definition of done. Without a target, Claude optimizes for sounding helpful — which is not the same as being useful.

Trainer note: A prompt is a brief, not a question. The shift from "asking" to "briefing" is the single biggest jump in prompting skill. Brief Claude the way you'd brief a capable contractor — and expect the same range of output quality based on how good the brief was.

Where prompts live in your day

One-shot prompts

A single message with the complete brief. Best for known patterns and tasks where you already know what you want. Fastest when the brief is in your head.

Solo brief

Conversational prompts

Build the brief across turns. Best for exploratory work — when you're not sure exactly what you want until you see a draft. Trade speed for discovery.

Build by turn

Saved prompts

Templates you reuse with variables. Best for recurring tasks (status updates, follow-up emails, condition explanations). Build once, ship dozens of times.

Build once

The prompting playbook

1

Lead with the goal

The first line should tell Claude what done looks like — not what the topic is. "Draft a 120-word reply that closes this thread" beats "I'm writing about Q4 results."

2

Give an example of good output

One sample of what you want is worth a paragraph of describing what you want. Paste a prior email you liked, a snippet of the right tone, a structure to copy.

3

Name the audience

Claude writes differently for a realtor partner, a borrower, a manager, and a compliance reviewer. Name the reader. Their priors are your output.

4

Specify format and length

Bullets vs. paragraphs, table vs. prose, 100 words vs. 500. The format you don't specify is the format you'll have to rework.

5

Ask Claude to ask you questions

When the brief feels thin, end with "Ask me three clarifying questions before drafting." Claude will surface what you forgot to include — and you'll have a better brief in 30 seconds.

Weak prompt

Write a marketing email for our new product.

Work-ready prompt

Audience: existing GMFS realtor partners on our preferred list. Offer: a new 95% LTV conventional product with no MI for FICO 740+. Tone: warm, professional, no hype. Length: 110 words. CTA: book a 15-minute call via my Calendly. Include a one-line subject. Sample of my recent voice in the next message.

Four prompt patterns that earn their keep

Role · Context · Task · Format

"You are a [role]. Here's the [context]. Do [task]. Output as [format]." The four-slot pattern that handles 80% of work prompts.

Before-and-after rewrite

"Here's the current version. Here's what's wrong with it. Rewrite to fix exactly those issues, without changing anything else." Cleaner than open-ended "make it better."

Structured extraction

"From the text below, extract every [thing] as a table with columns: name, date, amount, status." Turns unstructured input into structured output you can use.

Three options, one recommendation

"Give me three approaches to [decision]. For each: pros, cons, when to pick it. Then recommend one and explain why." Replaces vague "what should I do" with a real decision aid.

Five answer-quality red flags that point back to the brief

Employee rule: When the answer is bad, the prompt was usually the problem. Before blaming the model, re-read your prompt and ask which of the five signals above shows up. Fix the brief, run again, ship.

Six exercises that compound

The point is not to memorize templates. The point is to build a small library of prompts that work for the things you actually do.

  1. Open the prompt builder in this lesson. Toggle the chips on and off — watch how the brief grows. Save the strongest version somewhere you'll find it.
  2. Rewrite five of your recent prompts into the Role · Context · Task · Format pattern. Run each against the original and compare the outputs.
  3. Save three prompts you'll reuse weekly. Put them in a doc, a Slack canvas, a notes app — somewhere you can grab them in five seconds.
  4. Pair up with a teammate and trade your three saved prompts. Run their prompts on your work. Note what transfers and what doesn't.
  5. Deliberately break a prompt. Strip out the audience, the format, the length. Watch how the answer degrades. Now you know what each part was doing.
  6. Ask Claude to critique one of your prompts. Paste your prompt, ask "what would you add to this brief if you had to produce work-ready output?" Use the feedback.

Completion standard

You've finished this module when you have three saved prompts you actually use, can diagnose a weak answer by reading your own prompt, and can write a brief — not a question — on the first try.