MODULE 01

Prompt Writing Fundamentals

Master the five dimensions that separate pass-level outputs from reject-level ones. AI companies evaluate on specificity, context, role clarity, constraint definition, and output format. Learn to hit all five — consistently.

5 Lessons 1 Self-Assessment 3 Practice Exercises
1 2 3 4 5
LESSON 1 / 5

Clarity

Ambiguous prompts produce ambiguous outputs. The evaluator's rubric penalizes vague language — words like "good," "nice," "some," "a few." These don't constrain the model, so it fills the gap with generic filler.

A clear prompt uses concrete terms: numbers instead of quantities, specific names instead of categories, exact behaviors instead of general directions.

WEAK — Unclear
Write a blog post about social media marketing. Make it good and engaging for people who might be interested.
STRONG — Clear
Write a 650-word blog post targeting B2B SaaS founders who are new to content marketing. Cover three specific tactics: short-form video repurposing, newsletter guesting, and LinkedIn comment engagement. Tone: direct, data-grounded, no fluff.
Key move: Replace every adjective with a number, a name, or a concrete behavior.
LESSON 2 / 5

Specificity

Broad prompts generate surface-level outputs. Evaluators are calibrated to catch outputs that demonstrate "obvious coverage" without deep execution. The gap between a pass and a reject is usually 2-3 levels of specificity you didn't include.

Specificity means: naming the exact format, the exact scope, the exact edge cases, the exact audience's knowledge level, and the exact deliverable structure — before the model starts writing.

GENERIC — Surface level
Explain how to save for retirement.
SPECIFIC — Deep execution
Explain retirement savings for a 30-year-old software engineer earning $120k in a US HCOL city. Cover: 401k match (including the concept of vesting), Roth vs traditional IRA decisioning, HSA triple advantage, and why target-date funds beat stock-picking for this profile. Include the specific income thresholds where each strategy changes. End with a concrete monthly contribution target based on a $4,500/month after-tax income.
Key move: Name every dimension the model must address. If you leave one out, it stays shallow.
LESSON 3 / 5

Context

The same instruction lands differently depending on what the model knows about the situation. Without context, models default to generic — safe, shallow, broadly applicable. Evaluators test whether you can give the model enough situation to write as if it already knows what's going on.

Context includes: background information, intended use case, prior decisions already made, constraints the reader won't know, and the stakes of getting it wrong.

NO CONTEXT — Generic output
Write a follow-up email after a sales call.
RICH CONTEXT — Tailored output
Write a follow-up email for a 45-minute discovery call we just had with Mara Chen, Head of Revenue Ops at a 120-person Series B SaaS company (PointSync). Her primary pain: their current BI tool requires SQL for any ad-hoc report, slowing down the weekly executive review prep from 2 hours to 6. Her team size is 4 analysts + 8 GTM reps who need self-serve. Budget owner is the VP Finance. Decision timeline is end of Q3. Three competitors in the shortlist including us. Tone: consultative, not pushy.
Key move: Assume the model has no memory. Write everything it needs to act as if it was in the room with you.
LESSON 4 / 5

Role Definition

Role definitions aren't decorative — they set the model's decision-making framework. The same task done "as a junior copywriter" vs "as a CRO strategist" produces categorically different outputs. Evaluators assign separate rubric scores for whether you correctly invoked the right persona.

A good role definition has three parts: the persona (who it is), the perspective (how it thinks), and the boundary (what it won't do or say).

THIN — Decorative role
You are a helpful assistant. Write a product update email.
RICH — Decision-making role
You are a B2B email copywriter with 8 years of experience writing for HubSpot, Salesforce, and Outreach users. You write with the rhythm of a seasoned account executive — confident, specific about outcomes, never uses "leverage" or "synergy." You think in terms of what the VP of Sales actually cares about before they open their laptop. Write from their perspective, not yours. Do not include emoji. Keep sentences short.
Key move: Go beyond job title. Describe the decision-making criteria the persona uses — that's what makes the role actually stick.
LESSON 5 / 5

Output Formatting

Unstructured outputs get rejected. Evaluators score whether the output matches the format specified in the prompt. The bar isn't "any structure" — it's the structure that actually serves the use case. Bullets where paragraphs are needed is as bad as paragraphs where bullets are needed.

Format specifications include: structure type (paragraph, bullet, table, JSON, script), length constraints, section ordering, header presence, and any mandatory fields that must appear.

NO FORMAT — Evaluator rejected
Give me feedback on this resume.
EXPLICIT FORMAT — Evaluator scored
Review this resume and return a JSON object with exactly these fields: { "overall_score": number 1-10, "strengths": string[], "critical_gaps": string[], "priority_fixes": { "field_name": "what_to_change" }[], "verdict": "pass" | "borderline" | "reject" }. No additional fields. Verdict must be justified by the specific score.
Key move: Specify the exact format, length, and structure — then name the fields or sections that must appear. Leave zero ambiguity.

Prompt Scorer

Paste a prompt you've written and get scored on the five dimensions evaluators care about. This simulates the rubric-based scoring used in actual AI company assessments.

Practice Exercises

Review each AI output and score it using the rubric. Then compare against the evaluator's reasoning. These are drawn from real assessment scenarios.

EXERCISE 1 / 3 Task: Product Description
THE PROMPT (what was given to the AI)
Write a product description for a new wireless ergonomic keyboard. The description should be compelling and help people understand why it's worth buying. Target audience is remote workers. Length around 100 words.
THE OUTPUT (what the AI generated)
Introducing the ProType Wireless Keyboard — designed for the modern professional who demands both comfort and performance. This sleek, ergonomic keyboard features a split-key design that reduces strain during long work sessions, while the quiet mechanical switches provide satisfying tactile feedback without disturbing your colleagues. With multi-device Bluetooth connectivity and a 3-month battery life, you'll never miss a keystroke. Whether you're working from home, a coffee shop, or a co-working space, ProType keeps you productive and comfortable all day long. Upgrade your workspace today.
YOUR SCORING

Score this output on each dimension. Select a score (1-10) for each, then compare against the evaluator's verdict below.

EXERCISE 2 / 3 Task: Customer Support Reply
THE PROMPT
Write a customer support reply for a user whose order was delayed. Apologize and explain when it will arrive.
THE OUTPUT
Hi there, Thank you for reaching out! I'm really sorry about the delay with your order — that's not the experience we want you to have. We know how frustrating it can be to wait for something you need, and we appreciate your patience while we worked to get this sorted out. Good news: your order is now on track and should arrive within the next 3-5 business days. You'll receive a shipping confirmation email as soon as it's on its way. If you have any other questions, don't hesitate to reach out. We're always here to help! Warmly,
The Support Team
YOUR SCORING

Score this output on each dimension.

EXERCISE 3 / 3 Task: Data Analysis Summary
THE PROMPT
Analyze this dataset and write an executive summary for the VP of Finance. Include key findings, trends, and recommendations. The data covers Q1 2024 financial performance.
THE OUTPUT
## Q1 2024 Executive Summary **Overview** Q1 2024 showed mixed results with revenue growth offset by margin compression. **Key Findings** - Revenue increased 18% YoY, driven by enterprise segment expansion - Gross margin declined 4 points to 61%, primarily from product mix shift toward lower-margin services - Operating expenses grew 22%, outpacing revenue growth - Customer acquisition cost (CAC) rose 15% while customer lifetime value (LTV) remained flat **Trends** The data shows a concerning pattern where growth is being purchased through higher spending rather than efficiency gains. Enterprise deals are taking 40% longer to close, extending the sales cycle and increasing cost per deal. **Recommendations** 1. Restructure pricing for professional services to protect margins 2. Implement stricter SQA gates before enterprise proposals are sent 3. Evaluate CAC efficiency by segment — redirect spend from underperforming channels 4. Consider automating routine implementation tasks to reduce services headcount needs
YOUR SCORING

Score this output on each dimension.