Choose Strong Action Verbs — Make Prompts Do Things (Not Whisper Nice Wishes)

"The verb is the engine. If you pick 'do something', expect something vague. If you pick 'extract bullet list', expect bullets." — Your future prompt-writing overlord

---

Opening — Why verbs deserve their own therapy session

You already know the big-picture rules from Core Principles: set success criteria up front, decompose before execution, and iterate like you're sculpting a statue out of clay (and mild panic). Now zoom in: verbs are the control knobs that convert those principles into behavior. They tell a model what kind of mental gymnastic you want — summarize, classify, transform, invent, or debug.

Ask yourself: do you want the model to think, chop, rewrite, or judge? The verb you choose answers that. Pick a limp verb and you'll get limp results. Pick a sharp, well-defined verb and the model will sprint in the right direction.

---

The anatomy of a strong action verb

• Precise: It maps to a predictable operation (e.g., 'extract' vs 'talk about').
• Actionable: The model can accomplish it directly (e.g., 'generate 5 ideas' not 'be creative').
• Testable: You can check success automatically (e.g., 'list 10 facts' — count them).
• Composable: It fits into decomposition and iteration: you can chain 'extract' -> 'summarize' -> 'optimize'.

Quick checklist before you write a prompt

1. Do I have a clear success criterion? (Tie verb to it.)
2. Is the verb precise enough to be testable? (If not, refine.)
3. Can this verb be decomposed into smaller verbs that the model can execute stepwise? (If yes, do it.)
4. Is there a measurable output format? (Bulleted list, JSON, table, etc.)

---

Common verbs, what they do, and how to use them (table)

Verb	What it means the model should do	Example prompt artifact
Extract	Pull items or data from input	'Extract all dates as YYYY-MM-DD'
Summarize	Condense content, keep essence	'Summarize in 3 bullets'
Generate	Produce new content based on constraints	'Generate 5 subject lines, 6–9 words each'
Compare	Highlight similarities/differences	'Compare A and B; list 3 differences'
Classify	Assign categories or labels	'Classify each paragraph as fact/opinion'
Transform	Change format or viewpoint	'Transform into a formal email'
Critique/Evaluate	Judge against criteria	'Evaluate argument quality; score 1–5'
Plan/Outline	Create stepwise approach	'Outline a 6-week study plan'
Debug/Explain	Find errors or explain reasoning	'Debug this SQL query and explain fixes'
Synthesize	Combine multiple inputs into unified output	'Synthesize findings into one executive summary'
Optimize	Improve for a metric	'Optimize for clarity and 80% shorter'

Tip: Want the model to do multiple things? Decompose. Use one verb per step or chain them with explicit step numbers.

---

Examples — weak verbs vs strong verbs (and why it matters)

Weak: 'Tell me about the paper.'

Strong: 'Extract the paper's 5 key claims and present each as a single-sentence bullet with the page number.'

Why the strong one wins:

• It's actionable (extract) not vague (tell).
• It ties to success criteria (5 claims, one sentence each).
• It provides output format (bullets, include page number).

Another pair — creative task:

Weak: 'Make a headline.'

Strong: 'Generate 6 headline options for a tech audience, 6–9 words, one uses a question, one uses a pun; label tone as "urgent", "neutral", or "playful".'

Why better: constraints + multiple examples = lower variance.

---

Prompt templates and micro-patterns (copy-paste ready)

1. Extraction pattern

Input: {text}
Task: Extract {item_type} and present as {format}. Limit: {n} items.
Example: Extract all action items and present as a JSON array with fields {"task","owner","due"}.

2. Transform + optimize

Input: {draft}
Task: Transform into {style} and optimize for {metric}. Keep length <= {words}.
Example: Transform the draft into a professional email and optimize for clarity. Max 120 words.

3. Decompose and chain

Step 1: Extract key facts.
Step 2: Summarize extracted facts into 3 bullets.
Step 3: Generate a single-sentence executive summary.

---

Iteration and decomposition — verbs in a workflow

Remember Iteration & Refinement: start with a strong verb for the first pass, inspect results, then iterate with more targeted verbs. Example workflow for processing interview transcripts:

1. Extract — 'Extract all unique names and timestamps.'
2. Classify — 'Classify each quote by topic (product, customer, bug).'
3. Summarize — 'Summarize each topic into 2 actionable insights.'
4. Prioritize — 'Rank insights by impact and feasibility.'

This respects decomposition before execution: each verb maps to a unit of work the model can reliably perform.

---

Pitfalls and how to avoid them

• Using fuzzy verbs like 'analyze' or 'discuss' without constraints. Fix: specify what 'analyze' should output (metrics, bullets, score).
• Overloading one prompt with conflicting verbs. Fix: split into steps or clarify which verb is dominant.
• Forgetting output format. Fix: always pair the verb with the required format (JSON, bullets, table).

---

Closing — quick hits and an exercise

Bold moves you can use immediately:

• Replace 'explain' with 'summarize in N bullets' or 'list steps with reasons'.
• Replace 'improve' with 'optimize for X and output changed sections only'.
• Replace 'create' with 'generate N variants; each variant uses constraint set C'.

Exercise (5 minutes): take a sloppy prompt you or someone posted in chat and rewrite it using three strong verbs in a decomposed flow. Example conversion:

Sloppy: 'Help me with my marketing plan.'

Decomposed:

1. 'Generate 5 audience segments (1 sentence each).'
2. 'For each segment, list 3 high-impact channels.'
3. 'For the top channel per segment, draft a 30-day campaign outline.'

Final flourish: strong verbs are your prompt's steering wheel. They don't guarantee perfect output (iteration does), but they turn aimless wandering into purposeful action. Choose them like a surgeon chooses a scalpel — not a hammer.

---

Key takeaways

• Pick verbs that map to clear, testable operations.
• Pair verbs with output formats and success criteria.
• Use verbs to decompose tasks; iterate on each step.

Version yourself: next time you write a prompt, circle the verb. If you can't explain exactly what the model should produce in one sentence, swap the verb for a stronger one.