TIL · Field note

Prompt Thesaurus: 59 Action Verbs for Better LLM Outputs

  • prompt
  • LLM

When you open a prompt with a strong, specific verb, you’re basically flipping the model into the right “mode” before it starts thinking. Below is a breakdown of 59 verbs grouped by Reasoning, Transformation, and Creative Generation, each with a quick-use example.

1. Reasoning & Analysis

Use when you want the LLM to think, deduce, or measure.

  • Analyze: “Analyze this dataset and identify patterns.”
  • Assess: “Assess the risk factors in this business plan.”
  • Audit: “Audit my resume for clarity and relevance.”
  • Break down: “Break down the steps needed to deploy this app.”
  • Categorize: “Categorize these bug reports by severity.”
  • Compare: “Compare the performance of Redis vs. Postgres for caching.”
  • Contrast: “Contrast Agile and Waterfall for a non-technical audience.”
  • Diagnose: “Diagnose why this SQL query is slow.”
  • Evaluate: “Evaluate my portfolio website for accessibility.”
  • Examine: “Examine this API response for errors.”
  • Forecast: “Forecast our monthly revenue for the next 6 months.”
  • Identify: “Identify duplicate entries in this dataset.”
  • Inspect: “Inspect this Dockerfile for potential security risks.”
  • Interpret: “Interpret this log file and explain the issue.”
  • Judge: “Judge which of these three taglines is most compelling.”
  • Predict: “Predict which user actions lead to churn.”
  • Prioritize: “Prioritize these features based on user feedback.”
  • Rank: “Rank these marketing channels from most to least effective.”
  • Verify: “Verify if this JSON conforms to the schema.”

2. Transformation & Structuring

Use when you want the LLM to repackage or reshape information.

  • Adapt: “Adapt this blog post for a LinkedIn audience.”
  • Align: “Align these two datasets by matching user IDs.”
  • Classify: “Classify these customer reviews as positive, negative, or neutral.”
  • Convert: “Convert this HTML to Markdown.”
  • Distill: “Distill this 3-page article into 5 bullet points.”
  • Extract: “Extract email addresses from this text.”
  • Format: “Format this SQL output as a readable table.”
  • Group: “Group these GitHub issues by category.”
  • Highlight: “Highlight the main takeaways from this research paper.”
  • Index: “Index these blog posts by topic.”
  • Ingest: “Ingest this dataset into a vector database schema.”
  • Map: “Map these old API endpoints to the new ones.”
  • Normalize: “Normalize this CSV so dates follow ISO 8601.”
  • Organize: “Organize these notes into thematic sections.”
  • Outline: “Outline the structure of a technical proposal.”
  • Reformat: “Reformat this JSON with proper indentation.”
  • Refactor: “Refactor this React component for readability.”
  • Simplify: “Simplify this code without changing functionality.”
  • Sort: “Sort these job titles alphabetically.”
  • Structure: “Structure this raw transcript into sections and subsections.”

3. Creative Generation & Enhancement

Use when you want the LLM to create, improve, or elaborate.

  • Brainstorm: “Brainstorm 10 SaaS ideas for remote teams.”
  • Compose: “Compose a short welcome email for new users.”
  • Create: “Create a landing page headline for a fitness app.”
  • Design: “Design a 5-question onboarding survey.”
  • Elaborate: “Elaborate on this point for a more persuasive argument.”
  • Expand: “Expand this 50-word product description to 150 words.”
  • Generate: “Generate a JSON schema for a blog post object.”
  • Illustrate: “Illustrate how DNS works with an analogy.”
  • Imagine: “Imagine a future where AI manages city traffic.”
  • Improve: “Improve this sales pitch for a skeptical audience.”
  • Innovate: “Innovate on the classic to-do list app concept.”
  • Optimize: “Optimize this SQL query for speed.”
  • Recommend: “Recommend 3 low-cost hosting providers.”
  • Refine: “Refine this mission statement for clarity.”
  • Rewrite: “Rewrite this paragraph in active voice.”
  • Script: “Script a 30-second product video.”
  • Storyboard: “Storyboard a short video ad for a mobile game.”
  • Suggest: “Suggest improvements to our GitHub README.”
  • Visualize: “Visualize this sales data as a bar chart.”
  • Weave: “Weave these three user stories into a single case study.”

Key takeaway:

The first word of your prompt isn’t just decoration. It’s the mode selector. If you want sharper, more predictable outputs, lead with a strong action verb that matches your intent.