TL;DR - What works for drafting emails fails for analyzing data. Each domain has its own prompt pattern; recognizing the domain and reaching for its pattern is a fluency milestone.
Why it matters
A single "good prompt" habit doesn't transfer cleanly across writing, data, coding, and decisions. Knowing the right pattern per domain lets you get a usable result on the first try, even outside your usual role.
Patterns by domain
- Writing - persona + tone + format ("warm, 100 words, end with an ask").
- Research - ask for sources, opposing views, and a confidence note.
- Data - describe the columns, ask for specific calculations and one chart.
- Coding - state the language, inputs, expected output; ask for comments.
- Decisions - ask for options with trade-offs, not a single answer.
Worked example - a data prompt
Here is a sales export with columns: Date, Region, Product, Revenue.
Find the top region by month and flag any month that dropped >20%.
Output a short table plus 2 plain-English insights. Don't guess - if data is missing, say so.Notice how different that is from an email prompt.
Steal this - domain switch
Ask yourself: what KIND of task is this?
Writing -> role+tone+format Research -> sources+counterviews
Data -> columns+calcs Coding -> language+inputs+output
Decision-> options+trade-offs
Then reach for that pattern.Common mistakes (and the fix)
- Using the email pattern on a data task. Fix: describe the data and the exact calculation.
- Asking for one answer on a judgment call. Fix: request options with pros/cons.
- No "say so if unsure" guard on research/data. Fix: add it to curb hallucination.
Good to know
The right tool changes by domain too: Claude/ChatGPT for writing and reasoning, ChatGPT data analysis or Gemini in Sheets for data, Cursor/Copilot for code, Perplexity for sourced research. Same prompting principles, tuned to the job - which is exactly what the next three levels build on.