Prompt Types

Analysis Prompting

New to prompting?
Learn how to have the LLM help co-author your prompts or check out the common mistakes.

Analysis is one of the most valuable aspects of the LLM. It can score billions of bits of training data and compile them quickly. So why is so much LLM analysis pure junk? This is due to poor prompting. When doing analysis, the context is critical. Do you mean all time or just last month? Do you want everyone over 18? Over 30? Do you want imperial or metric? Are you talking to children or industry professionals?

Context to consider when analysis prompting

  • Model's training date: Ask it "What is your training date?" If you need historical data, the training data is fine. If you need the latest, you need to tell the model to get fresh data as of {date}
  • Type of analysis: financial, descriptive, correlation, summary, hive mind review, comprehensive, expert, trends
  • Data sources:attach a pdf or spreadsheet, point to formal sources online
  • Data scope all females, only European countries, customer support tickets about failing start button
  • Units of measure
  • Key metrics: PPM, session duration, ROI
  • Comparison baseline: last year, industry average
  • Time period:Which year, by quarter, all time
  • Audience: Automotive industry professionals, K-8 schoolteachers, COM101 students
  • Desired insights or hypotheses to test: root cause, projection, seasonality impact
  • Visualization style: heat map, chart, trendline
  • Output format: chart, XLS, bullet point list, 500 word summary
Data analysis for climate policy analyst 
You are a climate policy analyst reviewing wildfire incident trends in western North America from 2000 to 2024. Using satellite heat anomaly data and regional fire department incident reports:

1. Analyze trends in fire frequency, size, and seasonal timing by region (e.g., Pacific Northwest, British Columbia interior, California).
2. Correlate those trends with:
   - Average summer temperatures  
   - Drought index levels  
   - Land development or urban-wildland interface expansion  
3. Identify any statistically significant anomalies during La Niña or El Niño years.
4. Provide:
   - A regional heat map showing fire density over time  
   - A trendline chart for fire season start dates  
   - A short briefing summary suitable for government use
Simulated expert analysis on electrical repair proposal 
Simulate a review panel consisting of experienced electricians, licensed electrical engineers, and safety inspectors. Have them evaluate the attached repair instructions for accuracy, safety, and alignment with best practices. Highlight any red flags and suggest safer alternatives.
LLM self-diagnose bad analysis 
This prompt is not returning the intended output. Analyze it and explain why. Identify structural problems, ambiguous phrasing, or missing context that may be causing incorrect results. Do not rewrite it. First, repeat the instruction in your own words to confirm understanding. Then identify what additional context might be needed to improve performance.
Sales analysis 100 years of data 
You are an expert data analyst with access to U.S. historical sales and inflation data. Using annual candy sales figures from 1925 through 2024:

1. Fetch or ingest the original annual U.S. candy sales for each brand/product over that period.
2. Adjust every year’s sales to 2023 dollars using the CPI-U inflation index (sourced from the BLS).
3. For each brand/product, calculate:
   a. Annual real (inflation-adjusted) sales.
   b. Cumulative real sales over the entire period.
4. Rank all brands/products by their cumulative real sales and identify the Top 25.
5. Produce:
   - A table with columns: Rank, Brand/Product, Original Sales by Year, Adjusted Sales by Year, Cumulative Adjusted Sales.
   - A line chart of year-over-year real sales for the Top 10 brands.
   - A list of the five largest year-to-year percentage growth spurts among the Top 25.
6. Write a concise executive summary highlighting:
   - Legacy brands that dominated the early 20th century.
   - Post-war boomers and any modern upstarts.
   - Surprising dips or spikes tied to historical events (e.g., Great Depression, WWII, recent health trends).

Format your output as:
Step 1: Data sources & inflation adjustment methodology.
Step 2: Calculation & ranking process.
Step 3: Top 25 table.
Step 4: Visualizations.
Step 5: Executive summary.