Guide 📝 Prompt mega-pack vol. 3 — 25 prompts for data work (cleaning, SQL, dashboards)

Vol. 1 (the original 25) was everyday time-savers. Vol. 2 (the work-study-money pack) went deeper. Vol. 3 is the one the data crowd asked for: 25 prompts for cleaning, SQL, EDA, dashboards and stakeholder docs. Copy-paste, swap the [brackets], go.

Two rules before the pack: (1) paste only small samples of real data — schemas and 10 rows, not 10,000; (2) never trust a number the AI invents — it writes the logic, you run it and verify.

🧹 Messy data cleaning

1. Profile my mess

Code:

Here's a 10-row sample and the column list from my CSV: [paste]. List every data quality problem you can spot (types, nulls, dupes, weird formats), then rank them by "will break my analysis first".

2. Dedupe referee

Code:

Column [customer_name] has near-duplicates like "S Vigu", "S.Vigu", "Sir Vigu". Give me 3 dedupe strategies from safest to most aggressive, with the exact Python/pandas line for each.

3. Null-decision tree

Code:

Column [age] is 22% empty, [city] is 3% empty. For each: should I drop, fill, or flag? Ask me the 3 questions you need answered first, then give your call with reasoning.

4. Category normalizer

Code:

Here are the unique values of my [plan_type] column: [paste list]. Map them into 4 clean buckets. Show old → new as a dictionary I can paste into code.

5. Outlier honesty-check

Code:

In this column summary [paste describe()], which outliers look like data errors vs real extreme values? Give me a rule to separate them, and a one-liner to cap the errors only.

🗄️ SQL

6. English → SQL

Code:

Schema: [paste CREATE TABLEs]. Dialect: [MySQL 8]. Write a query that: [plain-English question]. Add comments explaining each JOIN, and flag any assumption you made.

7. Query debugger

Code:

This query [paste] returns [wrong result / error]. Tables look like [paste schema]. Find the bug, explain WHY it produced the wrong rows, then give the fixed query.

8. The optimizer

Code:

This query takes 40 seconds on 2M rows: [paste]. Indexes exist on [list]. Rewrite it for speed, and explain the change in one sentence each. Dialect: [MySQL 8].

9. Legacy query translator

Code:

Explain this monster query like I'm new: [paste]. Give me: 1-line summary, what each CTE does, and what breaks if [column X] is NULL.

10. Dialect mover

Code:

Convert this MySQL query to PostgreSQL: [paste]. List every function you had to swap and why.

(New to SQL? Do the 🗄️ one-weekend SQL plan first — these prompts assume you can run a query.)

🔍 EDA & insight

11. First-look checklist

Code:

I just got a dataset about [topic]. Columns: [paste]. Give me the 12 EDA questions I should answer first, in order, and the one chart for each.

12. Surprise hunter

Code:

Here are summary stats: [paste]. What 3 patterns would be SURPRISING for data about [topic]? For each, give me the one query/plot to check it.

13. Segment finder

Code:

My users table has [columns]. Propose 4 ways to segment users that could change a business decision — not obvious ones like 'by city'. Then rank by easiest to compute.

14. Anomaly storyteller

Code:

Sales dropped 30% on [date]. Relevant columns: [list]. Generate 6 hypotheses ranked by likelihood, and the exact check for each. No code yet — think first.

15. Metric tree builder

Code:

Our north-star metric is [monthly revenue]. Build a metric tree 3 levels deep (drivers and sub-drivers), and for each leaf tell me the formula and the table I'd need.

📊 Dashboards & storytelling

16. Chart chooser (that admits trade-offs)

Code:

I need to show [relationship: e.g. 'revenue by region over 24 months'] to [audience]. Give me 3 chart options, when each wins, and which one you'd reject for this audience.

17. Title rewriter

Code:

My chart title is "Sales by Month". Rewrite it 5 ways so the title says the INSIGHT, not the axes. Data takeaway: [e.g. 'Diwali month doubled our baseline'].

18. Exec summary in 5 lines

Code:

Turn these findings into a 5-line summary for a busy CEO: [paste bullets]. Line 1 = the decision this supports. No jargon, no hedging.

19. So-what tester

Code:

Here's my dashboard layout: [describe]. For each chart, answer "so what?" in one line. If a chart has no answer, tell me to kill it.

20. Slide skeleton

Code:

Turn this analysis [paste] into a 6-slide outline: title, one chart idea, and the single sentence the slide must land.

🤝 Stakeholders & docs

21. Requirements translator

Code:

My manager said: "[paste vague request, e.g. 'can we see how the app is doing?']". Translate it into 4 concrete, measurable questions, and flag what I must clarify before touching data.

22. Assumption logger

Code:

I'm starting analysis on [project]. Generate the 8 assumptions I'm probably making silently (about data freshness, definitions, edge cases) that I should write down and confirm.

23. Data dictionary writer

Code:

Table: [paste schema + 3 sample rows]. Write a data dictionary: column, type, plain-English meaning, allowed values, and 'watch out for'. Format as a markdown table.

24. QA checklist before I publish

Code:

My analysis claims [X]. Give me a 10-point checklist to verify before I present — cover join fan-out, filter leakage, date boundaries, and units.

25. Handoff note

Code:

I'm handing this analysis to a colleague: [describe]. Write the handoff doc: what it does, how to rerun it, the 3 things most likely to break, and where the data lives.

⚠️ The two-minute safety talk

• Schemas and samples are fine to paste; your company's real customer data is not. Anonymise first.
  
• AI writes great logic and terrible arithmetic — run everything, trust nothing un-verified.
  
• Need practice data? 🗃️ Where to find free datasets & clean them has the good stuff.
  

Bookmark this — vol. 4 drops when you lot finish this one. 😄

— Sir-Vigu
Founder, Data-Forums