40+ ChatGPT & Claude prompts for Data Analysis & Excel

**Role:** Senior analytics engineer who has reviewed 5,000+ SQL queries. You can read a 200-line CTE in 90 seconds and explain it to a non-technical PM.

**Context:** SQL query: [paste]. Schema overview: [tables and key columns the analyst should know]. Audience: [the person you're explaining to — PM, ops, exec].

**Task:** Explain the SQL.

1. One-sentence summary: what question does this query answer?
2. Walk the CTEs / subqueries in execution order. For each: what it does (in business terms), what it filters on, what it returns.
3. Final SELECT: what comes out — column meanings in business terms.
4. Edge cases the query handles (or doesn't): NULL handling, duplicates, ties.
5. Performance notes: indexes assumed, materializations, anything that would slow it down at 10x current data.

**Constraints:**
- No SQL jargon without a translation ("LEFT JOIN" → "keeps every row from the first table even if no match")
- Bullet points, not paragraphs
- Highlight any place the query could give misleading results
- ≤500 words

**Output format:** 5 sections · bullet format · ≤500 words.

Write dashboard requirements for a [audience: executives/ops team/data team] dashboard tracking [business area]. For each metric: (1) Definition (exactly how it's calculated). (2) Data source and refresh frequency. (3) Visualization type and why. (4) Target/benchmark/alert thresholds. (5) Drill-down dimensions. (6) Who owns it. Also define: layout priority (above/below the fold), filter defaults, and the single "north star" metric. Anticipate: 3 ways this dashboard will be misread and how to prevent it.

ROLE: You are a senior analytics engineer who writes production SQL for a [DATABASE_ENGINE] warehouse (e.g., Snowflake, BigQuery, Postgres).

CONTEXT: The stakeholder question is: "[BUSINESS_QUESTION]". The relevant tables and columns are:
[SCHEMA_DDL_OR_TABLE_DESCRIPTIONS]
Grain, primary keys, and known join keys: [GRAIN_AND_KEYS].

TASK (reason step by step before writing SQL):
1. Restate the question as a precise analytical ask, listing the metric(s), dimension(s), filters, and time window implied.
2. Identify which tables and join paths are needed and flag any fan-out or many-to-many risk.
3. Decide the correct aggregation grain to avoid double counting.
4. Write a single SQL query that answers the question, using CTEs for readability and inline comments on any non-obvious logic.
5. State 2-3 assumptions you made and how a different assumption would change the result.

OUTPUT FORMAT:
- Section 1: Interpreted ask (bullets)
- Section 2: Final SQL in a fenced ```sql block
- Section 3: Assumptions & caveats

CONSTRAINTS: Use ANSI-compatible syntax for [DATABASE_ENGINE]; never SELECT *; alias every table; qualify all columns; handle NULLs explicitly in filters and aggregates. Do not invent columns that are not in the provided schema; if a needed column is missing, say so instead of guessing.