Data Analysis
Analyse a Dataset With a Question
Analyse pasted data against a business question: findings, caveats, and the chart you should actually make.
The Prompt
analyse-a-dataset-with-a-question.md
You are an analyst. Answer the question with the data given. Do not invent rows. Question: {{question}} Data (CSV, table, or notes): {{data}} Audience: {{audience}} Grain / date range if known: {{grain}} Return: 1. Restate the question in measurable terms 2. How you treated the data (filters, missing values, duplicates) 3. Findings, each with the number from the data 4. Caveats (sample size, confounders, what the data cannot say) 5. One recommended chart and why 6. Next measurement if the question is still open If the data cannot answer {{question}}, say so in the first sentence and state what would.
Negative prompt
Included when you copy the prompt
Do not invent rows, metrics, or segments that are not in {{data}}. Do not treat correlation as causation or state a cause that the data cannot support. Do not bury the answer in methodology fluff or hedge every finding into meaninglessness. Do not recommend a chart that does not match the grain or question. Do not ignore missing values, duplicates, or small sample size. If the data cannot answer {{question}}, do not force an answer - say so first and state what data would.
Fill in your details below - this prompt updates as you type, then copy it.
Fill in your details
0 of 4 filled. Preview updates as you type - then copy the finished prompt (negative prompt included when present).
analyse-a-dataset-with-a-question.md
Prompt preview
You are an analyst. Answer the question with the data given. Do not invent rows. Question: {{question}} Data (CSV, table, or notes): {{data}} Audience: {{audience}} Grain / date range if known: {{grain}} Return: 1. Restate the question in measurable terms 2. How you treated the data (filters, missing values, duplicates) 3. Findings, each with the number from the data 4. Caveats (sample size, confounders, what the data cannot say) 5. One recommended chart and why 6. Next measurement if the question is still open If the data cannot answer {{question}}, say so in the first sentence and state what would.
Examples
Example input
Question: Which channel drove the highest revenue per session last week, and is paid search worth the spend? Data: channel,sessions,revenue,ad_spend organic,4200,18500,0 paid_search,1800,14200,3200 email,950,9800,0 social,2100,4100,900 Audience: growth lead Grain / date range: daily sessions rolled to week of 2026-08-04 to 2026-08-10
What you should get
1. Measurable question: Rank channels by revenue/session for week of Aug 4 - 10; for paid search, compare revenue to ad_spend. 2. Treatment: Used given aggregates as-is; no row-level dedupe possible. Assumed sessions and revenue are non-overlapping by channel. 3. Findings: - Email: $10.32/session (highest); revenue $9,800 on 950 sessions. - Organic: $4.40/session; $18,500 on 4,200 sessions (largest absolute revenue). - Paid search: $7.89/session; $14,200 revenue vs $3,200 spend → ~4.4x return on ad spend before other costs. - Social: $1.95/session; weakest efficiency. 4. Caveats: Channel attribution may double-count assisted conversions. No margin or LTV in the data. One week only - seasonality unknown. Ad spend only provided for paid and social. 5. Chart: Horizontal bar of revenue/session by channel, with a second series or annotation for ad spend on paid/social. 6. Next: Pull multi-touch attribution or last-click vs first-click for paid search over 4 weeks, plus contribution margin.
When to use this prompt
Use it when
- You need help with analysts.
- You need help with pms.
- You need help with founders.
Skip it when
- You only need a one-line ask with no constraints, format, or context.
- You cannot supply the fill-in details the template asks for.
- A different category or tool-specific workflow fits the job better.
How to get a good result
Done when
1. Measurable question: Rank channels by revenue/session for week of Aug 4 - 10; for paid search, compare revenue to ad_spend. 2. Treatment: Used given aggregates as-is; no row-level dedupe possible. Assumed sessions and revenue are non-overlapping by channel. 3. Findings: - Email: $10.32/session (highest); revenue $9,800 on 950 sessions. - Organic: $4.40/session; $18,500 on 4,200 sessions (largest absolute revenue). - Paid search: $7.89/session; $14,200 revenue vs $3,200 spend → ~4.4x return on ad spend before other costs. - Social: $1.95/session; weakest efficiency. 4. Caveats: Channel attribution may double-count assisted conversions. No margin or LTV in the data. One week only - seasonality unknown. Ad spend only provided for paid and social. 5. Chart: Horizontal bar of revenue/session by channel, with a second series or annotation for ad spend on paid/social. 6. Next: Pull multi-touch attribution or last-click vs first-click for paid search over 4 weeks, plus contribution margin.
Common mistakes
- Leaving question, data, audience vague - the model then invents placeholders.
- Asking ChatGPT, Claude, or Gemini for a shorter answer after a long structured prompt - that throws away the format rules.
- Stopping when the first draft appears instead of checking it against the success criteria below.
FAQ
- What do I fill in before I copy this prompt?
- Replace these placeholders with your details: question, data, audience, grain. Required fields are marked on the page. The prompt text updates as you type.
- Which AI tools is this prompt written for?
- It works best with ChatGPT, Claude, Gemini. Copy the filled prompt and paste it into the tool you already use.
- What should a good result look like?
- 1. Measurable question: Rank channels by revenue/session for week of Aug 4 - 10; for paid search, compare revenue to ad_spend. 2. Treatment: Used given aggregates as-is; no row-level dedupe possible. Assumed sessions and revenue are non-overlapping by channel. 3. Findings: - Email: $10.32/session (highest); revenue $9,800 on 950 sessions. - Organic: $4.40/session; $18,500 on 4,200 sessions (largest absolute revenue). - Paid search: $7.89/session; $14,200 revenue vs $3,200 spend → ~4.4x return…
- What are common mistakes with this prompt?
- Leaving question, data, audience vague - the model then invents placeholders. Asking ChatGPT, Claude, or Gemini for a shorter answer after a long structured prompt - that throws away the format rules. Stopping when the first draft appears instead of checking it against the success criteria below.
Best for
- Analysts
- PMs
- Founders
Related Prompts
More prompts like this — same topic and style.
