25 AI Prompts for Data Analysis That Drive Insights

Data analysis is where numbers turn into decisions — but only when the data is clean, the method fits the question and the findings are explained clearly. Well-written AI prompts for data analysis speed up every one of those steps, from auditing a new dataset to writing the SQL, checking your statistics and turning results into a story stakeholders will act on.

Below are 25 prompts across five stages of the workflow: exploring and cleaning data, planning the analysis, writing code, finding insights, and reporting results. Each one works in any model — paste it into a free AI chat, fill in the brackets, and verify the output against your own data before acting on it.

A vague request like ‘analyse this data’ gets a generic checklist back. The prompts that produce useful analysis give the model four things:

• Context: what the data represents, where it came from, and the decision it needs to inform
• Structure: column names, data types, row counts or a small sample — never the full sensitive dataset
• Constraints: your tools (SQL dialect, Excel version, Python libraries) and the level of rigour required
• Output format: a table, a numbered plan, runnable code or a one-paragraph summary — say which

Model choice matters too. Claude Sonnet 5 tends to be strong at careful interpretation and spotting weak assumptions, while GPT-6 Astra is reliable for structured plans and complex, multi-step analysis. Running the same question through both is a quick way to catch blind spots.

Most analysis errors start before the analysis does. These prompts help you understand a new dataset and fix it before drawing conclusions. If the data dictionary arrives as a long PDF, condense it first with an AI PDF summarizer so you can paste the key definitions into the prompt. For quick checks on a small column sample, a lightweight model like Gemini 3.1 Flash Lite is usually enough.

1. Dataset orientation brief

I have a new dataset: [describe the source, number of rows and columns, and key variables]. My goal is to answer [business or research question]. List the first questions I should ask about this data, the quality checks to run immediately, the variables most relevant to my goal, and the questions this dataset cannot answer.

2. Data quality audit

Audit this dataset for quality issues. Here are the column names, data types and a 20-row sample: [paste]. Check for missing values, duplicates, inconsistent formats, impossible values and outliers. Return a table with each issue, the affected column, how serious it is and a suggested fix.

3. Step-by-step cleaning plan

Create a step-by-step cleaning plan for this dataset: [describe columns and known issues]. I will work in [Excel / SQL / Python pandas]. For each step, explain what to do, why it matters and the exact function or code to use. Flag any step that could remove valid data.

4. Missing values strategy

Column [name] in my dataset is [X]% missing. The data is about [describe], and the column will be used for [describe]. Compare the options — dropping rows, simple imputation, model-based imputation or a missing-indicator flag — and recommend one, explaining how each would bias my results.

5. Exploratory analysis plan

Design an exploratory data analysis plan for [describe dataset]. Include the summary statistics to calculate, the distributions and relationships worth plotting, the segments to compare, and the three patterns most likely to be worth investigating further. Keep it to what I can finish in one afternoon.

Use these when you need to choose a method, test a hypothesis or check whether a result actually holds up. GPT-5.6 Sol is a solid pick for step-by-step statistical reasoning, and an AI math calculator is handy for double-checking sample sizes and test statistics.

6. Business question translator

Translate this stakeholder question into an analysis plan: [question as the stakeholder asked it]. My available data: [describe]. Restate it as a specific, measurable question, name the metrics that best capture it, outline the approach, and flag any gap between what is being asked and what my data can answer.

7. Statistical method selector

I want to know whether [describe the relationship or difference]. My data: [describe variable types, sample size and how it was collected]. Recommend the most appropriate statistical test or model, list its assumptions, explain how to check each one, and suggest an alternative if an assumption fails.

8. Correlation vs causation check

I found that [describe the relationship]. The data is [observational / experimental / quasi-experimental]. Assess how strong the case for causation is, list the most plausible alternative explanations, and write a one-sentence finding that claims exactly as much as the evidence supports.

9. A/B test designer

Help me design an A/B test for [feature, message or price change]. Hypothesis: [describe]. Primary metric: [describe]. Current baseline: [value]. Estimate the sample size needed, name the main threats to validity (sample ratio mismatch, novelty effects, peeking, multiple comparisons), and tell me how to read a borderline result.

10. Regression output interpreter

Explain this regression output in plain language: [paste output]. The model predicts [outcome] using [predictors]. Tell me which coefficients matter and what they mean in real units, whether the model fits well, any warning signs such as multicollinearity, and what I can and cannot conclude.

These prompts turn a plain-English question into working code. Code-focused models such as DeepSeek V4 Pro handle multi-step queries and pandas scripts well, while DeepSeek V4 Flash is faster for small fixes and one-line formulas. Always run the result on a small sample first.

11. SQL query writer

Write a [PostgreSQL / BigQuery / MySQL] query to [describe what you need]. Tables: [table names, columns and how they join]. Add comments explaining each step, handle NULLs explicitly, and tell me how to check that the result is correct.

12. SQL query optimiser

This query runs slowly on [number] rows: [paste query]. Explain why it is slow, rewrite it for better performance, and suggest indexes or structural changes. Confirm that the rewritten query returns exactly the same results.

13. Excel formula builder

In Excel, I need a formula that [describe the calculation]. My data sits in [describe columns and ranges]. Give me the formula, explain how it works piece by piece, and offer a version using [XLOOKUP / LET / dynamic arrays] if it is simpler.

14. Python pandas script

Write a Python script using pandas that loads [file type], [describe the transformations: filtering, grouping, merging, pivoting], and outputs [describe the result]. Add comments, use clear variable names, and include a quick check that row counts make sense after each step.

15. Code explainer and debugger

Here is code I inherited: [paste SQL or Python]. Explain what it does step by step, point out any bugs or logic errors, flag anything that could silently produce wrong numbers, and suggest a cleaner version.

Results only matter once they point to a decision. These prompts help you separate real signal from noise. Gemini 3 Pro handles longer reasoning chains when you are weighing several explanations, and asking Grok 4.5 the same question gives you a useful second opinion.

16. Insight extractor

Here are my analysis results: [paste or describe]. The decision they need to inform: [describe]. Pull out the 3–5 most decision-relevant insights, separate strong findings from suggestive ones, flag anything that could be a confounder or data artefact, and phrase each insight as a recommendation.

17. Anomaly investigator

My data shows [describe the spike, dip or unusual pattern] during [time period]. The normal baseline is [describe]. Rank the most likely explanations, tell me how to test each one against the data, and help me decide whether this is a real event or a data quality issue.

18. Segmentation and cohort analysis

Suggest a segmentation for [customers / users / products] using these variables: [list]. My goal is to [describe]. Recommend the segments or cohorts to compare, the metrics to compare them on, and how to tell whether differences between segments are meaningful rather than random.

19. Trend and forecast check

Here is [metric] by [day / week / month] for the last [period]: [paste data]. Describe the trend, seasonality and any structural breaks, suggest a simple forecasting approach for the next [period], and explain how confident I should be in it.

20. KPI and metric designer

Design a metrics framework for [team, product or goal]. Recommend 3–5 core metrics, mark each as a leading or lagging indicator, define exactly how to calculate it, and flag the metric most likely to be gamed if it becomes a target.

Good analysis still fails if nobody understands it. These prompts help you choose the right chart, build the narrative and prepare for tough questions. For long draft reports, an AI summarizer can cut the text down first, and Claude Haiku 4.5 is quick at producing several versions of a summary for different audiences.

21. Chart selector

I want to show [describe the comparison, trend, distribution or relationship] to [audience]. My data: [describe variables]. Recommend the best chart type, explain why it beats the obvious alternatives, and give specific guidance on axes, labels, colour and what to leave out.

22. Data story builder

Turn these findings into a data story for [audience] who need to decide [decision]: [describe results]. Open with the most important finding, build a logical case toward one recommendation, address the likely objections with data, and end with a clear next step.

23. Executive summary writer

Write a 150-word executive summary of this analysis for [audience]: [paste key findings]. Lead with the conclusion, include no more than three supporting numbers, state the main limitation in one sentence, and end with the recommended action.

24. Dashboard layout planner

Plan a dashboard for [audience] tracking [goal]. Recommend the top-level KPIs, the supporting charts, the filters users need, and the layout from top to bottom. Tell me what to remove so the dashboard answers one question well.

25. Analysis sanity check

Act as a sceptical senior analyst reviewing my work before I present it: [describe the question, data, methods and conclusions]. Identify questionable methodological choices, conclusions that go beyond the data, alternative interpretations I missed, and the three questions a stakeholder is most likely to ask — with how to answer each.

Chat Smith gives you Claude, GPT, Gemini, Grok and DeepSeek in one place — from fast everyday models like Gemini 3.5 Flash to advanced reasoning models — so you can write the SQL with one model, check the interpretation with another and compare answers side by side. Save the prompts you use most as templates and they are ready for every new dataset.

Once your findings are ready, the AI email writer turns the summary into a stakeholder update in seconds. The analysis still needs your judgement — these prompts just get you to a sound answer faster.