22 Claude Prompts for Data Analysis (2026)

Give me a first look at this dataset. Use the analysis tool to compute, don't estimate.

The question I care about is [question].
<data>
[attach the CSV/Excel, or paste it; describe the columns]
</data>

Do this:
- Data-quality check first: row count, columns and types, missing values, duplicates, obvious anomalies.
- A quick statistical summary of the key columns.
- The 3 things that stand out most.
- What I should ask next given my question.

Answer a specific question from my data. Compute it with code and show your work.

Question: [precise question, e.g. "what was revenue by region in Q2 vs Q1?"].
<data>
[attach or paste the data]
</data>

Do this:
- Give the exact answer with the numbers.
- Show the calculation/logic so I can verify it.
- Note any assumption you made (date ranges, filters, how you handled blanks).
- Flag if the data can't actually answer this.

Find the narrative hidden in my data. Think step by step.

Audience for the story: [exec / team / customers].
<data>
[attach or paste the data]
</data>

Do this:
- Surface the 3 most surprising or decision-relevant patterns.
- For each: the finding, the evidence (numbers), and the "so what".
- Note any correlation I'd be wrong to read as causation.
- End with the single headline finding, in one sentence.

Break my data into meaningful segments and compare them. Use code for exact figures.

<data>
[attach or paste the data]
</data>
Segment by: [dimension, e.g. plan, region, cohort] and compare on [metric].

Do this:
- Show each segment's key metrics in a table.
- Call out the biggest differences and whether they're likely meaningful.
- Note segments too small to trust.
- Suggest the one segment worth a closer look and why.

Find anomalies and outliers in this data. Compute with code.

<data>
[attach or paste the data]
</data>
Context: [what "normal" looks like, if you know].

Do this:
- Identify outliers and unusual patterns, with the rows and values.
- For each, offer the most likely explanations (real signal vs. data error).
- Tell me which to investigate first.
- Don't drop anything silently — list what you flagged.

Give me a working spreadsheet formula and teach me why it works.

Tool: [Excel / Google Sheets]. My columns: [e.g. dates in A, amount in B, category in C]. I want to [what to calculate].

Do this:
- The exact formula to paste, using real cell references.
- Each part explained in one line.
- A cleaner alternative (pivot table, QUERY, SUMIFS) if there is one.
- Edge cases to watch (blanks, text vs. numbers, ranges).

My spreadsheet formula is wrong. Fix it.

Tool: [Excel / Google Sheets].
<formula>
[paste the formula]
</formula>
What it should do: [intent]. What it returns instead: [error or wrong value].

Do this:
- Tell me exactly what's wrong and why.
- Give the corrected formula.
- Point out anything fragile that will break later.

Tell me how to set up a pivot table to answer my question.

Tool: [Excel / Google Sheets]. My data has columns: [list]. I want to see [what].

Do this:
- Exactly what to put in Rows, Columns, Values, and Filters.
- Which aggregation (sum, count, average) and why.
- How to sort or format it to read clearly.
- One follow-up view worth adding.

Help me design a spreadsheet model.

Goal: [what it should calculate, e.g. a simple budget / pricing / forecast model]. Inputs I have: [list]. Outputs I want: [list].

Do this:
- Lay out the tabs/sections and what each holds.
- List the input cells vs. calculated cells.
- Give the key formulas.
- Add checks that warn me if an input looks wrong.
Keep it simple enough that I can maintain it.

Turn my data into a clean chart as an Artifact.

<data>
[attach or paste the numbers]
</data>
The point I want it to make: [message].

Do this:
- Pick the chart type that best shows that point; say why in one line.
- Render it as an interactive Artifact (self-contained).
- Label axes, add a clear title, format numbers readably.
- Suggest one alternative view of the same data.

Build a small dashboard as an Artifact from this data.

<data>
[attach or paste the data]
</data>
Audience: [who]. The 3-4 things they care about: [metrics].

Do this:
- A single self-contained HTML Artifact with 3-4 charts + key numbers up top.
- Choose the right chart per metric; keep it clean and readable.
- Add a one-line takeaway under each chart.
- Note any metric the data can't support yet.

Critique this chart and make it clearer.

<chart>
[attach a screenshot of the chart, or describe/paste it]
</chart>
What I want it to communicate: [message].

Do this:
- Say whether the chart type and design serve that message.
- Flag anything misleading (truncated axes, wrong scale, chart junk).
- Suggest specific fixes, and a better chart type if needed.
- If useful, rebuild the improved version as an Artifact.

Explain what this chart is really showing.

<chart>
[attach a screenshot or image of the chart]
</chart>

Do this:
- Tell me the main takeaway in one sentence.
- Point out what the chart is NOT showing that I might assume.
- Flag any way the visual could be misleading.
- List 2 questions I should ask about the underlying data.

Explain a statistical concept or result in plain language.

Explain [concept or result, e.g. "p-value", "our A/B test's 3% lift with p=0.08"] to [me / my team].

Do this:
- A one-sentence plain-English meaning.
- What it does and does NOT let me conclude.
- One concrete example.
- The mistake people make when interpreting this.

Sanity-check my experiment. Think step by step.

Setup: [what was tested, metric, variants]. Results:
<results>
[sample sizes, conversions/values per variant]
</results>

Do this:
- Compute the difference and whether it's likely real vs. noise (with reasoning).
- Flag issues: too-small sample, peeking, novelty effect, unequal groups.
- Tell me what I can honestly claim from this.
- What I'd need to be more confident.

Make a defensible forecast from my history. Think step by step and compute with code.

<data>
[paste the time series, e.g. monthly revenue]
</data>
Forecast [metric] for [period].

Do this:
- State the trend and any seasonality you detect.
- Give base, optimistic, and conservative projections with the assumptions behind each.
- List what could break the forecast.
- Be explicit that this is an estimate, not a guarantee.

Help me pick the right way to analyze this. Think step by step.

My question: [what I want to know]. My data: [describe what you have — columns, size, type].

Do this:
- Recommend the analysis/approach that fits, and why.
- Warn me about assumptions it requires and whether my data meets them.
- Suggest a simpler fallback if the ideal approach is overkill.
- Tell me how I'd know the result is trustworthy.

Write an executive summary of my analysis.

Audience: [exec / board / client]. They have [time] and care about [priority].
<findings>
[paste your findings or the analysis]
</findings>

Do this:
- Lead with the headline finding and the recommended action.
- 3-5 supporting points, each with the number behind it.
- The one risk or caveat they must know.
- Keep it to [length]; no jargon.

Turn this analysis into a recommendation. Think step by step.

<findings>
[paste the findings]
</findings>
The decision on the table: [decision].

Do this:
- State a clear recommendation.
- Tie each supporting reason to a specific finding.
- Give the strongest counterargument and why you still recommend it (or don't).
- Note what would change your mind.

Clean this messy data into something usable. Use code where it helps.

<raw>
[paste or attach the raw data]
</raw>
I want it as [target format, e.g. a table with columns X, Y, Z].

Do this:
- Normalize inconsistent formats (dates, casing, units, spelling).
- Flag duplicated or suspect rows instead of silently dropping them.
- Output the clean version I can paste into a spreadsheet.
- List every transformation you applied so I can audit it.

Help me define a metric that actually reflects what I care about.

What I'm trying to measure: [the real thing]. Data I have access to: [sources].

Do this:
- Propose 2-3 candidate metrics with exact definitions and formulas.
- For each, what it captures and how it could be gamed or misleading.
- Recommend one, plus a guardrail metric to watch alongside it.
- How often to review it.

Help me understand why a metric moved. Think step by step; compute with code.

Metric: [metric]. It went from [X] to [Y] over [period].
<data>
[attach the relevant data — by segment, over time]
</data>

Do this:
- Decompose the change: which segments/drivers account for how much of it.
- Rank the contributing factors by size.
- Separate what the data shows from what's a hypothesis.
- Tell me the one thing to check next to confirm the cause.