I want to analyze data.

Every question is optional. The ones marked Recommended matter most.

Question. What do you want the data to tell you? Recommended

For example: Which customer segments drove the drop in repeat orders last quarter?

Decision. What decision or action will this inform? Recommended

For example: Where to focus a $5k win-back campaign.

The data. What is one row, which columns matter, what period, where did it come from? Recommended

For example: One row per order, Jan 2025 to Sep 2026. Columns: order_date, customer_id, segment, total.

Upload a spreadsheet file to your AI tool, not a screenshot, when exact values matter.

Definitions. How do you define the key terms? Recommended

For example: 'Repeat customer' means 2 or more orders in 12 months.

Output. What should you get back, and for whom? Recommended

For example: 3 charts and a 5-point summary for my operations manager.

Known issues. Any problems you know about in the data?

For example: October has missing segment values.

Comparisons. What should be compared or broken down?

For example: This quarter against the same quarter last year, by segment.

Your hunch. What do you expect to find?

For example: I think the price rise hit the budget segment.

Method. Any methods you want, or want avoided?

For example: Simple percentages; no forecasting.

Show work. Do you want the calculations or code shown?

For example: Yes, so I can check them.

More detail, if you have it

How should the AI start?

Ask only if neededInterview me firstJust go

Saved on this device only

How to brief an AI to analyze your data

Use this form when you have a spreadsheet, an export or survey results and want answers from them: what changed, what drove it, what to do next. What you tell the AI about the data matters as much as the question.

Why the recommended questions matter

Question. Ask what you want the data to tell you, not “look at this”. Anthropic’s Claude Academy example asks what decisions the results will inform and what patterns you want to surface.
Decision. Say what the analysis will inform. Anthropic’s Claude Academy example on program data starts by asking who the results are for and what decisions they will inform.
The data. Say what one row is, which columns matter, what period it covers and where it came from. The same Claude Academy example asks for the time periods and the measures that matter.
Definitions. Define the terms the question depends on, such as what counts as a repeat customer. A Carnegie Mellon study of underspecified prompts found that simply listing more requirements does not reliably help, so define the few that matter.
Output. Say what you want back and who will read it. The Claude Academy example asks whether the audience is board members, funders or staff, because that changes how simple the charts should be.

If you expect a particular answer, add “Your hunch”. The form turns it into something to test, because Anthropic’s research on sycophancy found that AI assistants tend to agree with the user.

A worked example

An online shop wants to know why repeat orders fell. The form composes:

I want my data to answer this: Which customer segments drove the drop in repeat orders last quarter?

<context>
The decision this informs: Where to focus a $5k win-back campaign.
Definitions: 'Repeat customer' means 2 or more orders in 12 months.
What I expect to find: I think the price rise hit the budget segment.
Treat what I expect as something to test. Say so if the data does not support it.
</context>

<source_material>
About the data: One row per order, Jan 2025 to Sep 2026. Columns: order_date, customer_id, segment, total.
</source_material>

<task>
Check the data for quality problems first and tell me what you find. Use code for calculations. State your assumptions.
</task>

<output_format>
What I want back, and for whom: 3 charts and a 5-point summary for my operations manager.
</output_format>

Today's date is November 10, 2026. If the information here is not enough to answer well, say so and tell me what you would need. If something missing would materially change the result, ask me up to 3 short questions first. Otherwise proceed and state your assumptions.

What the form adds for you

It asks the AI to check the data for quality problems before drawing conclusions, as the Claude Academy example advises, to use code for calculations, and to state its assumptions.