Your starting point
Prerequisites: Use the ideas and vocabulary from lesson 36; revisit its worked examples if needed.
Definition: An analytical prompt specifies the question, available evidence, definitions and desired checks. A useful AI response is a proposal to inspect, not a substitute for source data or an independent calculation. This academy provides authored guidance without requiring a live AI connection.
Attempt the question before delegating
Write your hypothesis and a small calculation first. This gives you a reference against which to compare an AI response. Asking “analyze everything” encourages a broad narrative that may not match your decision. Instead, name one question and ask the assistant to identify missing information before making claims. You should understand the requested operation well enough to check a simple example.
Ask for boundaries and alternatives
Specify the row grain, units, period and denominator. Tell the assistant to separate observations, calculations, assumptions and recommendations. Request one alternative explanation and a list of claims that cannot be supported by the supplied fields. Asking for a confident answer can reward style rather than accuracy; asking for inspectable steps makes errors easier to find.
Keep responsibility with the learner
A model may not follow every instruction and can produce plausible but incorrect calculations. Copying a long chain of reasoning is not verification. Recalculate a selected result with the workbench or by hand and inspect the source records. Use prompts to obtain a hint, critique or checking question after your attempt, then explain the result independently in your own words.
Two worked examples
A bounded rate request
A useful prompt says: “These synthetic rows contain campaign, clicks and conversions. I calculate 10/100 = 10% for A. Check the denominator and compare with B, which has 9/30. Do not infer causation or customer satisfaction. Identify any additional fields needed for a cost comparison.” The constraints make unsupported claims easier to detect.
A learning-oriented request
Instead of “write my report,” try: “Here is my proposed finding and the calculation that supports it. Ask one question about a possible alternative explanation. Wait for my answer before suggesting a revision.” The learner still has to select evidence and defend the claim rather than merely approve generated prose.
Where this becomes useful
A beginner analyst can use AI as a reviewer of a narrowly defined question while retaining a local, verified workbook. A non-sensitive schema or synthetic example is often enough to request useful guidance.
Make an attempt
Use AI Analysis Verification Challenge. Write a prompt that includes your own calculation, the data definitions and one restriction on unsupported claims. Then inspect the authored case response and identify what you would independently verify.
Misconception check
Claim: A sophisticated prompt guarantees a correct analysis.
Inspect the correction
Prompts can improve clarity, but the response still needs verification against source data and mathematical rules.
Independent application
Write and test a prompt using only synthetic or non-sensitive information. Include a question, schema, first attempt and verification request. Record one useful suggestion, one claim requiring checking and your final independent explanation.
Completion records practice and self-review. It is not certification of statistical competence.
