Opening the spreadsheet and making charts is the fastest route to a dead end. A good analysis starts with the decision you need to make, then works backward to the data. These five ChatGPT prompts build that plan in order, so you spend your time on the analysis, not on guessing what to do first.
Why analysis plans stall
- The analysis starts with the data instead of the question, so it produces interesting facts that answer nothing.
- Data quality gets skipped, and a conclusion later falls apart because of duplicate rows or missing values nobody checked.
- Findings stop at the number and never reach a recommendation, leaving the reader to guess what to do.
Prompts that fix it
1. Analysis plan from a question
I need to answer this question: [business question]. The decision it informs is [decision to make]. Build an analysis plan that lists what to measure, what to compare, and what a result in each direction would tell me. Start from the question, not the data.What it does: Anchors the whole analysis to the decision instead of the dataset.
How to use: replace [business question] and [decision to make]; keep the rest as is.
2. Data quality and cleaning plan
Here is my dataset: [dataset]. Known issues so far are [known issues]. Write a cleaning checklist covering duplicates, missing values in [key fields], and inconsistent formats. For each check, tell me what to look for and how to decide whether a row is safe to keep.What it does: Catches the problems that quietly break a result later.
How to use: replace [dataset], [known issues], and [key fields]; keep the rest as is.
3. Method selection
I want to answer [question] using [data type]. Help me pick the method that fits: compare two groups, track a trend over time, or find what drives an outcome. Explain why each method does or does not fit, and name the assumption I would need to check first.What it does: Matches the method to the question and surfaces its assumptions.
How to use: replace [question] and [data type]; keep the rest as is.
4. Analysis outline with checks
Write a step-by-step analysis outline for testing this hypothesis: [hypothesis]. Use the variables [variables]. After each step, add a check that would tell me the result is unreliable. Keep the steps in the order I should run them.What it does: Adds a self-check to every step so problems surface early.
How to use: replace [hypothesis] and [variables]; keep the rest as is.
5. From findings to recommendation
Here are my results: [analysis result]. My audience is [stakeholder], and the decision on the table is [decision]. Write three findings, one caveat, and one recommendation. Keep the language plain and tie every finding back to the decision.What it does: Turns raw results into a conclusion someone can act on.
How to use: replace [analysis result], [stakeholder], and [decision]; keep the rest as is.
Rewrite the ask
| Weak ask | Better ask |
|---|---|
| Analyze this data and tell me what you find. | Plan an analysis for [business question] that supports [decision to make]. List what to measure, what to compare, and the checks that would make a result unreliable. |
| Is this data clean? | Write a cleaning checklist for [dataset] covering duplicates, missing values in [key fields], and format issues, with a rule for keeping or dropping each row. |
Both fixes come down to one habit: state the decision first. Everything else in the plan follows from it.
Keep going
- More Data & Analysis prompts
- Claude prompts for chart critiques
- Gemini prompts for ROI calculations
- ChatGPT prompts for keyword research
GuPrompt keeps every prompt copy-paste ready. No account needed.