AI Prompts for Statistical Analysis and Interpretation
Prompts that help you choose a test, check assumptions, read output and write up results. AI is good at explaining and drafting, but it can make arithmetic and logic mistakes, so always verify the numbers.
Rules for using AI with statistics
- Give context, not raw sensitive data. Describe your variables and paste summary output instead of personal records.
- Verify every number. Re-run the calculation in your software or with a trusted calculator, such as our free tools.
- Ask it to explain its reasoning so you can spot a wrong assumption.
- Know the basics yourself. You must be able to defend your choice of test to a supervisor or reviewer.
1. Choose the right test
I want to answer this research question: [question]. My outcome variable is [variable, type: continuous/ordinal/binary/categorical]. My predictor or grouping variable is [variable, type, number of groups]. The groups are [independent/paired]. Sample size is [n]. Recommend the most suitable statistical test, explain why, list its assumptions and name a non-parametric alternative if assumptions fail.
Compare the answer with our guide to choosing a statistical test.
2. Check the assumptions
I plan to run [test]. List the assumptions in plain language, tell me how to check each one in [Excel/SPSS/R/Python] and what to do if an assumption is violated. Then explain what a Shapiro-Wilk p-value of [value] and a Levene test p-value of [value] mean for my choice.
3. Interpret software output
Below is output from [software] for [test]. Explain each number in plain language: the test statistic, degrees of freedom, p-value, confidence interval and effect size. Then tell me what the result means for my research question and what it does not prove. Output: [paste summary table, no personal data]
4. Report results in APA style
Write a results sentence in APA 7 style for an independent-samples t-test. Group A: M = [ ], SD = [ ], n = [ ]. Group B: M = [ ], SD = [ ], n = [ ]. t([df]) = [ ], p = [ ], 95% CI of the difference [ , ], Cohen’s d = [ ]. Use correct rounding and italics notation, and add one sentence of interpretation.
5. Explain a result to a non-specialist
Explain this finding to a programme manager with no statistics background in three sentences, without jargon: [paste result]. Include how confident we can be and one limitation.
6. Plan a sample size
I want to compare two groups on a continuous outcome. I expect a medium effect (d = 0.5), want 80% power and alpha = 0.05. Explain how to calculate the sample size per group, give the formula, and list what would change the answer. Do not give a final number without showing the working.
For proportion-based surveys, use our sample size calculator.
7. Write Excel formulas
My data are in Excel: column A is Group (Online/In-person), columns C to G are five Likert items scored 1 to 5, and some cells are blank. Give me formulas to (1) count missing values per row, (2) compute the mean score only for rows with no missing items and (3) calculate the mean and standard deviation by group. Use functions that work in Excel 2016.
8. Write R or Python code
Write [R/Python] code to read a CSV with columns [names], recode items [list] as reverse-scored on a 1 to 5 scale, compute Cronbach’s alpha for items [list], and compare the mean scale score between two groups with Welch’s t-test and a Mann-Whitney test. Add comments and print results to the console.
Run the code on a small test dataset you can check by hand before trusting it on your real data.
9. Find mistakes in your analysis plan
Act as a statistics reviewer. Here is my analysis plan: [paste]. List problems such as wrong test choice, multiple comparisons, ignoring clustering, missing-data handling, treating ordinal data as continuous and over-interpreting p-values. Rank them by seriousness.
10. Write the limitations
My study used [design], sample [n, how selected], and [analysis]. Write a limitations paragraph of 120 words that is honest and specific, covering sampling, measurement, missing data and generalisability, without undermining the study.
Common mistakes with AI and statistics
- Trusting a calculated p-value or mean without re-checking it.
- Letting the tool pick a test without telling it the data type and design.
- Accepting “significant” as “important”. Always ask for effect sizes and confidence intervals.
- Pasting a full dataset with personal details.
For a worked example with real output, see our survey analysis worked example. If you would like a specialist to run and check the analysis, see statistical analysis services or survey data analysis.
Frequently asked questions
Can AI do my statistical analysis for me?
It can explain methods and draft code, but it can make errors and does not know your study. You should run the analysis in proper software and verify the output.
Is it acceptable to use AI to write my results section?
Use it for wording and structure only, and follow your institution’s rules on AI use. The numbers and interpretation must come from your own verified analysis.
Why does AI give different answers to the same question?
Language models are not deterministic. If two answers disagree, treat that as a warning and check a textbook or software output.
Which prompt helps beginners most?
The test-choice prompt and the interpret-output prompt, used together with a statistics guide.
Related guides, tools and services
Want an expert to check or run your analysis?
Send us your question and data description. We reply with a clear plan and a quote.
Discuss your project on WhatsApp