ChatGPT Prompts for Data Science: 12 for EDA to Evaluation

ChatGPT prompts for data science help you plan the work around your own data: framing the question, exploring the dataset, building features, and checking a model fairly. The 12 prompts below move from problem framing to evaluation and write-up. Paste column names and summary stats instead of raw private data, and never treat a number the AI writes as a result until your own code produces it.

Share only schemas and summary stats, never raw customer records or confidential data, and report metrics you actually computed rather than estimates from the AI.

Frame the problem

1. Question to task

My business question is [question] and the decision it drives is [decision]. Restate it as a prediction or analysis task, the unit of prediction, and when the prediction is made.

2. Is ML needed?

For this question [question] and data [columns], tell me whether a simple rule, a dashboard, or a model is the right first step, and why.

3. Questions for the data owner

Here are my columns [columns]. List the questions I should ask the data owner about how each column is collected and when it becomes known.

Explore the data

4. EDA checklist with pandas

Write an EDA checklist with pandas snippets for this dataset [columns and types]: missing values, duplicates, target balance, outliers, and time trends.

5. Read my summary stats

Here is my df.describe() and df.info() output [paste]. Point out anything suspicious and what to check next. Do not guess values that are not shown.

6. Leakage check

The prediction is made at [moment]. For each column [columns with timing], say whether it is known at that moment, and flag any that could leak the target.

Build and evaluate

7. Feature ideas

Suggest 5 to 8 features for predicting [target] using only columns known at prediction time [columns]. Give a one-line reason for each.

8. Baseline and first model

Suggest a simple baseline and one first model for [task], with a scikit-learn pipeline stub that uses these columns [numeric and categorical lists].

9. Metric that fits the decision

The team will act on [how predictions are used, such as calling the top 200]. Which metric fits, and why could accuracy mislead with this class balance [rate]?

10. Validation split

My data covers [time span] with one row per [unit]. Design a train, validation, and test split that respects time and avoids the same entity leaking across splits.

Explain the results

11. Results table template

Make a results table template for these models [list] and metrics [list]. Leave every value blank for me to fill in.

12. Plain-language readout

Here are my real results [paste table]. Write a short readout for [audience]: what the model does, how well, its limits, and the next step. Use only my numbers.

Plan the whole project in one prompt

The flagship Data Science Prompt: EDA Plan, Features, and Model Checks takes your question, columns, summary stats, target, column timing, and stack. It returns problem framing, an EDA checklist, leakage flags, feature ideas, a baseline and model stub, the metric and validation split, and a blank results table. Its churn example drops a cancel reason column that would leak the answer.

Tips for data science with ChatGPT

Share the schema and summary stats rather than raw customer data, and follow your company's data policy. Run every snippet yourself, since AI can write code that looks right but uses the wrong column or a leaky split. Ask why a metric fits your decision before you report it.

Related prompts

Write analysis code with the Python data analysis pandas script generator or clean inputs with the pandas ETL contract from a messy CSV. For spreadsheet work, read ChatGPT prompts for Excel data analysis, and if you are new to the language, start with the ChatGPT prompt for learning Python.

Find more in the Coding category, look through Browse, or post your own on Share.