If you've tried using ChatGPT for data analysis, you've probably experienced both its surprising capabilities and its frustrating limitations. It can write a Python script that cleans your dataset in seconds. It can also confidently return a completely wrong number.

This post covers what ChatGPT actually does well for data work, where it breaks down, and what purpose-built AI analytics tools do differently.

What ChatGPT Can Actually Do With Data

ChatGPT is a large language model. It was trained on text — including enormous amounts of code, documentation, and data-related writing. That makes it genuinely useful for certain data tasks:

Writing code

Ask ChatGPT to write a Python script using pandas, and it usually produces working code. This is legitimately useful if you know Python well enough to review and run the output.

Explaining concepts

"What's the difference between correlation and causation?" "How do I interpret a p-value?" ChatGPT is excellent at plain-language explanations of statistical concepts.

Cleaning data schemas

Paste a messy CSV header row and ask it to suggest clean column names. It handles this well.

Writing summaries

If you paste in a table of numbers, ChatGPT can write a paragraph summarizing the key trends. This is genuinely time-saving.

Formula help

"How do I write a SUMIF formula that groups by month?" ChatGPT answers formula questions better than most Google searches.

Where ChatGPT Falls Short for Data Analysis

1. It Doesn't Have Direct Access to Your Data

ChatGPT can't connect to your Google Sheets, Postgres database, or CSV file on your desktop. You have to manually paste data into the conversation, which means:

• You're limited by context window size (large datasets won't fit)
• You have to manually re-paste data every time
• There's no live connection — stale data is a constant risk
• Sensitive data gets pasted into a third-party chat interface

ChatGPT's file upload feature (in Plus/Pro plans) helps with smaller files but doesn't solve the connection problem for ongoing analysis.

2. It Hallucinates Numbers

This is the critical one. ChatGPT is trained to produce plausible-sounding text. When asked to calculate something, it will often return a confident-looking number that is simply wrong — especially for multi-step calculations, aggregations across many rows, or percentages derived from other percentages.

The model doesn't actually run calculations the way a spreadsheet or database does. It predicts what the answer should look like based on its training data. For small, simple calculations it usually gets it right. For anything complex, you need to verify.

Always verify ChatGPT's numeric outputs before presenting them. This is not optional.

3. The Context Window Limits Dataset Size

Even with file uploads, ChatGPT's context window limits how much data it can reason about at once. A dataset with 50,000 rows and 20 columns simply won't be analyzed accurately — the model either truncates the data or reasons about a sample without telling you.

4. It Can't Run Persistent Analysis

Every ChatGPT conversation starts fresh. There's no concept of a saved dashboard, a repeatable query, or a report that updates when new data comes in. Every analysis is one-off, which means every time you need the same insight again, you start over.

5. No Visualization Without Code

ChatGPT can write matplotlib or seaborn code to generate a chart, but you need a Python environment to run it. It can't show you a chart directly unless you're using the Advanced Data Analysis feature (Code Interpreter) in ChatGPT Plus. And even then, the charts are basic and not interactive.

The Code Interpreter Feature

ChatGPT's Advanced Data Analysis (formerly Code Interpreter) is the most powerful option for data work within ChatGPT. You upload a file, and ChatGPT runs actual Python code against it — which means real calculations, not hallucinated numbers.

It's genuinely impressive for one-time exploratory analysis. The limitations:

• Requires a paid ChatGPT Plus or Pro plan
• Maximum file size limits apply

ChatGPT vs. Purpose-Built AI Analytics Tools for Data Work

If you've tried using ChatGPT for data analysis, you've likely seen both sides of it: it can write a Python script in seconds, yet also return a confident but completely wrong number.

This summary breaks down what ChatGPT actually does well for data work, where it breaks down, and how purpose-built AI analytics tools (like Qunta) differ.

What ChatGPT Can Actually Do With Data

ChatGPT is a large language model trained on text, including code, documentation, and data-related writing. That makes it genuinely useful for several data-adjacent tasks: