Data analysis used to require a degree. Now it requires a subscription.

There's a dirty secret in business: most companies are sitting on data they've never properly analysed. Sales figures in spreadsheets. Customer feedback in survey responses. Website analytics in dashboards nobody reads. The data exists. The insights don't, because extracting insights used to require someone who could write Python or R, and hiring that person costs £40,000-60,000 per year.

AI has changed this equation completely. Not in the "AI will replace data analysts" way that LinkedIn influencers claim, but in the "you can now ask meaningful questions about your data without writing code" way that's genuinely useful.

I'm going to walk you through analysing a real dataset using the three major AI tools: chatgpt, claude, and gemini. Same data, same questions, honest comparison of the results.

The dataset: 12 months of sales data

For this walkthrough, I'm using a CSV file with 12 months of fictional (but realistic) sales data for a small e-commerce business. The columns include: date, product name, category, quantity sold, unit price, total revenue, customer region, and marketing channel. About 2,400 rows covering January through December.

This is the kind of data that most small businesses have in a spreadsheet somewhere. The kind that gets exported from Shopify, WooCommerce, or a POS system. Nothing exotic.

ChatGPT with Code Interpreter: the best overall experience

ChatGPT's Code Interpreter (now called "Advanced Data Analysis") is the gold standard for this use case. Here's the workflow.

Step 1: Upload the file. Click the attachment icon in ChatGPT, upload your CSV. ChatGPT will acknowledge the file and show a brief preview.

Step 2: Ask for an overview. The first prompt: "Analyse this sales data and give me the top 5 most important insights. Show me any trends, patterns, or anomalies you find."

What ChatGPT does behind the scenes: it writes Python code (using pandas), runs it in a sandboxed environment, analyses the data, generates charts, and presents the results in natural language. You see the insights, not the code (though you can expand it if you want).

The output I got: Revenue peaked in November (Black Friday effect), with a 340% spike in electronics sales The "Organic Search" marketing channel had the highest ROI, while "Paid Social" had declining returns after Q2 Product category "Home & Living" showed consistent month-over-month growth of 8%, suggesting a reliable trend rather than seasonal spikes The North West region underperformed by 23% compared to similarly-sized regions, which could indicate a distribution or marketing gap Average order value dropped 15% between Q1 and Q4, driven by a shift toward lower-priced products in the best-sellers

Step 3: Go deeper. "Create a chart showing monthly revenue by category. Use a stacked bar chart."

ChatGPT generates the chart directly in the conversation. The visualisation is clean, properly labelled, and actually useful. You can ask for modifications: "make it a line chart instead," "add a trend line," "show only the top 3 categories."

Step 4: Ask specific business questions. This is where it gets powerful. "If I could only run marketing on two channels next quarter, which two would you recommend based on this data and why?" ChatGPT analyses the channel performance data, calculates cost per acquisition and return per channel, and gives a specific recommendation with reasoning.

Step 5: Export. "Create a summary report of your findings as a downloadable PDF." ChatGPT generates a formatted report that you can download and share with your team.

ChatGPT verdict: The best tool for data analysis by a clear margin. The Code Interpreter gives it a genuine execution environment, so it's not just describing what it would do, it's actually running the analysis. Charts are generated in real-time. The iterative workflow (ask a question, get an answer, ask a follow-up) feels natural and productive.

Claude: better explanations, no execution environment

claude's approach to data analysis is fundamentally different. You can upload a CSV (Claude supports file uploads), and Claude will analyse the structure and content. But Claude doesn't have a code execution environment like ChatGPT's Code Interpreter. It reads the data, reasons about it, and provides analysis, but it can't run Python scripts to crunch numbers.

What Claude does well:

The initial analysis prompt ("give me the top 5 insights") produced insights that were, in some ways, more nuanced than ChatGPT's. Claude was better at contextualising the findings: "The November revenue spike is almost certainly Black Friday-driven, which means it's not indicative of organic growth. Your underlying monthly growth rate, excluding November, is closer to 4% rather than the 12% the raw numbers suggest."

Claude also excelled at explaining why patterns might exist, not just identifying them. Where ChatGPT said "Paid Social had declining returns after Q2," Claude said "Paid Social's declining returns after Q2 could indicate audience fatigue, increased competition in your ad categories, or seasonal shifts in social media usage patterns. I'd look at your ad creative refresh frequency and CPM trends to narrow down the cause."

What Claude doesn't do well:

No charts. Claude can describe what a chart would show and even write the Python code to generate one, but it can't execute that code and display the result. For someone who wants visual analysis, this is a significant limitation.

The numerical analysis is also less reliable. Without a code execution environment, Claude is doing mental arithmetic on 2,400 rows of data. For broad pattern recognition, it's fine. For precise calculations ("what was the exact month-over-month growth rate for Home & Living?"), ChatGPT's actual computation is more trustworthy.

Claude verdict: Better at explaining insights and providing business context. Worse at precise calculations and visualisation. Best used as a complement to ChatGPT: use ChatGPT to run the numbers and generate charts, then paste the key findings into Claude and ask "what should I do about this?"

Gemini: good integration, middling analysis

gemini's data analysis strengths depend heavily on how you access it.

Gemini in Google Sheets is genuinely useful. If your data is already in a Google Sheet, you can use Gemini's built-in features to ask questions about the data directly within the spreadsheet. "What's the average order value by region?" generates an answer right there. It can also suggest formulas, create pivot tables, and build charts within Sheets.

Gemini as a standalone chatbot is less impressive for data analysis. You can upload files, but the analysis depth doesn't match ChatGPT's Code Interpreter or Claude's reasoning. The insights I got were accurate but surface-level. More "summary statistics" and less "here's what this means for your business."

Where Gemini wins: If your data lives in Google Sheets and you want quick answers without exporting anything, Gemini is the most frictionless option. The Sheets integration means zero setup. Just highlight your data, ask a question, get an answer.

Where Gemini loses: For serious analysis of large datasets, the standalone chatbot can't compete with ChatGPT's execution environment. And for nuanced interpretation, Claude is meaningfully better.

Gemini verdict: Best for quick, in-situ analysis within Google Workspace. Not the tool for deep analytical work.

The prompts that work best for data analysis

Through testing, I've found these prompts consistently produce the best results across all three tools:

For initial exploration: "Upload this CSV and give me: (1) a summary of the dataset structure, (2) the top 5 most actionable insights, and (3) any data quality issues you notice."

For trend analysis: "Show me the monthly trend for [metric]. Is this growth seasonal or structural? What's the underlying trend if you remove seasonality?"

For comparison: "Compare performance across [dimension, e.g., regions, channels, categories]. Which is the strongest performer and which is underperforming relative to its potential?"

For anomaly detection: "Find any anomalies or outliers in this data. Are there any data points that don't fit the general pattern? For each anomaly, suggest whether it's likely an error, a one-off event, or something worth investigating."

For actionable recommendations: "Based on this data, what are the three most impactful changes I could make to improve [metric] next quarter? Be specific and support each recommendation with data."

The honest limitations

AI data analysis is powerful but not magic. Here are the things it genuinely cannot do:

It can't tell you why something happened. AI can identify that sales dropped 20% in March. It cannot tell you that this was because your warehouse flooded and you couldn't fulfil orders for two weeks. Correlation is not causation, and AI has no access to the real-world context behind your numbers. You have to bring that.

It can be wrong about specific numbers. This is especially true for Claude and Gemini, which don't have code execution environments. Even ChatGPT's Code Interpreter occasionally makes errors in its Python code. Always sanity-check important numbers manually.

It can't handle messy data gracefully. If your spreadsheet has merged cells, inconsistent date formats, missing values in key columns, or duplicate entries, the AI will struggle. Clean your data first. At minimum: consistent column headers, no merged cells, consistent date and number formats.

It generates insights, not decisions. "Region X is underperforming" is an insight. "We should increase marketing spend in Region X by 30%" is a decision that requires business context AI doesn't have. Use AI to surface patterns. Use your judgement to decide what to do about them.

The recommended workflow Clean your data in a spreadsheet first. Consistent formats, no blanks in key fields, clear column headers. Upload to ChatGPT for the initial analysis, charts, and specific calculations. Paste key findings into Claude and ask "given these findings, what questions should I be asking?" and "what business actions would you recommend?" Use Gemini in Google Sheets for ongoing, quick questions about data you work with regularly.

This three-tool approach gives you the best of each: ChatGPT's computational power, Claude's analytical depth, and Gemini's convenience. Total cost: about $56/month if you're on paid tiers, or free if you use the free tiers and accept the usage limits.

That's dramatically cheaper than a data analyst, and for the 80% of analysis tasks that involve "help me understand this spreadsheet," it's genuinely just as good.