AI Data Analysis Without a Data Scientist

For AI data analysis, a small business no longer needs a data scientist: you upload a CSV or Excel file to ChatGPT, Claude or Gemini, ask your question in plain language and get summaries, tables and charts back. For many analyses, ChatGPT and Claude write and run Python code in the background, according to their help pages. That saves you the programming, but not the checking. The AI can pick the wrong method, skip rows or copy a figure incorrectly. So for every figure you pass on, you recalculate at least one point yourself. And you only upload customer data if you know what happens to it.

I work with AI tools every day, and my fourteen AI employees run on Claude Team. One of them, Sebastian, my AI employee for SEO and GEO, analyzes data for me: search queries, clicks and impressions from Google Search Console and Bing Webmaster Tools. I show how he does it further down as an example. I checked the tools' features on the vendors' help pages on October 29, 2026; I did not run my own head-to-head test for this article. All the basics about AI are collected on my AI Basics page.

What AI data analysis does for a small business

Instead of pivot tables, formulas or hiring someone who knows Python, you now describe what you want to know and the AI does the number crunching. OpenAI lists exactly these tasks for ChatGPT: summarizing rows, columns, trends and outliers, creating tables for review, building charts from uploaded data, running Python-based calculations and statistical analysis, and explaining assumptions and results.

For a small business, that could mean: which services brought in the most revenue last quarter, which customers haven't ordered for a year, or how material costs per job have developed. You don't need to know a single formula for this. For formulas and the AI tools inside the spreadsheet, read AI for Excel: Formulas, Analysis, Limits; this article covers data in general.

Which AI analyzes data: what the vendors document

The table summarizes the help pages as of October 29, 2026. It does not say which tool calculates better.

ChatGPT

OpenAI recommends clean data: descriptive headers in the first row, one record per row, one table per sheet, no empty rows or columns splitting the data. According to OpenAI, the environment in which ChatGPT runs Python code for this cannot access the internet, so you upload any external data yourself. OpenAI also admits that on the first try ChatGPT may choose a method or chart type that does not match your intent. Then you name the exact column, grouping or calculation.

Claude

In Claude the feature is called code execution and file creation. According to Anthropic it is available on all plans, including the free one, on the web, in the desktop app and on mobile. Claude can write Python scripts for analysis, process CSV and TSV files and create charts as images. On top of that come interactive charts right in the chat, still in beta according to Anthropic and only on web and desktop. Upload a CSV and ask what the data shows, and Claude builds such a chart.

Anthropic names a risk itself: with network access turned on, hidden instructions in files could trick Claude into sending data outside. Its advice is to watch Claude and stop it if something unexpected happens.

Gemini

According to Google, you can upload up to 10 files per prompt to Gemini, most file types up to 100 MB. In the web app you can turn an uploaded spreadsheet into a chart and then change its type and labels. Google warns that with files that are too large, the answer may miss connections or details.

How to analyze data with AI: a six-step workflow

I recommend this workflow regardless of the tool.

1. Make a copy and trim it. Export only the columns you need for the question. Leave out names, email addresses and phone numbers, or replace them with customer numbers.
2. Clean up the data. One table per file, column names in the first row, one row per transaction. That is the preparation OpenAI explicitly recommends.
3. Have it describe the file first. Start with: "Describe this file: how many rows, which columns, what time period, are any values missing?" If the row count does not match your source, the AI has not read everything.
4. Ask one concrete question. Not "analyze this", but for example: "Sum of net amount per service and month for July to September, sorted descending, as a table and as a bar chart."
5. Have it show its working. Ask the AI to calculate with code and explain the steps. OpenAI advises reviewing code, outputs and assumptions for Python-based analysis.
6. Recalculate. Check at least one total by hand or in your own spreadsheet. If it matches, that proves nothing about the rest, but a gross error will stand out.

An example, made up: a painting business exports a year of invoices as a CSV, without customer names, and asks: "Which three job types bring in the most revenue per working day?" The owner then checks one job type against her own bookkeeping.

An example from my own work: how Sebastian analyzes search data

Sebastian is himself an AI that analyzes data without a data scientist. He reads search queries, clicks, impressions and indexing status from Search Console and Bing Webmaster Tools, read-only. How to set up Search Console for this is covered in Setting Up Google Search Console.

Three rules make his analyses reliable, and they apply to any AI data analysis:

• Every figure has a source. For Sebastian, every figure comes from a saved tool result in his project folder, not from memory.
• Measure before and after the same way. He measures before a change and afterwards with the same setup. One of his early analyses showed: technology done, the bottleneck is content and internal linking. What the measures achieved afterwards is documented, with sources, on his team page.
• Check separately. In my team, whoever created something does not check it. On September 18, 2026, an outside cross-check found factual errors that Sebastian's own proofreading had missed. Since then it has been part of every large delivery.

Why you should put several sources side by side: Microsoft Copilot cited kevinwelter.com 183 times in the three months up to September 16, 2026, and the most cited page was not the one with the most Google clicks.

AI reporting: turning one analysis into a regular report

Once an analysis works, you will want it every month. That is AI reporting on a small scale: same question, same columns, new data. Save your prompt with an exact description of the columns and the calculation rule. According to Anthropic, Claude can output the report straight away as a Word or PDF file with charts, or as an Excel spreadsheet.

If an export changes unnoticed, say a renamed column or a duplicated period, the AI keeps calculating anyway. So step 3 above is mandatory for every report.

The limits of AI data analysis

Wrong figures. A language model can copy figures incorrectly or make them up when it answers without code. Calculating with code lowers that risk, but the method can still be wrong: wrong filter, duplicate rows, wrong column. How to spot such errors is covered in AI Hallucinations: Why AI Makes Things Up.

Incomplete reading. Google writes that details may be missed in large files. So check the row count and preferably split large files.

Customer data. Whatever you upload is processed by the vendor under the rules of your plan. Trim data to what is necessary and replace names with numbers. What you can show an AI is explained in AI and Privacy: What the AI Gets to See.

Hidden instructions. Anthropic warns about prompt injection, meaning instructions hidden inside files. Don't analyze files of unknown origin with network access turned on.

Frequently asked questions

Which AI is best for data analysis?

There is no universal ranking. According to the vendors, ChatGPT and Claude calculate with their own Python code, Gemini and ChatGPT create charts from uploaded spreadsheets, and Claude also creates interactive charts in the chat. Use the tool your business has already approved and test it on a file whose result you know.

Can I upload Excel files directly?

Yes. OpenAI explicitly lists .xls, .xlsx and .csv, Anthropic lists CSV and TSV and also creates Excel files, and Google speaks of spreadsheets in general. A simple table with one header row and no merged cells works most reliably.

Is AI data analysis free?

Partly. According to Anthropic, Claude offers code execution and file creation on all plans, including the free one. Gemini has rolling limits on uploads according to Google, and with ChatGPT the available file types and limits depend on model, plan and workspace according to OpenAI.

Can I upload customer data to ChatGPT or Claude?

This isn't legal advice, just the operator's perspective: it depends on your plan, your contracts and your data. The safest option is to use customer numbers instead of names and upload no personal data at all.

Where to go next

Take a spreadsheet whose result you already know, for example last month's revenue. Upload a trimmed copy, ask for that same figure and compare. If it matches, ask the next question you have never been able to answer. How an AI employee checked search data against a paid tool is shown in my SE Ranking review. How others analyze their data with AI is something we discuss in my community AI Employee Framework.