The short answer.

Your AI assistant can analyse your files and data: ChatGPT, Claude, Copilot and Gemini each read spreadsheets, PDFs and exports, run real calculations by writing and running code, draw charts and, in most cases, hand back a spreadsheet or document you can download. Each maker does it in its own way and in its own place, and the trade-offs are the same everywhere: large or messy files lose detail, the assistant chooses a method you need to check, and some data does not belong in the assistant at all. Used with those three things in mind, file analysis is one of the most useful features most people have never opened.

What changes when you attach a file.

When you attach a file, the assistant stops guessing at arithmetic and starts computing it. In a plain chat, an assistant writes numbers the way it writes sentences, and it can get a sum wrong. With a spreadsheet attached and file analysis on, the assistant writes a short program, runs it on your data and reports what the program returned. That is the difference between basic Excel and an advanced Excel user: the same tool, used for work it was built to do.

Three things become possible that most people never try: asking plain questions of an export (which customers ordered less this quarter than last), getting a chart without building it, and receiving a cleaned or reshaped file back instead of a paragraph about one.

What each maker offers, under its own name.

Each of the four makers offers file analysis, under different names and in different places. Makers rename and move these features often; what follows was read on their own help pages in November 2026.

  • ChatGPT.OpenAI’s help centre describes data analysis with ChatGPT: you upload spreadsheets, PDFs or text files, and ChatGPT runs calculations in code, builds tables and draws charts, some of which you can make interactive. For producing finished files, ChatGPT Work creates and edits Word documents, Excel workbooks and Google Docs and Sheets from source material you attach. Editing an Excel workbook you already have open runs through Codex, OpenAI’s coding tool, in the desktop app, which is less of an everyday route.
  • Claude.Anthropic’s feature is called code execution and file creation. Claude runs code on what you upload and produces Excel workbooks with working formulas, PowerPoint decks, Word documents and PDFs, plus charts as images, all of which you download. Anthropic’s own page warns that a file or web page from outside can carry hidden instructions that try to steer Claude, and advises watching what it does.
  • Copilot.Copilot in Excel answers plain questions about the tables in a workbook, and its advanced analysis mode writes and runs Python on a new sheet, with the code there to expand and read. Microsoft also offers Analyst, an agent in Microsoft 365 Copilot that takes Excel and CSV files and returns a written report with charts and tables. Your administrator may need to switch Analyst on. Copilot’s strength is that the work stays inside the Excel files your company already keeps.
  • Gemini.Google’s help page is titled “Attach & analyze files in Gemini Apps”: you attach documents and spreadsheets, ask questions, and ask for a chart you can then adjust. Gemini in Google Sheets does the same work inside a sheet, and shows its analysis steps so you can follow the method. Gemini’s strength is the same as Copilot’s, for companies that live in Google Sheets.

The real choice is not which maker does it best. It is whether your people work in a separate chat with a file attached, or inside the spreadsheet they already use. ChatGPT and Claude are strongest at producing a new file from several sources. Copilot and Gemini are strongest when the data already sits in Excel or Google Sheets.

What file analysis does well.

File analysis does well at the questions a manager would ask an analyst, if the company had one.

  • First reads of an export.What changed since last month, which rows look unusual, where the gaps are.
  • Tidying.Merging two exports, splitting a name column, standardising dates, removing duplicates, and returning the cleaned file.
  • Charts for a meeting.A clear chart of the trend you care about, without building it by hand.
  • Reading PDFs into tables.Pulling the figures out of supplier statements or reports that arrive as PDFs, when the PDF holds real text.

The trade-offs to know before you rely on it.

File analysis has four trade-offs, and every maker’s own help pages name some of them.

  • Size and shape.Every product has limits on how large a file can be and how much it reads closely. Google’s page says plainly that a large file can make Gemini miss details spread through it; OpenAI’s says large or formula-heavy files may not be fully analysed. Send one sheet or one month, not the whole workbook.
  • Messy inputs.Scanned PDFs, photos of documents and sheets with merged cells and notes in the margins read poorly. Copilot in Excel does not handle unstructured data at all. A clean export from the system beats a formatted report.
  • The method.The arithmetic is done in code and is reliable; the choice of what to calculate is not. The assistant may pick the wrong column, drop rows it judged empty, or choose a chart you did not mean. Every maker tells you to review the method, and the code or steps are there to read.
  • What should not go in.Payroll, health information, client records under a confidentiality agreement and anything your contracts restrict need a decision before anyone uploads them. Is it safe to put company data into ChatGPT, Claude, Copilot or Gemini? sets out what decides it.

Two smaller trade-offs catch people out. Charts made this way are usually pictures, so they do not update when the data changes. And in Gemini in Google Sheets, the conversation is lost if you reload the page, so insert what you want to keep.

Questions to ask before you trust the numbers.

Ask these before an analysis from an AI assistant reaches a decision, a client or a board pack.

  1. Is this data allowed in this account?Work account or personal one, and does the information fall under a rule or a contract.
  2. Did it read the whole file?Ask the assistant how many rows it used and compare with the file.
  3. What did it calculate, exactly?Ask for the method in plain words, or open the code or the steps.
  4. Does one figure check out by hand?Pick one total and confirm it in the original spreadsheet.
  5. Which rows did it leave out, and why?Blanks, duplicates and outliers are where answers quietly change.
  6. Who signs it off?One person, by name, who reads the result before it is used.

How do we keep quality up when AI prepares the work? covers checking more broadly.

A landscaping company reads its own job data.

Take a 45-person landscaping company. The operations manager exports last season’s jobs from the scheduling system each month and builds a summary by hand.

Illustrative example.

She attaches the export to the company’s assistant and asks which crews ran over their estimated hours, by job type, with a chart for the Monday meeting. The assistant returns a table, a chart and a cleaned workbook. She asks how many jobs it counted, finds it dropped the jobs with no estimate, and asks it to list those separately. She checks one crew’s total against the export, then takes the chart to the meeting. The same request, written once and saved, becomes something she runs every month.

Where FullyPowered AI comes in.

This is what we help companies do: level up what they get from the AI they already pay for, by planning which work goes into it, developing the workflows and deploying them to the people who run them. FullyPowered Planner is where that plan lives, and How organizations plan their use of AI sets out the method.