Analyze text and data
On this page
Use ChatGPT as an on-demand analyst
A contract to dissect, survey results to interpret, a spreadsheet no one has time to mine: this is where ChatGPT delivers its most dramatic value. In seconds, it produces a structured analysis that would take you an hour to write. But you must ask the right question. “Analyze this document” is a command with no destination—you will get a polite overview that is rigorously useless.
Analyze text
Targeted extraction is the simplest and most profitable use case. You provide a long text and specify exactly what you want pulled out:
“Here is the minutes from the executive committee meeting. Extract:
- Decisions made
- Assigned actions (with owner and deadline)
- Items deferred to the next meeting”
Sentiment and tone analysis applies to anything coming from your customers or teams. It delivers real value when you ask for qualification and prioritization at the same time, turning a pile of comments into an action plan:
“Here are 10 customer reviews of our product. For each, indicate: sentiment (positive/neutral/negative), the main subject of the complaint or compliment, and an action priority (high/medium/low). Present the result as a table.”
Document comparison works on the same principle. Submit two versions of the same text and ask not just for a list of differences, but for their interpretation: “Here are two versions of our return policy. Identify the major differences and state which one is more favorable to the customer.” Finally, critical analysis involves assigning ChatGPT an adversarial role, which yields feedback far more useful than a complacent opinion:
“Here is my business plan. Act as a skeptical investor. Identify the 5 most significant weaknesses in this plan and propose an improvement for each.”
Analyze structured data
You can paste a table directly into the conversation, whether it comes from Excel, Google Sheets, or simple aligned text. The mandatory condition is to state what the numbers represent: a column of figures without a legend allows no serious analysis, and ChatGPT will fill the blanks with assumptions.
“Here are my monthly sales over 12 months: Jan: 45,000 | Feb: 38,000 | Mar: 52,000 | Apr: 41,000 | May: 63,000 | Jun: 58,000 Jul: 35,000 | Aug: 28,000 | Sep: 55,000 | Oct: 62,000 | Nov: 71,000 | Dec: 85,000
Analyze this data: identify the general trend, exceptional months (good and bad), seasonality, and give me a reasonable projection for next January.”
For larger datasets, do not paste: attach the file using the + button. ChatGPT reads Excel spreadsheets (.xlsx, .xls), CSV, and TSV files, then switches to data analysis: it writes and executes Python code in an isolated environment to calculate, group, filter, or plot a chart. Bar, line, pie, and scatter plots can even become interactive. Prepare the file as you would for a colleague: explicit column headers on the first line, one record per line, a single table per sheet. Then ask precisely what you want: “Calculate the mean, median, standard deviation, and annual growth rate of these sales figures, then plot their month-by-month evolution.”
A serious caveat applies here. A calculation done “in head” within the response text can be wrong while being presented with confidence. For any figure that drives a decision, explicitly request that the calculation go through data analysis, review the displayed code and assumptions, and cross-check the result with your spreadsheet. Also beware of scanned or photographed tables: OpenAI warns that exact values are not reliably extracted from them; prefer a spreadsheet or text file. A wrong number in a report costs infinitely more than the thirty seconds of verification you saved.
Analyze long documents
For a contract, report, or academic paper, do not paste the text into the conversation: attach the document (PDF, DOCX, or TXT) using the + button. On the Free plan, count on three files per day. ChatGPT can then summarize it, answer targeted questions about its content, extract a specific clause or section, and compare it to other documents you provide. Lesson 14 will dive deeper into this mechanics.
Again, the precision of your request determines the value of the result:
“I have attached a service contract. Identify: the provider’s obligations, late penalties, termination conditions, and anything that could be problematic for us as the client.”
Your turn
Choose your exercise. Three paths, depending on what looks most like your daily work. If your material is textual, take a recent email or work document and have it extract the actions to take, ranked by priority. If you work with numbers, take your last six months of data—sales, KPIs, budget—and ask for a trend analysis with key points of attention. If you prefer to test the mirror function, paste a document you wrote yourself and ask for an evaluation listing three strengths and three weaknesses. In all three cases, verify at least one number or quote from the response against the source.
Official sources: Data analysis with ChatGPT (OpenAI Help Center: formats, data preparation, Python, charts, limits) · Uploading files and audio to ChatGPT (import formats and limits) · Does ChatGPT tell the truth? (calculations and verification)
Key takeaways
- ChatGPT can analyze text (extraction, sentiment, comparison, critique) and data (trends, calculations, statistics)
- Always provide context with your data: what they represent, what you are looking for
- For a number that matters, have data analysis (Python) do the calculation and cross-check in your spreadsheet
- Attach long documents and spreadsheets with the + button rather than pasting them
- Structure your analysis requests with specific questions rather than a vague “analyze this”