Having AI Analyze Your Data: How to Write Prompts That Pull Insights from Uploaded Excel and CSV Files

The Era of AI Data Analysis: Why Learn It Now

In the past, analyzing a single set of sales data meant building an Excel pivot table, writing VLOOKUP formulas, and drawing charts by hand. It took experts 2–3 hours and beginners half a day or more.

Now you can upload a file to ChatGPT or Claude and get the same result in 5 minutes with a single line of prompt.

Follow this guide to the end, and you can handle the entire process on your own: upload an Excel or CSV file to AI → request an analysis → turn the insights into a report.

Before you start: check your files and tools

Supported tools and plans

• ChatGPT: File upload and Advanced Data Analysis (formerly Code Interpreter) are available on the Plus plan ($20/month) or higher

• Claude: File upload is available on the Pro plan ($20/month) or higher. The free plan also allows limited file attachments

💡 Pro tip: On a free plan, you can paste your data into the prompt box as text. Once you pass 100 rows, though, uploading a file is far more accurate.

Prepare the file format

These are the file formats AI reads best:

• CSV (.csv) — Most recommended. Fewer encoding problems and faster processing

• Excel (.xlsx) — If the workbook has multiple sheets, specify which sheet to analyze

• TSV (.tsv) — Tab-delimited text files are also recognized

File cleanup checklist before uploading

Before you upload, be sure to check the items below:

• Is the first row a column header? (e.g., Date, Product name, Revenue, Customer name)

• Are there any merged cells? → Unmerge them, then save

• Are there any blank rows or columns in the middle of the data? → Delete them

• Does the Korean text encoding display correctly? → Choose UTF-8 when saving as CSV

• Does the file contain any sensitive information (national ID numbers, card numbers, etc.)? → Delete or mask it

💡 Pro tip: When saving from Excel to CSV, choose **File → Save As → File format: "CSV UTF-8 (Comma delimited) (*.csv)"**. If you choose plain CSV, Korean text may appear garbled.

Step 1: Upload your file

How to upload in ChatGPT

1. Log in at chat.openai.com

2. Start a new conversation (click "New chat" at the top left)

3. Click the 📎 paperclip icon (or the + button) to the left of the message box at the bottom

4. Select **"Upload from computer"**

5. Choose your prepared CSV or Excel file and click Open

6. When the file name appears above the message box, the upload is complete

7. Now type your prompt and press Enter

How to upload in Claude

1. Log in at claude.ai

2. Start a new conversation

3. Click the 📎 paperclip icon in the message box at the bottom

4. Select a file, or drag and drop it into the input box

5. Once you see the file attached, enter your prompt

💡 Pro tip: Uploads may fail if the file is larger than 10MB. For large files, extract only the columns and rows you need, save them as a separate file, and upload that. For example, if you only need the most recent 6 months out of 3 years of data, trim the file to that period.

Step 2: Ask the AI to Understand the Data Structure (First Prompt)

Right after uploading the file, don't ask for the analysis yet. The key is to have the AI understand the structure of your data first. If you skip this step, the AI may misread columns or produce off-target results.

Prompt Example 1: Understanding the Data Structure

Please analyze the attached CSV file. First, summarize the following:

1. Total number of rows and columns

2. Each column’s name, data type (number/text/date), and share of missing values

3. A sample of the first 5 rows

4. The full date range of the data (if there is a date column)

5. Basic statistics for numeric columns (mean, median, min, max)

Please present the results in a clean table.

Example of the AI’s Response

The AI returns a table like this:

💡 Pro tip: If any column has 10% or more missing values, ask the AI a follow-up: *"How should I handle the missing values?"* It can suggest whether to fill them with 0, replace them with the average, or exclude those rows.

Step 3: Request the Core Analysis (Second Prompt)

Once you’ve confirmed the data structure, it’s time for the real analysis. Here, the quality of your prompt determines the quality of the result.

Bad Prompt vs. Good Prompt

Prompt Example 2: In-Depth B2B Sales Data Analysis

Please analyze the attached sales data from the following perspectives.

Analysis requests:

1. Monthly revenue trend: A table with month-over-month change (%) and a line chart

2. Revenue ranking by customer: Revenue, number of transactions, and average deal size for the top 10 customers

3. Share by product category: Visualize as a pie chart

4. Outlier detection: A list of transactions more than 3 times above or below the average

5. Seasonal patterns: Analyze whether revenue is concentrated in particular months

Output format:

• For each item, include both the key Finding and the business Implication

• Format tables as Markdown tables

• Use thousands separators for numbers, and show amounts in Korean won (₩)

Role: You are a B2B sales data analyst with 10 years of experience. Write in the tone of a report to the head of sales.

Prompt Example 3: Customer List Segmentation Analysis

Analyze the attached customer list data and perform customer segmentation.

Analysis criteria:

• Purchase frequency: Classify as high/medium/low based on the number of transactions in the last 12 months

• Purchase amount: By total revenue, top 20% (VIP), middle 60% (regular), bottom 20% (likely dormant)

• Most recent transaction: By days since the last transaction (within 30 days / 31–90 days / 91+ days)

Deliverables:

1. RFM analysis matrix (Recency, Frequency, Monetary)

2. Number and share of customers in each segment

3. List of customers at risk of churning (no transactions in the last 90+ days, with a past history of at least one transaction per month)

4. Recommended sales strategy for each segment (2–3 lines each)

Summarize the results in a table, then also create a one-page summary that can be used directly in a sales team meeting.

💡 Pro tip: Assigning a **"Role"** in your prompt makes a big difference in the quality of the results. Instead of *"You are a data analyst,"* be specific, such as *"You are a senior analyst who has handled revenue analysis at a B2B SaaS company for 5 years."*

Step 4: Request Visualizations

Creating Charts in ChatGPT

ChatGPT's Advanced Data Analysis feature automatically runs Python code to generate charts directly.

1. Continuing from the earlier analysis conversation, enter the following:

Based on the analysis above, please create the following charts:

1. Monthly revenue trend line chart (Y-axis: revenue, X-axis: month, with a same-period prior-year comparison line)

2. Horizontal bar chart of revenue for the top 10 customers (sorted in descending order)

3. Pie chart of revenue share by product category

Chart style: white background, a clean sans-serif font, titles and axis labels in English, and a unified blue color palette suited to business presentations.

2. ChatGPT runs the Python code and outputs the charts as image files

3. Click a generated chart image to download it

Visualization Alternatives in Claude

Claude does not currently generate chart images directly, but it offers the following alternatives:

• Text-based visualization: Visual summaries using ASCII charts or Markdown tables

• Code provided: Claude generates Python (matplotlib, plotly) or Google Sheets formula code for you to run yourself

• Artifacts feature: Claude’s Artifacts can generate interactive charts built on HTML/JavaScript

💡 Pro tip: If a chart isn’t quite right, make detailed edit requests such as "Tilt the X-axis labels 45 degrees", "Move the legend below the chart", or *"Change the colors to our company brand colors (#2E5BFF, #00C9A7)"*. The AI remembers the earlier context, so you can keep refining.

Step 5: Summarize Insights and Generate a Report

Once the analysis and visualization are done, finish by asking for everything in a form you can use for business decisions.

Prompt Example 4: Executive Summary

Based on the analysis so far, write a one-page executive summary report.

Include:

• 3 key findings (each with a one-line summary and supporting figures)

• 2 risk factors that need attention

• 3 action items that can be executed immediately (name the responsible department)

• Next month’s sales forecast (based on the current trend)

Format: Mostly bullet points, with numbers and percentages emphasized and jargon kept to a minimum

Length: One A4 page or less

Tone: Concise, confident business-report style

Before/After Comparison: Report Quality

Advanced Use: Automating Repetitive Tasks

Save Your Prompt Templates

If you run the same data analysis every week or month, save your prompts as templates.

Recommended approach:

• Keep prompt templates in a notes app or Notion

• Reuse them by changing only the file name and date

• Save your default analysis framework in ChatGPT’s "Custom Instructions" or Claude’s "Project" feature

Multi-File Comparative Analysis

Please compare and analyze the two attached files.

• File 1: Sales data for the first half of 2024 (January–June)

• File 2: Sales data for the second half of 2024 (July–December)

Compare:

1. Change in total revenue, first half vs. second half

2. Change in purchase frequency by customer (identify churned and new customers)

3. Growth and decline by product category

4. Trends that were absent in the first half but emerged in the second half

Summarize the results in a comparison table, and give me 3 recommendations for the first-half 2025 sales strategy.

💡 Pro tip: To upload two files at once in ChatGPT, click the paperclip icon and add the files one at a time. In Claude, you can also drag and drop multiple files in one go.

Common Problems and How to Fix Them

Problem 1: "Unable to read the file"

• Cause: An encoding problem or a corrupted file

• Fix: Open the file in Excel, then Save As → CSV UTF-8 to re-save it

Problem 2: Korean column names appear garbled

• Cause: The file was saved with EUC-KR encoding

• Fix: Open the file in Notepad, then choose Save As → Encoding: UTF-8

Problem 3: The AI analyzes the wrong columns

• Cause: Column names are ambiguous or use abbreviations

• Fix: Name the columns explicitly in your prompt. Example: “The ‘amt’ column is revenue and the ‘qty’ column is quantity.”

Problem 4: Results are too generic and superficial

• Cause: The prompt is too abstract

• Fix: Specify concrete numbers, comparison criteria, and the output format. Instead of “Analyze this,” try “Calculate the month-over-month change and highlight in red any category that fell by 10% or more.”

Security Notes: What You Must Do

Before uploading data to an AI tool, always check the following:

• ⚠️ No personal information: Delete or mask personally identifiable information such as customers’ national ID numbers, phone numbers, and email addresses before uploading

• ⚠️ Check for confidential company data: Confirm that your company’s information security policy allows uploading data to external AI services

• ⚠️ API vs. web interface: The ChatGPT API does not use your data for training, but policies for the free web version may differ. Set Settings → Data Controls → “Improve the model for everyone” to OFF

• ⚠️ Test with dummy data first: Before uploading real data, we recommend validating your prompt with fictitious data that has the same structure

💡 Pro tip: Replace customer names with “Customer A, Customer B, Customer C,” or alter the original figures by multiplying revenue amounts by a constant factor before uploading. This protects your data without hurting analysis accuracy. The AI analyzes patterns and ratios, so as long as the proportions hold, the quality of the insights is unaffected.

Practical Checklist: Finish Your Data Analysis in 5 Minutes

Follow these steps every time you analyze data to get results quickly and without mistakes:

• Step 1 (30 sec): Prepare the file — check headers, unmerge cells, save as UTF-8

• Step 2 (30 sec): Upload the file — click the paperclip icon and select the file

• Step 3 (1 min): Enter a structure-check prompt — row count, column count, data types, missing values

• Step 4 (2 min): Enter the core analysis prompt — specify the analytical angle, output format, and role

• Step 5 (1 min): Request visualizations and a summary — generate charts and a one-page report summary

• Wrap-up: Download and verify the results — cross-check the numbers, then add them to your report

Key Takeaways: 5 Principles of a Good Prompt

Keep these five principles in mind and apply them to every prompt, and the AI will work not as a simple calculator but as your own dedicated data analyst. Today, upload one Excel file you already have, copy the prompt template above, and get started. The moment you see the first result, the way you think about data analysis will change completely.