How to use AI for spreadsheets and data analysis

What AI can do with spreadsheets

AI can turn a shopping list into a table, find duplicate orders, explain an Excel formula, or compare expenses across two months.

You don't need to learn complicated functions to get started. Describe your task in plain language and provide the data in a clear format.

ChatGPT, Claude, Gemini, DeepSeek, and other AI chatbots can work with spreadsheet data. AI assistants are also available inside Excel and Google Sheets. Capabilities vary by tool: some explain data in text, while others can perform calculations and modify files.

Task	Can AI help?	What to verify
Create a spreadsheet template	Yes	Required columns
Fill a table from text	Yes	All records are included
Standardize date and number formats	Yes	Original values
Find duplicates and missing values	Yes	Matching criteria
Group data	Yes	Group totals
Compare time periods	Yes	Change calculation
Write an Excel formula	Yes	Result in the cell
Choose a chart	Yes	Whether it represents the data accurately
Calculate results from a large dataset	Use caution	Independent calculation
Build a statistical forecast	Use caution	Method and assumptions

AI might correctly identify which metrics you need but still make an arithmetic error.

A better approach is to separate the tasks. Use AI to define the question, prepare the data, and explain the result. Use Excel or Google Sheets for calculations you can independently verify.

This is especially useful for sales data, budgets, schedules, order lists, and reports.

How to choose your workflow

There are two main approaches.

The first is to use a regular AI chat alongside Excel. Share a table or part of one, then ask for a formula, instructions, or a written summary. Verify the output before transferring it to your spreadsheet.

The second is to use an AI assistant built into a spreadsheet application. Depending on its capabilities, it may work directly with cells and ranges.

The first approach is enough for the exercises in this lesson. It lets you understand each step and doesn't depend on a particular version of Excel.

How to share spreadsheet data with AI

Before analyzing a spreadsheet, you need to preserve its structure: headers, rows, values, and units of measurement.

There are four common ways to share the data.

Paste a table as text

Select the cells in Excel or Google Sheets, copy them, and paste them into the chat.

When you copy cells normally, columns are often separated by tabs. If the chat doesn't recognize the column boundaries correctly, use CSV instead.

Example data:

Date	Category	Amount, $
2026-10-01	Groceries	12.50
2026-10-02	Transportation	3.20
2026-10-03	Groceries	8.40

Explain what each row represents.

I'm sharing a table of personal expenses. Each row represents one purchase. Amounts are in US dollars, and dates use the YYYY-MM-DD format. First, check the number of rows and the column names. Do not perform any calculations until you have confirmed the table structure.

This prompt follows the approach from the lesson “How to write AI prompts”: define the task, provide context, set constraints, and specify the output format.

Attach a CSV file

CSV stores spreadsheet data as plain text. Values are separated by commas, semicolons, or another delimiter.

For example:

Date;Category;Amount
2026-10-01;Groceries;12.50
2026-10-02;Transportation;3.20
2026-10-03;Groceries;8.40

To create a CSV file in Excel, go to “File,” select “Save As,” and choose a CSV format. In Google Sheets, go to “File,” then “Download,” and select CSV.

Before uploading, open the exported file and inspect the first few rows. Pay particular attention to special characters, decimal separators, and column delimiters.

CSV doesn't preserve formatting, charts, or the structure of a multi-sheet Excel workbook. An export typically contains one sheet with its current values.

If the original workbook contains multiple sheets, prepare the data you need separately.

Upload a screenshot or PDF

Sometimes the table is in a PDF report, an image, or a photo of a document.

An AI model with image support can recognize visible cells. If a PDF contains extractable text, that text can also be used as a data source.

Prompt:

Extract the table from the attached document. Preserve the headers, row order, and numeric values. Show the result in Markdown. Mark unreadable cells as “needs review”. Do not infer missing numbers from context.

First, compare the extracted table with the original.

Check decimal places, negative numbers, total rows, and column boundaries. Small text in a screenshot may be misread.

Text-based PDFs and scanned PDFs are different. If a file consists only of images of pages, text extraction may fail. In that case, you'll need image recognition through a supported AI model or a separate tool.

Explain what the columns mean

Suppose your spreadsheet has columns labeled “Date,” “Amount,” “Expenses,” and “ACoS”.

AI doesn't know exactly which expenses are included or what an abbreviation means in your report.

Provide a definition. For example: “ACoS means advertising cost of sales, calculated as advertising spend divided by attributed sales revenue, expressed as a percentage.”

If your table includes refunds, discounts, or empty cells, explain how they should be handled.

Without this information, the model could perform a mathematically correct calculation using the wrong interpretation of a metric.

Consider file size and data security

In Mia chat, you can attach CSV, TXT, and PDFs with extractable text, up to 60,000 characters per file. A single text message is limited to 10,000 characters. Models with image support can read images.

Mia doesn't accept XLSX files. Save the sheet you need as CSV.

If the data is too large, split it into parts with identical headers. Number the parts and first verify that all of them have been read.

Remember that separate calculations for individual parts don't replace a complete calculation across the entire dataset.

How to create a spreadsheet from scratch

AI can suggest a spreadsheet structure even when you don't have any data yet.

Suppose you want to track your personal budget. Start by deciding what information you need to record.

Prompt:

Create a template for tracking personal expenses over one month. Include columns for date, category, description, and amount in US dollars. Show three fictional rows as examples. Do not invent any additional expenses. Output the table in Markdown.

Result:

Date	Category	Description	Amount, $
2026-10-01	Groceries	Grocery shopping	12.50
2026-10-02	Transportation	Transit fare	3.20
2026-10-03	Phone	Mobile phone service	5

This is a template. Delete the sample records before entering your real data.

A project plan needs different columns: task, assignee, due date, and status.

For tracking orders, useful columns include date, order ID, amount, and payment status.

If you're unsure which fields you need, ask AI to suggest them and explain their purpose. Then remove any unnecessary columns.

How to turn notes into a table

Original text:

“On Monday, I spent $12.50 on groceries. On Tuesday, transportation cost $3.20. On Wednesday, I paid $5 for my phone service.”

Prompt:

Turn this note into a table. Include columns for day of the week, category, and amount in US dollars. Preserve all values. Do not add calendar dates because the specific week isn't given.

Day	Category	Amount, $
Monday	Groceries	12.50
Tuesday	Transportation	3.20
Wednesday	Phone	5

AI extracted the information from the sentences and organized it into columns.

You can use the same approach to structure notes, task lists, and requests. If the source text doesn't contain a particular value, instruct AI to leave that cell empty.

How to transfer the result to Excel

You can copy a Markdown table from an AI response. However, copying and pasting doesn't always place each value in the correct cell.

If all the data ends up in one column, ask AI to return it as tab-separated text or CSV.

Output the same table as CSV using “;” as the delimiter. Preserve the headers and values. Do not add anything.

Then import the text data into Excel or Google Sheets. Choose the correct delimiter during import.

After importing, check that amounts are recognized as numbers. If a value is stored as text, formulas may produce incorrect results.

How to clean spreadsheet data before analysis

Before running calculations, standardize inconsistent dates, category names, and number formats.

Below are six fictional orders. The original data uses different date formats, capitalization, and number formats.

Date        Category          Amount
01.10.26    Books             12
2026-10-02  books             8,51
03.10.2026  BOOKS             7
04.10.26    Office supp.      4.50
2026-10-05  Office supplies   3,00
06.10.2026  books             15

Date        Category          Amount
2026-10-01  Books             12.00
2026-10-02  Books              8.51
2026-10-03  Books              7.00
2026-10-04  Office supplies    4.50
2026-10-05  Office supplies    3.00
2026-10-06  Books             15.00

The cleaned table still contains all six rows. Dates and amounts now use consistent formats. The abbreviation “Office supp.” was replaced with “Office supplies” according to a predefined rule.

Prompt:

Standardize this table. Dates: YYYY-MM-DD. Categories: “Books” and “Office supplies”. Amounts: US dollars with two decimal places. Preserve every row, the original row order, and all numeric values. List the changes separately.

Don't ask AI to correct numeric values at the same time unless you know which values are wrong.

For example, an amount of 1.50 might look suspicious next to 15, but you shouldn't change it without checking the original source.

How to find duplicates and missing values

Matching amounts don't necessarily mean two records are duplicates.

Two orders worth $15 could belong to different customers. Unique order IDs are more reliable for detecting duplicates.

Prompt:

Find records with identical order IDs. Show the matching IDs and their rows. Do not delete anything. Separately list any empty cells in the “Amount” column.

If there are no unique order IDs, define which combination of fields should be used to identify duplicates before running the check.

Missing values also need to be handled separately.

An empty amount might mean payment hasn't been made, the amount is still unknown, or the value was missing from the export. AI shouldn't automatically replace it with zero.

How to find unusual values

Suppose typical expenses are a few dollars, but one row contains $500.

Ask AI to flag the record without deleting it.

Show all rows where the amount exceeds $100. Include the row number and amount. Do not correct or exclude any matching records.

An unusual value might represent a legitimate large purchase. Verify it against the original source.

You can take a similar approach with text columns. Ask AI to standardize city names, remove extra spaces, or split full names into separate fields. Define the rules first and preserve the original data.

How to analyze data and draw conclusions

Once the data is clean, you can start analyzing it.

Begin with one specific question: What were total sales? Which day had the highest sales? Which category grew? Where did expenses exceed revenue?

Let's use fictional weekly sales data from a store.

Day	Sales, $	Expenses, $
Mon	120	70
Tue	90	60
Wed	150	85
Thu	110	72
Fri	180	100
Sat	220	130
Sun	130	80

How to get a numerical summary

First prompt:

Calculate total sales for the week. State how many rows were included and show the arithmetic expression. Use only the “Sales” column.

Correct calculation:

120 + 90 + 150 + 110 + 180 + 220 + 130 = $1,000.

Next, calculate expenses in a separate prompt.

70 + 60 + 85 + 72 + 100 + 130 + 80 = $597.

Difference:

1,000 − 597 = $403.

This is the difference between the reported sales and expenses. You cannot call it net profit if taxes and other costs are unknown.

How to compare metrics

Next question:

Find the days with the highest and lowest sales. Show the amounts and the difference between them. Do not speculate about why they differ.

Result:

• Highest: Saturday, $220.
• Lowest: Tuesday, $90.
• Difference: $130.

Then ask for findings:

Write two findings based on the table. Support each with specific numbers. Do not invent reasons for increases or decreases.

A suitable response:

“Total sales for the week were $1,000. After the listed expenses, $403 remained.”

“Saturday's sales were $130 higher than Tuesday's. The table doesn't explain why.”

How to group rows by category

In a real spreadsheet, the same category might appear dozens of times.

Instead of reviewing every row, ask AI to group the records.

Let's use the cleaned order data from the earlier example:

Category	Number of records	Total, $
Books	4	42.51
Office supplies	2	7.50
Total	6	50.01

Prompt:

Group the cleaned orders by category. For each category, show the number of rows and the total amount. Add a grand total. Show which original amounts were included in each group.

For books, the calculation is:

12 + 8.51 + 7 + 15 = $42.51.

For office supplies:

4.50 + 3 = $7.50.

Grouping helps you identify which category contributed the most to the total.

When to use a pivot table

If you need to repeat the same grouping whenever new data arrives, a pivot table in Excel or Google Sheets is useful.

It lets you group records by category, date, or other attributes and calculate totals without rewriting the original rows.

AI can explain how to set one up.

I have columns labeled “Date”, “Category”, and “Amount”. Write step-by-step instructions for creating a pivot table in Excel. Categories should appear in the rows, and the sum of expenses should appear in the values. Do not modify the original data.

First, verify which columns are selected for grouping and which are used for calculations.

Pivot tables are especially useful when you need to update a report after adding new rows. AI helps you define the structure, while the spreadsheet application performs the calculations.

How to compare two periods

Overall totals don't always tell the full story.

Suppose a fictional store's sales by category changed as follows:

Category	September, $	October, $
Books	200	300
Office supplies	300	250
Gifts	100	150
Total	600	700

Total sales increased by $100.

Percentage increase:

(700 − 600) / 600 × 100% = 16.67%.

But office supplies sales fell by $50 even though overall sales increased.

Prompt:

Compare September and October sales by category. Show the absolute change in US dollars and the percentage change. Use September as the baseline. Highlight categories with negative growth. Do not speculate about the reasons for these changes.

When calculating percentage changes, always specify the baseline period.

If the original value is zero, the standard percentage change formula cannot be used because division by zero is undefined.

Also, distinguish percentages from percentage points. If a share increases from 10% to 15%, that's an increase of 5 percentage points, or 50% relative to the original share.

How to distinguish correlation from causation

Suppose advertising expenses increased at the same time as sales.

AI might suggest that advertising caused the sales growth. However, those two metrics alone aren't enough to support that conclusion.

Sales could also have been affected by product selection, seasonality, discounts, or other factors.

Ask AI to separate observations from hypotheses:

Identify metrics that changed at the same time. Do not claim that one metric caused another to change. Explain what additional data would be needed to test the hypothesis.

Forecasts require even more caution. One week of data isn't enough to reliably predict sales for the following month.

How to combine two tables

Sometimes the information you need is spread across different tables.

For example, the first table contains order IDs and product codes, while the second contains product codes and product names.

The task is to match records using a shared identifier.

Prompt:

I have two tables. The first contains order IDs and product codes. The second contains product codes and product names. Explain how to add the product names to the first table by matching product codes. Do not match records based on row order. Show how to identify codes that have no match.

Lookup functions, including VLOOKUP and XLOOKUP, are suitable for this task.

If the same product code appears multiple times in the reference table with different names, the match will be ambiguous. Check that the identifiers are unique first.

As a beginner, the main principle to understand is that tables should be joined using a verifiable common field, not because their rows appear similar.

How to generate and verify an Excel formula

AI can explain existing formulas and suggest new ones.

In your prompt, specify the application, cell ranges, locale, and expected result.

Let's use this table:

A: Product	B: Category	C: Amount, $
Notebook	Office supplies	8
Novel	Books	12
Dictionary	Books	15
Pen	Office supplies	5
Sketchbook	Books	7

The headers are in row 1. The data occupies rows 2–6.

We need to calculate the total sales amount for the “Books” category.

Prompt:

Write a SUMIF formula for English-language Excel using commas as argument separators. Categories are in B2:B6, and amounts are in C2:C6. I need to sum the amounts for the “Books” category. Explain the arguments and verify the result manually.

Formula:

=SUMIF(B2:B6,"Books",C2:C6)

Manual verification:

12 + 15 + 7 = $34.

If Excel shows a different number, check the category values and the cell range boundaries.

Useful Excel functions to know

Function	Purpose	Example
IF	Test a condition	Flag a payment
SUMIF	Sum values that meet a condition	Expenses by category
VLOOKUP	Find a value in a lookup table	Retrieve a product price
XLOOKUP	Find a matching value	Retrieve a product name by code

You can ask AI to choose the right function once you've described the result you need.

VLOOKUP depends on how the columns are arranged. XLOOKUP isn't supported in every version of Excel.

Formula syntax varies by application and locale. English-language Excel often uses commas as argument separators and English function names. Other regional settings may use semicolons, and localized versions may use translated function names.

For Google Sheets, specify the spreadsheet's locale and the function names you're using.

How to understand someone else's formula

Sometimes you inherit a spreadsheet with a formula written by someone else, and the formula spans several lines.

Instead of deleting it, ask AI to explain it.

Explain the following Excel formula step by step. Identify which cells it references, what conditions it checks, and what result it returns. Do not modify the formula. List possible causes of errors separately.

If an error occurs, share the error message and a small sample of the data.

Don't just ask AI to “fix the formula”. Explain what result you expected and what actually happened.

How to verify a formula

1. Enter the formula in the correct cell.
2. Check that the application accepts the syntax.
3. Compare the result with a manual calculation.
4. Change one input value and verify that the formula recalculates.
5. Check what happens when there are no matching records.

If you copy a formula down a column, some cell references may need to be fixed using the $ symbol.

For example, $B$2:$B$6 keeps the original range unchanged when the formula is copied.

AI can explain which references to lock in your spreadsheet. But you still need to test the copied formula in the spreadsheet application.

How to highlight values in a spreadsheet

Color highlighting helps you spot large expenses and unusual values.

Suppose you want to highlight amounts above $100.

Excel and Google Sheets have a feature called conditional formatting for this purpose.

Prompt:

Write instructions for Excel explaining how to highlight cells in column C with a red fill when the amount exceeds $100. The data is in rows 2–100. Specify which range to select and which condition to apply.

AI should explain the steps in the application's interface.

Mia chat doesn't change the formatting of the original Excel file. You apply the rule yourself in the spreadsheet application.

For repetitive tasks, you can use VBA macros, Google Apps Script, regular expressions, or Power Query. These tools can automate data processing but require additional knowledge.

For now, you only need to describe the task clearly and verify the result. You don't need to learn programming or automation yet.

How to verify calculations and findings

The lesson “AI hallucinations: why AI makes mistakes and how to check” covers five verification steps: identify the claim, find the original source, check the numbers, verify the reasoning, and compare the result with the original question.

Use the same process for spreadsheets. The original spreadsheet is your source of numerical data, and you should also verify calculations in Excel or Google Sheets.

Common errors

Error	How to detect it
Missing rows	Compare record counts
Mixed-up columns	Check the headers
Rounding errors	Compare the original numbers
Invented values	Find each number in the source
Incorrect date format	Clarify the month and day
Incorrect decimal separator	Check numeric cells
Incorrect grouping	Verify which records belong to each category

Pay particular attention to missing values.

For example, if a spreadsheet contains five orders but amounts are known for only four, the average for the four known orders and the average across all five orders represent different metrics.

You cannot replace a missing amount with zero without explaining why.

How to classify verification results

Your verification should produce one of three outcomes.

Verified. The number matches an independently checked calculation, and all required rows are included.

Disproven. There is a specific discrepancy, such as an incorrect total, a missing row, or a faulty formula.

Unverified. There isn't enough information to reach a conclusion. For example, it's unclear which expenses are included, or the original spreadsheet is incomplete.

If the model makes a mistake, point out the exact discrepancy.

Excel shows total expenses of $597, but you calculated $602. Show the original rows and numbers you used to calculate your total. Identify the cause of the discrepancy without changing any data.

Verify the AI model's explanation against the original spreadsheet as well.

For accounting, tax, and other financially significant documents, checking the results with an AI chatbot alone isn't sufficient.

How to choose the right chart for your data

Charts help you see differences between categories, trends over time, and how individual values contribute to a total.

AI can recommend a chart type, describe the axes, and provide instructions for Excel or Google Sheets.

Task	Chart type
Daily sales	Column chart
Monthly trends	Line chart
Category shares	Pie chart
Sales and expenses	Clustered column chart

A clustered column chart works well for comparing sales and expenses over a week.

Prompt:

Recommend a chart for comparing sales and expenses using the seven-day table. Specify the chart type, horizontal and vertical axes, units of measurement, and series names. Then provide instructions for Excel. Do not add any metrics that aren't in the data.

Expected description:

• Type: Clustered column chart.
• Horizontal axis: Days of the week.
• Vertical axis: Amounts in US dollars.
• Series: “Sales” and “Expenses”.
• Title: “Weekly sales and expenses”.

In Excel, select the data, including the headers. Open the “Insert” tab and choose a clustered column chart.

In Google Sheets, go to “Insert”, select “Chart”, and choose the appropriate type in the Chart editor.

After creating the chart, check the labels, axis scales, and whether the plotted values match the original spreadsheet.

Pie charts work best for showing parts of a whole. They're generally less useful for comparing changes over time.

What you can do in Mia chat

In Mia, you can upload a CSV file, get a written summary, or ask for a formula or chart description.

The chat responds with text. You can transfer tables from its responses to Excel or Google Sheets.

Mia doesn't create XLSX files or generate finished chart images. You'll need to build the chart in a spreadsheet application.

Models in Mia chat also don't search the internet. Provide the source data you want to analyze.

Practice your spreadsheet skills

Complete these three exercises in order: clean a table, analyze metrics, and generate a formula.

All data is fictional. You can use the free Mia chat to complete the exercises.

Paid GPT, Claude, Gemini, and Grok models are also available in the same conversation for credits. You can switch models without starting a new chat.

Exercise: Clean up orders

The first exercise contains six orders with inconsistent date formats, category names, and amounts.

Create a cleaned table. Verify that the order IDs and row order haven't changed.

Expected result: six records, total amount $50.01.

Compare the cleaned rows with the originals. If the model changes a numeric value, ask it to restore the original.

Exercise: Analyze sales

The second exercise contains seven days of sales and expenses.

Calculate both totals, the difference between them, and the days with the highest and lowest sales. Ask AI to show every addition.

Expected results: sales of $1,000, expenses of $597, and a difference of $403.

The highest sales occurred on Saturday: $220.

After verifying the calculations, request two findings. Don't accept explanations for sales growth without additional data.

Exercise: Generate an Excel formula

The third exercise asks you to write a SUMIF formula for the “Books” category.

Check the B2:B6 and C2:C6 ranges, argument separators, and result.

Correct total: $34.

If the amount for “Novel” increases by $1, the result should become $35.

How to write a prompt for your own spreadsheet

Choose your task, how you'll share the data, the output format, and the verification method.

The prompt builder will assemble instructions that you can customize with your own data.

For calculations, always specify the units of measurement, time period, and rules for handling missing values. For formulas, include column names and row numbers.

If you want to modify an existing spreadsheet, save a copy of the original data before making changes.

Test your knowledge

Checklist before using AI-generated numbers

Frequently asked questions

Which AI tools work with Excel?

ChatGPT, Claude, DeepSeek, and other AI chatbots can explain formulas, transform data, and provide instructions for Excel. The ability to upload and modify files directly depends on the specific tool.

For simple tasks, a regular AI chat that accepts text or CSV files is enough.

Can I upload an Excel file to Neuromia?

Mia doesn't accept XLSX files. Save the sheet you need as CSV and attach the exported file.

TXT and PDF files are also supported. Text attachments have a limit of 60,000 characters.

Can AI create a ready-to-use spreadsheet?

Yes. AI can generate a table as text that you can transfer to Excel or Google Sheets.

Mia doesn't create finished XLSX files. If the pasted data ends up in a single column, ask for CSV or tab-separated text.

Can I trust AI-generated calculations?

You should verify totals independently.

AI might skip a row, mix up categories, misinterpret a decimal separator, or make an addition error.

Ask AI to show its calculations, then compare the total with a formula in your spreadsheet.

How do I get an Excel formula from AI?

Specify the column names, cell ranges, task, and locale.

For example: “Write a formula for English-language Excel that adds the amounts in C2:C6 when B2:B6 contains 'Books'. Use commas as argument separators.”

After receiving the formula, test it in Excel.

How many spreadsheet rows can I paste into AI?

The number of rows depends on the length of the data and the chat's limits.

In Mia, a single message can contain up to 10,000 characters, while a text attachment can contain up to 60,000 characters.

For large tables, use CSV or process the data in parts. Verify that every part is complete and that all records are included in the final dataset.

Is it safe to upload spreadsheets containing work data?

Remove unnecessary personal and confidential information before uploading.

Use fictional data for learning. For work spreadsheets, check whether your organization's policies allow you to share information with the selected tool.

If the data cannot be shared, use an anonymized example with the same structure.

What to learn next

After analyzing a spreadsheet, you have metrics you can explain to others: total sales, expenses, the difference between them, and daily comparisons.

The next lesson, “How to make a presentation with AI: step-by-step”, shows you how to turn these results into slides.

Save the results from the second exercise: sales of $1,000, expenses of $597, a difference of $403, and the highest daily sales of $220 on Saturday.

You can use these numbers to create a slide with summary metrics, a chart comparing daily figures, and a slide presenting verified findings.