How to Analyze Data With ChatGPT: Excel, CSVs, Charts, and Reports

How to Analyze Data With ChatGPT: Excel, CSVs, Charts, and Reports

Business data often begins as rows and columns rather than useful answers. Sales exports, marketing reports, customer lists, inventory files, survey results, financial models, and operational spreadsheets may contain valuable information, but someone still needs to clean, compare, calculate, visualize, and interpret it.

ChatGPT can assist with this process by analyzing structured data, explaining columns, finding missing values, calculating metrics, comparing groups, identifying unusual records, creating charts, and turning findings into understandable reports.

The goal is not to replace careful analysis. Incorrect column definitions, missing rows, poor data quality, inappropriate statistical methods, or misunderstood business context can still produce misleading conclusions.

This guide explains practical ChatGPT data analysis workflows using Excel files, CSVs, tables, charts, and business reports while keeping important calculations and decisions reviewable.

You can also read how to use ChatGPT for work or 25 practical ChatGPT workflows for small businesses.

Quick Answer: What Data Can ChatGPT Help Analyze?

ChatGPT can help investigate many forms of structured business data, including:

Sales transactions.
Marketing campaign exports.
Customer lists.
Product and inventory data.
Survey responses.
Website-performance exports.
Support tickets.
Operational metrics.
Project data.
Financial and budgeting spreadsheets prepared for appropriate review.
Employee or workforce data where privacy and employment controls allow it.

A good workflow normally follows this sequence:

Prepare → Inspect → Clean → Analyze → Visualize → Verify → Communicate.

The final step is important. A chart or calculated number becomes useful only when its definition, assumptions, limitations, and business meaning are understood.

What ChatGPT Can Do With Structured Data

Depending on the available ChatGPT tools and account capabilities, data-analysis tasks may include:

Summarizing rows and columns.
Explaining a dataset's structure.
Finding missing or duplicate values.
Grouping and filtering records.
Calculating sums, averages, medians, counts, percentages, and rates.
Comparing periods or customer segments.
Merging datasets through shared identifiers.
Finding possible anomalies or outliers.
Creating tables.
Creating charts.
Running code-backed calculations.
Performing statistical analysis.
Turning findings into executive summaries or reports.

Exact features can change, so consult current official ChatGPT documentation when a particular file type, integration, or spreadsheet capability is essential.

Start With the Decision, Not the Spreadsheet

Uploading a large file and asking ChatGPT to find something interesting can produce observations, but it may not answer the question your business actually needs to solve.

Start with a decision such as:

Which marketing channel deserves more budget?
Which products are losing sales?
Which customer group has the strongest repeat-purchase rate?
Why did support volume increase last month?
Which invoices need review?
Which locations have unusually high cancellation rates?
Which inventory items may need attention?

Then state which data should support that decision.

Example:

I am trying to decide which marketing channels should receive more budget next quarter based on conversion rate, acquisition cost, revenue, and customer quality. First inspect this dataset and tell me whether the available columns are sufficient to answer that question.

Prepare Your Spreadsheet Before Uploading It

Clean structure makes analysis easier to verify.

Use:

One descriptive header row.
One record per row.
Consistent date formats.
Consistent currency and units.
Clear column names.
Stable customer, order, campaign, or product identifiers.
Separate columns for separate concepts.

Avoid where possible:

Multiple unrelated tables inside one sheet.
Blank rows dividing a dataset.
Merged cells inside data tables.
Important values stored only inside images.
Several meanings combined into one column.
Totals mixed with transaction rows.
Unexplained abbreviations.

If changing the source spreadsheet could create risk, keep the original untouched and prepare a separate analysis copy.

Create a Data Dictionary

One of the most useful things you can provide is a short explanation of what each important column means.

Example:

order_id = unique order identifier
customer_id = customer identifier
order_date = date order was completed
gross_revenue = customer payment before refunds
refund_amount = refunds issued against the order
channel = first attributed marketing channel
country = customer billing country

This prevents terms such as revenue, active customer, qualified lead, conversion, profit, and completed order from being interpreted incorrectly.

When two teams define the same KPI differently, ask ChatGPT to identify the conflict instead of choosing a definition automatically.

Step 1: Inspect the Dataset Before Analyzing It

Begin with an exploratory inspection rather than immediately requesting conclusions.

Useful prompt:

Inspect this dataset before doing any business analysis. Report:
1. Number of rows and columns.
2. Column names and likely data types.
3. Date range.
4. Missing values by column.
5. Duplicate rows and duplicate identifiers.
6. Suspicious values or formatting problems.
7. Columns that appear to require definitions.
8. Questions you need answered before calculating KPIs.

Do not make business recommendations yet.

This first pass often reveals problems that would otherwise contaminate every later calculation.

Step 2: Clean Missing Values Carefully

A blank cell does not always mean the same thing.

It might mean:

Information was never collected.
The value is genuinely zero.
The value does not apply.
A system failed to record it.
The customer declined to provide it.
A join between two datasets failed.

Do not automatically replace every blank with zero.

Useful prompt:

Identify missing values and group them by column. Before filling or removing anything, explain the possible meaning of each missing field and recommend which cases require business clarification.

Step 3: Find and Review Duplicates

Duplicates can distort customer counts, revenue totals, conversion rates, inventory records, and operational reporting.

Ask ChatGPT to distinguish:

Exact duplicate rows.
Repeated customer IDs.
Repeated transaction IDs.
Several legitimate purchases by the same customer.
Records that look similar but represent different events.

Example:

Find duplicate order_id values. For each duplicate, show the complete rows side by side and identify whether they appear identical, conflicting, or potentially legitimate. Do not delete anything.

Review the records before removing data.

Step 4: Check Data Types and Formats

A number stored as text or a date interpreted incorrectly can break an otherwise simple analysis.

Review:

Dates.
Currencies.
Percentages.
Decimal separators.
Negative values.
Boolean fields.
Category labels.
Country and region names.
Identifiers with leading zeros.

Example:

Check whether every field has a suitable data type. Flag dates stored as text, percentages stored inconsistently, numeric columns containing text, and IDs that should remain strings.

Step 5: Calculate Business KPIs

Once definitions and data quality are understood, ask for the required metrics explicitly.

Examples include:

Revenue.
Order count.
Average order value.
Conversion rate.
Refund rate.
Repeat-customer rate.
Customer acquisition cost.
Cost per lead.
Gross margin where valid cost data exists.
Ticket-resolution time.
Cancellation rate.
Inventory turnover where the required data is available.

Example prompt:

Using the definitions below, calculate the requested KPIs by month. Show the formula or calculation method for each metric before presenting the results. If the dataset does not contain the information required for a metric, mark it as unavailable instead of estimating it.

Ask ChatGPT to Show the Calculation

Never rely only on a polished KPI table when the number affects an important decision.

Ask for:

The formula.
Columns used.
Filters applied.
Rows excluded.
Date range.
Treatment of missing values.
Treatment of refunds or cancellations.
Any assumptions.

Useful prompt:

For every KPI in this table, show its definition, formula, source columns, filters, exclusions, and assumptions. Then identify which calculations I should independently verify before using this report.

Step 6: Compare Periods Correctly

Businesses frequently compare week over week, month over month, quarter over quarter, or year over year.

Before comparing, check whether periods are actually comparable.

Possible problems include:

A partial current month compared with a complete previous month.
Different numbers of business days.
Seasonality.
Holiday effects.
Campaign launches.
Price changes.
Tracking changes.
Missing data.

Prompt:

Compare monthly performance, but first identify whether each period is complete and comparable. Flag partial periods, major data gaps, or structural changes that could make the comparison misleading.

Step 7: Segment the Data

An overall average can hide important differences between groups.

Useful segments may include:

Marketing channel.
Product category.
Customer type.
Country or region.
New versus returning customers.
Device type.
Sales representative.
Support category.
Subscription plan.
Order size.

Example:

Calculate conversion rate, average order value, and refund rate by acquisition channel. Include sample size for every channel so a very small group is not presented as equally reliable.

Step 8: Identify Trends

Trend analysis should distinguish an ongoing pattern from one unusual period.

Ask ChatGPT to examine:

Direction.
Magnitude.
Duration.
Seasonality.
Changes in volatility.
Sudden breaks.
Differences between groups.

Example:

Analyze the monthly trend in revenue and order count. Identify sustained increases or declines separately from one-month spikes. Then list possible explanations as hypotheses, not confirmed causes.

Step 9: Find Anomalies and Outliers

An unusual value may indicate fraud, an operational problem, a data error, or simply a legitimate exceptional event.

Ask ChatGPT to flag—not automatically remove—unusual records.

Example:

Scan this transaction file for unusual values, duplicate identifiers, sudden vendor changes, unusually high amounts, and patterns that differ materially from the rest of the dataset. Create a review table and explain why each row was flagged. Do not label anything as fraud.

Human investigation should determine what the anomaly actually means.

Step 10: Create Useful Charts

Charts should answer a question rather than decorate a report.

Common choices include:

Line chart: Trends over time.
Bar chart: Comparing categories.
Stacked bar: Composition across groups.
Scatter plot: Relationships between numerical variables.
Histogram: Distribution of values.
Box plot: Spread and potential outliers.

Prompt:

Create the three charts that best explain the change in monthly sales. Give each chart a descriptive title, label axes and units clearly, and explain in one paragraph what the chart shows and what it does not prove.

Do Not Choose Charts Only Because They Look Impressive

A visually complex chart can make a simple message harder to understand.

Before accepting a chart, check:

Does the axis start at an appropriate value?
Are units clearly labelled?
Are dates ordered correctly?
Are categories readable?
Are important differences visible without exaggeration?
Is the sample size shown when relevant?
Does colour communicate information rather than decoration?
Could a simple table explain the result more clearly?

The goal is understandable evidence, not visual complexity.

Step 11: Analyze Multiple Files Together

Some questions require more than one dataset.

Examples include:

Customers + orders.
Campaigns + leads.
Products + inventory.
Invoices + payments.
Support tickets + customer accounts.

Before merging, identify the shared key such as customer_id, order_id, product_id, or campaign_id.

Example:

I have uploaded customers.csv and orders.xlsx. Determine the safest shared field for joining them. Before merging, report duplicate keys, unmatched records, and any many-to-many relationship that could inflate totals.

Incorrect joins are a common source of duplicated totals.

Step 12: Analyze Marketing Data

Marketing analysis should connect spending and activity with meaningful outcomes.

Possible questions include:

Which channels generate qualified leads?
Which campaigns have the strongest conversion rate?
How does cost per acquisition change over time?
Which campaigns have high traffic but weak sales?
Which channels create repeat customers?

Prompt:

Analyze performance by channel. Calculate spend, leads, conversions, conversion rate, cost per lead, cost per acquisition, and revenue where valid data is available. Flag metrics that cannot be calculated reliably. Then identify three questions the marketing team should investigate.

Step 13: Analyze Sales Data

Sales analysis may include:

Revenue by period.
Units sold.
Average order size.
Product mix.
New versus returning customers.
Regional performance.
Salesperson performance.
Discount effects.
Refund patterns.

Example:

Compare this quarter with the previous quarter. Separate changes caused by order volume from changes caused by average order value. Then segment the result by product category and customer type.

Step 14: Analyze Customer Data

Customer datasets can help businesses understand behaviour, but privacy and appropriate use matter.

Possible analysis includes:

New versus repeat customers.
Purchase frequency.
Average customer value.
Product preferences.
Cancellation patterns.
Support frequency.
Geographic distribution.

Avoid making unsupported assumptions about sensitive personal characteristics.

Prompt:

Segment customers using only purchase frequency, recency, order value, and product category. Do not infer age, income, ethnicity, health status, or other personal attributes that are not explicitly and appropriately available.

Step 15: Analyze Operational Data

Operations teams may use data to identify delays, capacity issues, quality problems, and recurring exceptions.

Examples include:

Order fulfilment time.
Appointment cancellations.
Support response times.
Inventory shortages.
Production delays.
Project completion time.
Supplier performance.

Prompt:

Analyze processing time by workflow stage. Identify the stages with the largest delays, compare medians as well as averages, and flag whether a small number of extreme cases are distorting the result.

Step 16: Use Statistics Carefully

ChatGPT can assist with statistical methods, but selecting a calculation mechanically does not make the conclusion valid.

Questions to clarify include:

What hypothesis is being tested?
What type of variables are involved?
Is the sample representative?
Are observations independent?
Are important assumptions satisfied?
Is the sample large enough?
Are several comparisons being tested at once?

Ask ChatGPT to explain why it selected a method and which assumptions the method requires.

For important scientific, medical, financial, or policy work, use a qualified statistician or subject-matter expert.

Correlation Is Not Automatically Causation

Two variables moving together does not prove that one caused the other.

For example, sales may increase during a marketing campaign, but seasonality, pricing changes, new inventory, competitors, holidays, or another campaign may also contribute.

Useful prompt:

Identify correlations in this dataset, but do not describe them as causal. For each important relationship, list possible confounders and additional evidence required before making a causal claim.

Step 17: Create Forecasts With Clear Assumptions

Forecasts are estimates, not guaranteed future results.

Before forecasting, review:

Historical coverage.
Seasonality.
Outliers.
Structural changes.
Missing periods.
Recent price changes.
Market changes.
Whether older data remains relevant.

Example:

Build a simple eight-week demand forecast from this historical data. Explain the method, assumptions, limitations, and confidence range. Compare it with a simple baseline and do not present the forecast as guaranteed demand.

Step 18: Turn Analysis Into a Business Report

A strong report explains decisions rather than simply reproducing calculations.

Use a structure such as:

Objective
What question was investigated?

Data
Which sources, periods, and records were used?

Quality notes
Which missing values, exclusions, or limitations matter?

Findings
What does the evidence show?

Interpretation
Why might the findings matter?

Risks
What could make the interpretation wrong?

Actions
Which decisions or additional investigations are recommended?

Verification
Which calculations or assumptions still require human review?

Executive KPI Report Prompt

Analyze the uploaded dataset and create an executive KPI report.

First:
- Inspect data quality.
- Confirm KPI definitions.
- Identify incomplete periods.
- State any exclusions.

Then provide:
1. KPI summary table.
2. Month-over-month change.
3. Year-over-year change where sufficient data exists.
4. Three useful charts.
5. Five key findings.
6. Three risks or anomalies.
7. Five recommended follow-up questions.

Separate observed facts from hypotheses. Do not invent reasons for changes that the dataset cannot demonstrate.

Marketing Analysis Prompt

You are supporting a marketing analyst.

Using this campaign dataset, calculate performance by channel and campaign.

Include:
- Spend
- Impressions
- Clicks
- Leads
- Purchases
- Conversion rate
- Cost per lead
- Cost per acquisition
- Revenue
- ROAS only where revenue attribution is valid

Before calculating, identify missing fields, attribution limitations, duplicate records, and incomplete periods.

Then recommend the next questions to investigate rather than automatically recommending budget changes.

Sales Analysis Prompt

Analyze this sales dataset to understand what changed between the last two complete quarters.

Break the change into:
- Order volume
- Average order value
- Product mix
- New vs returning customers
- Geography
- Refunds

Show calculations, sample sizes, and exclusions. Identify unusual records separately and explain which findings are observations versus possible explanations.

Data-Cleaning Prompt

Audit this dataset for quality problems before analysis.

Report:
- Missing values
- Exact duplicates
- Duplicate IDs
- Inconsistent categories
- Invalid dates
- Impossible numerical values
- Mixed units
- Text in numeric fields
- Suspicious outliers
- Unmatched records

Do not modify or remove records yet. Create a proposed cleaning plan showing the reason and risk for every suggested change.

Chart Creation Prompt

Create visualizations for this dataset, but choose charts based on the analytical question rather than appearance.

For each chart provide:
- Question being answered
- Chart type
- Columns used
- Filters
- Units
- Key observation
- Limitation

Keep the visualization readable and do not exaggerate differences through misleading axes.

Anomaly Review Prompt

Scan this dataset for records that deserve manual review.

Look for:
- Duplicate identifiers
- Unusually large values
- Unexpected negative amounts
- Sharp changes from historical behavior
- Unusual category combinations
- Missing information in otherwise complete records

Create a review table with record ID, reason flagged, evidence, severity, and recommended next check.

Do not label any record as fraud or error without independent verification.

Ask ChatGPT to Critique Its Own Analysis

After receiving the first analysis, challenge it.

Useful follow-up prompts include:

Which findings depend most heavily on assumptions?
Which calculations should I independently verify?
What data-quality issue could materially change the conclusion?
Which segment has too little data for a reliable comparison?
What alternative explanation fits these results?
What conclusion are we not justified in making?
Which additional columns would improve this analysis?

This review step often adds more value than requesting another chart.

Verify Results Before Making Decisions

For important analyses, check the result against the original data.

Verify:

Record counts.
Grand totals.
Key formulas.
Date ranges.
Filters.
Join keys.
Excluded rows.
Currency and units.
Definitions.
Unexpected values.

Spot-check individual records behind major findings.

If a report claims one product category generated the majority of revenue, manually trace several source rows and confirm the aggregation before acting on it.

Protect Sensitive Data

Business datasets may include private customer, employee, financial, health, or operational information.

Remove or anonymize unnecessary sensitive fields before analysis where possible.

Be particularly careful with:

Passwords and credentials.
Banking information.
Government identifiers.
Health data.
Payroll information.
Private customer records.
Employee performance records.
Legal documents.
Payment-card information.
Confidential business metrics.

Review your organization's account settings, contractual obligations, privacy policies, retention requirements, and applicable laws before processing sensitive data.

ChatGPT in Excel and Google Sheets

Current ChatGPT experiences can also support spreadsheet work directly inside Excel and Google Sheets for eligible users and workspaces.

These tools may assist with tasks such as understanding an existing workbook, building formulas, updating tables, analyzing large multi-tab spreadsheets, and answering questions from spreadsheet context.

Because spreadsheet-native features, permissions, and availability may change, treat this as an optional workflow rather than making it the foundation of your analysis process.

The evergreen skill remains the same: define the decision, understand the data, inspect assumptions, verify calculations, and document the result.

Common ChatGPT Data Analysis Mistakes

Avoid these problems:

Uploading messy data and immediately asking for conclusions: Quality problems remain hidden.
Not defining columns: Important metrics may be interpreted incorrectly.
Replacing every blank with zero: Missing information is misclassified.
Removing outliers automatically: Legitimate exceptional events disappear.
Ignoring sample size: Small groups appear more reliable than they are.
Trusting a chart without checking its data: Visual errors become business conclusions.
Confusing correlation with causation: Relationships are described as proven causes.
Using incomplete periods: Month-to-month comparisons become misleading.
Skipping formula verification: Incorrect definitions appear authoritative.
Uploading unnecessary sensitive information: Privacy risk increases.
Using an AI-generated forecast as certainty: Planning ignores uncertainty.

A Seven-Step ChatGPT Data Analysis Workflow

1. Define:
Write the business decision and KPI definitions.

2. Prepare:
Structure the spreadsheet and create a data dictionary.

3. Inspect:
Review rows, columns, missing values, duplicates, date range, and formatting.

4. Analyze:
Calculate the required metrics and segment the results.

5. Visualize:
Create charts that answer specific questions.

6. Verify:
Check formulas, totals, assumptions, exclusions, joins, and source records.

7. Communicate:
Turn confirmed findings into an understandable report containing limitations and next actions.

Data Analysis Quality Checklist

Before using the result, confirm that:

The business question is clearly defined.
The dataset covers the correct period.
Column definitions are understood.
Missing values were reviewed.
Duplicate identifiers were investigated.
Units and currencies are consistent.
Incomplete periods are labelled.
KPIs use approved definitions.
Filters and exclusions are documented.
Dataset joins were checked.
Charts use appropriate scales.
Sample sizes are considered.
Correlation is not presented automatically as causation.
Important calculations were independently checked.
Private information was handled appropriately.
A responsible person has reviewed the business conclusion.

When deeper analysis is required, you can post a data-analysis project on UstadWork or browse data, Excel, reporting, and automation specialists.

When to Hire a Data Analyst

ChatGPT can assist with many routine analytical tasks, but professional support becomes more important when the work involves:

Large or complex databases.
Business-critical financial decisions.
Statistical modelling.
Experimental design.
Forecasting with major financial consequences.
Healthcare or regulated information.
Machine learning.
Complex attribution.
Data engineering.
Executive dashboards tied to several systems.
High-stakes customer or employee decisions.

A capable data professional should explain the data source, methodology, assumptions, limitations, validation process, and how the output supports the business decision.

Use our freelancer hiring guide before comparing candidates.

Official Resources for Current Features

ChatGPT file support, spreadsheet tools, integrations, plans, and workspace controls can change. Check current official documentation when exact functionality matters.

OpenAI Academy: Analyzing Data With ChatGPT
OpenAI Help: Data Analysis With ChatGPT
OpenAI Help: Extracting Insights With Data Analysis
OpenAI Help: Supported File Types

Turn Data Into Decisions Without Skipping Verification

ChatGPT can make spreadsheet analysis more approachable by helping inspect datasets, clean common issues, calculate metrics, build charts, compare segments, identify unusual records, and communicate findings.

The strongest results still begin with a clearly defined business question and well-structured data. Important conclusions should show their formulas, assumptions, filters, and limitations so another person can review them.

Effective ChatGPT data analysis is therefore not simply upload a spreadsheet and trust the answer. It is a repeatable process of preparing, inspecting, analyzing, visualizing, verifying, and communicating evidence responsibly.

Continue with ChatGPT for Work, ChatGPT for Small Business, the Deep Research tutorial, or post a data project on UstadWork.