You can analyze marketing data with ChatGPT by uploading a spreadsheet and asking questions in plain English. Behind the scenes it writes and runs Python code to do the math, so the numbers come from real calculations rather than guesswork. Used carefully, it turns an afternoon of pivot tables into twenty minutes. Used carelessly, it produces confident charts from data it misunderstood.
This guide shows how to get reliable answers from it, with prompts built around the reports marketers actually export: ad platform performance, GA4 data and CRM leads.
What the feature can and can’t do
According to OpenAI’s help article on data analysis, ChatGPT accepts spreadsheets (.xls, .xlsx, .csv), PDFs and text-based data files such as JSON. It runs Python in a sandboxed environment to calculate, transform and chart data. Bar, line, pie and scatter charts can be interactive; other chart types come back as static images.
Two limits matter for marketers. It can’t connect to live data sources or make web requests from inside the analysis, so you work with exports, not live dashboards. And file size and upload limits vary by plan and model, so very large raw exports may need trimming first.
Before you analyze marketing data with ChatGPT, clean the file
Most wrong answers come from messy files, not from the model. Five minutes of preparation prevents most of them:
- One header row, with clear column names. “Cost (USD)” beats “Cost” when you have several currencies.
- No merged cells, subtotals or notes in the middle of the data.
- Dates in one consistent format.
- Remove personal data you don’t need. Names, emails and phone numbers rarely help the analysis, and many privacy laws, from GDPR to India’s DPDP Act, expect you to minimize what you share.
Step 1: make it describe the data back to you
Before asking anything clever, ask:
“Describe this file. List each column, its data type, the date range, and any columns with missing or odd values. Don’t analyze anything yet.”
This catches most problems early: a spend column read as text because of currency symbols, a date column in two formats, or two campaigns with near-identical names that should be one.
Step 2: ask specific questions
Vague prompts get vague analysis. Compare these:
| Vague | Specific |
|---|---|
| “Analyze my campaign data.” | “Calculate cost per lead by campaign for August and September, and flag any campaign where it rose more than 25% month on month.” |
| “What’s working?” | “Rank countries by conversion rate, only including countries with at least 200 clicks.” |
| “Make a chart.” | “Line chart of daily spend and daily conversions on two axes for the last 60 days.” |
The minimum-volume condition in the second example is the one most people forget. Without it, a country with 3 clicks and 1 conversion tops the table.
Step 3: check the working
Ask it to show the code, or to state the formula it used. Then spot-check one number against the source file yourself. Cost per lead should be total cost divided by total leads for the period, not the average of each day’s cost per lead. Models occasionally average ratios when they should sum and divide, and the difference can be large.
Prompts for common marketing files
Ad platform export (Google Ads, Meta, DV360)
- “Group by campaign and calculate spend, clicks, conversions, CTR, CPC and CPA. Sort by spend.”
- “Find campaigns with spend above the median and CPA above the account average. These are my review list.”
- “Compare weekday against weekend performance for CPA and conversion rate.”
GA4 export
- “Which landing pages have more than 500 sessions but an engagement rate under 40%?”
- “Show sessions and key events by session default channel group, month by month.”
If you’re not sure which GA4 reports to export in the first place, this guide to tracking website traffic in GA4 covers the useful ones.
CRM or lead export
- “Join this lead export to the ad export on the campaign column and calculate cost per qualified lead, where qualified means status is SQL or Won.”
- “How many days, on average, from lead created to deal won, by source?”
A worked example
Say a travel agency exports three months of Google Ads data across campaigns targeting the UK, UAE and India. After the describe step, the question is: “Which campaigns spent the most per booking, excluding any with fewer than 10 bookings?” ChatGPT returns a table showing one generic “holiday packages” campaign costing three times more per booking than the destination-specific ones. The agency moves budget into the destination campaigns and asks a follow-up: “Chart weekly cost per booking for these four campaigns so I can track the change.” Total time: about fifteen minutes.
Know when to stop
Once you’re asking the same questions every week, a spreadsheet template, a Looker Studio dashboard or a small automation is better than re-uploading files to a chat. The best time to analyze marketing data with ChatGPT is during exploration and one-off questions. Recurring reporting deserves a proper pipeline. That’s the kind of thing we build under AI and automation at Digital Poojari.