ChatGPT Data Analysis for Marketing Exports

#01How should you prepare an export?

OpenAI's data analysis article asks for descriptive headers in the first row, one row per record and no empty rows or columns splitting the data. For marketing exports, also:

• Delete the totals row at the bottom of GA4 and ad platform exports.
• Use one date format and say which it is (DD/MM or MM/DD).
• Keep currency in one column with a stated currency.
• Remove names, emails and any personal data from CRM exports.

#02Which analyses are worth running?

#03Which prompts should you copy?

FIRST LOOK
Attached: [file]. Before analysing, describe the columns,
row count, date range, and any data problems (blanks,
duplicates, odd values). Don't calculate anything yet.

CHANNEL QUALITY
Join sessions.csv and leads.csv on date and source/medium.
For each channel with at least 200 sessions, show sessions,
leads, lead rate and change versus the previous period.
Show the code, then a table, then a bar chart of lead rate.

CAMPAIGN READ
Compare the 4 weeks before [date] with the 4 weeks after,
matching weekdays. Report the change in [metric] and list
other things that could explain it (seasonality, tracking
changes, other campaigns). Don't claim causation.

CHECK YOUR WORK
List every filter, join and assumption you used. Show the
row counts before and after each step.

Our view: the "first look" prompt saves the most time. Most bad analysis comes from a data problem nobody spotted, not from a wrong formula.

#04How do you check the results?

Ask for the code. Look at the row counts after each join. Recompute one total in your spreadsheet. If a number will go to leadership, rebuild it in a sheet you own; ChatGPT then becomes the fast explorer, not the system of record.

For AI referral tracking, see tracking AI traffic in GA4 and the GA4 AI referrals channel group.

#05Worked example: Ledgerline's channel review

Ledgerline is a fictional accounts-payable tool. Details are illustrative.

Ledgerline uploaded a GA4 export and a CRM lead export. The first-look prompt found the CRM file used US date format and the GA4 file used UK format; fixing that changed which week the webinar leads landed in. The channel-quality table then showed organic search sent fewer sessions than paid social but had a much higher lead rate. The team recomputed the two lead rates in a spreadsheet, confirmed them and shifted next quarter's content budget.

#06What should you check before the next lesson?

• Clean one export to the checklist above.
• Run the first-look and check-your-work prompts.
• Recompute one figure yourself.
• Claude users can compare Claude CSV analysis for GA4 and ads.

Next: Deep research for competitive intelligence. Or return to the ChatGPT for Marketers track.

#07Common questions

Can ChatGPT analyse a GA4 export? Yes. Export a report to CSV or a spreadsheet, check the headers are in the first row and remove summary rows, then upload it and ask specific questions. ChatGPT runs Python on the file.

Can I trust ChatGPT's numbers? Check them. Ask it to show the code and the intermediate tables, and recompute one or two totals yourself in a spreadsheet.

Does ChatGPT connect directly to Google Analytics? Check the Plugin Directory for what is available to your plan and workspace. File exports are the dependable route and keep you in control of what is shared.