ChatGPT, Claude, Gemini + WooCommerce

Talk to your store like you’d talk to an analyst.

“Which campaigns are losing me money?” “Show me the best hour to launch ads.” “What do my repeat customers come back for?” “Which landing page actually sells this product?”

You ask. The AI digs through your orders, builds the chart, and tells you what to do. The whole flow takes minutes. No SQL, no spreadsheets, no BI tool, no analyst.

PixelYourSite makes it possible by exporting every WooCommerce or EDD order as a single AI-ready CSV with the full attribution stack: traffic source, landing page, UTMs (first-visit and last-visit), products, totals, and profit if you use Cost of Goods. Drop that file into ChatGPT, Claude, Gemini, or Copilot and start asking.

What you can export

Two CSVs, both available on the same page.

The ChatGPT-friendly file. Comprehensive export of every order in the date range with full attribution. Use this one for AI analysis.

• Where: PixelYourSite Pro → WooCommerce Reports (or Easy Digital Downloads Reports) → Export all.
• Filename: yourdomain-chatgptfile-woo-2026-01-01_to_2026-03-31.csv.

The current-view export. Whatever filters and segments you have active in the reports UI, exported as CSV.

• Same page, Export current button.
• Useful when you’ve already narrowed down to a campaign or source and want to share just that slice.

The chatgptfile contains, per order: order ID, date and time, total, currency, payment method, customer email, billing city/state/country, first-visit traffic source, first-visit landing page, first-visit UTMs (source, medium, campaign, term, content), last-visit traffic source, last-visit landing page, last-visit UTMs, products in the order (IDs, names, quantities, prices), and — if you use WooCommerce Cost of Goods — the cost per product and the profit per order.

Which AI tool to use

You need an AI assistant that can actually run code on the file you upload. Plain chat can only read snippets. Here’s what works:

Tool	Tier needed	Notes
ChatGPT	Plus, Team, or Pro (Advanced Data Analysis)	The most polished workflow. Generates charts you can download.
Claude	Free or paid (Analysis tool / artifacts)	Strong at narrative analysis and chart artifacts. Free tier handles small files.
Gemini	Advanced (code execution)	Good at charts. Tight integration with Sheets and Drive.
Microsoft Copilot	Copilot Pro / Microsoft 365	Works through Excel for large files.
Mistral Le Chat	Pro (code interpreter)	Good European-hosted alternative if you care about data residency.
Perplexity	Pro	Decent for one-off questions, weaker for multi-chart reports.

Free chat-only tiers (no code execution) won’t run analysis on a CSV. They’ll guess based on the first few rows and the file structure. Use a tier with code execution.

Read the customer data warning first

The chatgptfile contains customer emails and addresses. Before uploading:

• Check the AI provider’s data policy. Most paid tiers let you opt out of training. Check the setting before uploading customer data.
• Strip PII you don’t need for the analysis. Open the CSV, delete the customer_email, billing_city, billing_state, and billing_country columns if you’re only analysing channels and products. The attribution data is what matters for marketing reports — the contact details aren’t.
• Use a privacy-respecting tool for high-volume sites. Mistral Le Chat (EU-hosted), self-hosted models, or Anthropic’s Workbench API with zero-data-retention all work if your privacy bar is high.
• GDPR. If your visitors are in the EU, you’ve already promised them their data won’t be shared with third parties without consent. AI vendors are third parties. The safest practice is to drop the PII columns before upload, every time.

Set the AI up for the analysis

Start the conversation with a single setup prompt so every follow-up question has the right framing. This works in all the tools above:

You’re a marketing data analyst. I’m going to upload a CSV with WooCommerce orders, full attribution (first-visit and last-visit), products, and profit data. For every analysis: show me a clear chart, explain what it means in two sentences, and tell me one specific action to take. Use the cost_of_goods and profit columns when present. Use first_visit_* columns for first-touch attribution and last_visit_* columns for last-touch.

Then upload the CSV and ask any of the questions below.

Prompts that produce useful answers

The original strategy page had generic prompts. These are sharper.

Channel and campaign performance

Which traffic sources produced the most revenue last quarter, and which produced the most profit? Show both as a single bar chart with two bars per source.

Compare first-touch vs last-touch attribution for the top 10 campaigns. Which campaigns get more credit on first-touch than last-touch — those are the ones opening the door but not getting credit in ad-platform reports.

Find UTM campaigns that got at least 10 sessions but produced fewer than 2 orders. Those are likely budget I should cut or messaging I should fix.

Product and customer insight

What’s the average order value per traffic source? Highlight sources where AOV is at least 30% above the site average — those are the visitors I should send more of.

Which products are most often bought together? Show the top 10 product pairs, with the number of orders that contained both.

Group customers by their first-visit traffic source. For each group, calculate: average orders per customer, total lifetime revenue, and repeat-purchase rate. Which source produces the best long-term customers?

Time patterns

Build a heatmap of order count by hour of day and day of week. Which hours and days produce the most orders? When should I schedule ads to peak just before?

Show a daily revenue chart for the period. Mark days with revenue more than two standard deviations above or below the mean and tell me what those days have in common (same campaign, same landing page, same product).

Landing page tests

For each product that was bought through more than one landing page in the period, compare conversion rate (orders / sessions if available, otherwise order share) per landing page. Which landing pages convert best for which products?

Look at orders where first_visit_landing_page is different from last_visit_landing_page. For those, find the most common first → last landing path. Those are the visitor journeys I should optimise.

Profit-focused (needs WooCommerce Cost of Goods)

Calculate profit per campaign, then rank campaigns by profit-per-acquisition (profit ÷ orders). Identify the most profitable acquisition channel even if it’s not the highest revenue.

Which products have a high revenue contribution but a low profit margin? Show a quadrant chart with revenue on X and margin on Y. Products in the bottom-right are the candidates for a price increase or a sourcing review.

Compute “profit after estimated ad cost.” Assume Meta CPM of US$15 and a 2% click-through rate, and Google paid CPC of US$1.20. Show profit per traffic source after subtracting estimated ad cost.

Anomaly and quality checks

Find orders with unusual patterns: orders with a payment method that’s rare for the rest of the data, orders from countries you’ve barely seen before, orders with an AOV more than 5x the mean. Those are worth a manual look — fraud, B2B whales, or data errors.

Find traffic sources that appear in first_visit_* but never in last_visit_*. Those are sources that introduce visitors but never close. Worth checking if those visitors are coming back through a different path you haven’t tagged.

Follow-up questions that get more out of any answer

After any chart the AI gives you, try one of these:

Same chart, but only for products in category X.

Same chart, but split by device category if it’s in the data.

Now give me the same insight as a one-paragraph executive summary I can paste into Slack.

What would the chart look like if I removed the top three outliers? Does the conclusion change?

What’s the smallest action I could take this week based on this data?

A monthly workflow

Set up once, run every month:

1. End of month: WooCommerce Reports → Export all for last month’s date range.
2. Open your AI of choice. Drop the file in.
3. Paste the setup prompt above.
4. Run three or four of the prompts above that match your current questions.
5. Save the charts and the executive summary.
6. Decide one thing to change this month based on what the data shows.

The whole cycle takes 30 minutes once you’re set up.

What this won’t tell you

• Session-level data (sessions, bounce rate, pages per session). The export is order-level. For session metrics, use GA4 and the WooCommerce First-Party Reports.
• Visitor identity beyond email. No fbp, no external_id, no GA4 client ID in the file. Those are pixel-level signals, not order-level.
• Ad spend. The export has revenue and (optionally) profit, but not what you spent in Meta or Google Ads. Combine with platform exports for ROAS.
• Causation. AI gives you correlations, not causation. A campaign with high profit might be selling to people who would have bought anyway. Use the AI to find candidates, then run real tests.