How to Track ChatGPT, Claude, and Perplexity Traffic in GA4 and Cross-Reference with Sales

Discovering that your visitors come from AI assistants like ChatGPT, Claude, or Perplexity is just the first step. The question that truly matters for business: how much revenue do these visitors generate and how does this compare to organic or paid traffic?

Kondado is a data pipeline platform that replicates your Google Analytics 4 into SQL-ready tables for analysis. With the Kondado MCP connected to ChatGPT or Claude, you identify AI visitors in GA4 and automatically cross-reference with orders from your ERP, all in natural language, without writing SQL queries or opening multiple spreadsheets.

Why Tracking AI Traffic in GA4 Is Different from Other Channels

When someone clicks a link inside ChatGPT, Claude, or Perplexity, the browser records the domain as the source. The problem: each AI uses a different domain and, by default, GA4 classifies all of them as generic "Referral", mixing with dozens of other websites.

This hides valuable behavior. Visitors arriving through AI curation tend to have:

1. More qualified intent - they went through algorithmic curation before clicking
2. Different engagement - longer sessions and distinct bounce rates
3. Potentially higher conversion - they came from contextual recommendation, not generic search

Without grouping these sources, you cannot compare AI performance against other acquisition channels.

How to Create a Custom Channel Group for AI Traffic in GA4?

GA4 allows you to create custom channel groupings that override the default classification. To isolate AI visitors:

1. Go to Admin → Data Settings → Channel Groups
2. Click Create New or edit the default group
3. Add a new condition with:
   • Source: Matches regex
   • Value: chatgpt.com|perplexity.ai|claude.ai|gemini.google.com|copilot.microsoft.com
4. Name the channel: "AI Assistants"

This grouping appears in the sessionDefaultChannelGroup dimension in tables replicated via Kondado.

Which GA4 Tables Do I Need to Replicate for AI Traffic Analysis?

Kondado replicates 13 GA4 tables. For channel and revenue analysis, the essential ones are:

Primary table:

• ga4_traffic_acquisition_daily: contains sessionSourceMedium, sessionDefaultChannelGroup, totalRevenue, transactions, engagedSessions

Complementary tables:

• ga4_landing_page_analysis: cross-references landingPagePlusQueryString with traffic source
• ga4_ecommerce_revenue_daily: details purchaseRevenue by campaign and channel
• ga4_events_tracking_daily: custom events like conversions and keyEvents

Replication uses date cursor with incremental mode, updating automatically according to the schedule defined in the pipeline.

How to Connect Kondado MCP to Ask Questions in Natural Language

The Kondado MCP (Model Context Protocol) exposes your replicated data as an API that ChatGPT, Claude, and other compatible clients query via OAuth.

Connection steps:

1. In ChatGPT or Claude, go to Settings → MCP/Pipelines
2. Add a new MCP server with URL: https://mcp.kondado.io/mcp
3. Authorize via OAuth on your Kondado account - you will be redirected to https://app.kondado.com.br/oauth/authorize
4. The session is linked to your Kondado Via - all queries will be read-only on your integrated data

Once connected, you can ask questions in natural language about your GA4 data.

What to Ask ChatGPT/Claude About AI Traffic?

Here are practical analyses the MCP answers by cross-referencing GA4 with other data sources:

How much revenue came from AI this month?

Ask: "What is the totalRevenue and transactions from the 'AI Assistants' channel in the last 30 days?"

The LLM queries ga4_traffic_acquisition_daily filtering sessionDefaultChannelGroup = 'AI Assistants' and aggregates by period.

Do AI sessions convert better than organic traffic?

Ask: "Compare engagedSessions and conversion rate between 'AI Assistants' and 'Organic Search' channels in the last quarter."

The LLM cross-references engagement metrics and transactions between both channels.

Which landing page receives the most AI traffic?

Ask: "Which landingPagePlusQueryString had the most sessions from sessionSourceMedium containing 'chatgpt' or 'perplexity' last month?"

The LLM queries ga4_landing_page_analysis filtering sessionSourceMedium.

How to Cross-Reference GA4 with Your ERP to Close the Sales Loop?

The revenue GA4 records (totalRevenue) does not always match invoiced orders in your ERP due to:

• Timezone differences between platforms
• Cancelled or refunded orders that GA4 does not immediately deduct
• PIX or boleto purchases that only confirm after payment

For real billing data, cross-reference with your orders table:

Ask: "Compare GA4 transactions from the 'AI Assistants' channel with the count of orders from your ERP in the same period - what is the percentage gap?"

Tables involved:

• ga4_ecommerce_revenue_daily: transactions, purchaseRevenue filtered by channel
• orders (ERP): issue_date, total_value aggregated by day

The LLM performs the JOIN by date and calculates the divergence, helping identify whether AI traffic generates real sales or just purchase events in GA4.

Interpretation Pitfalls to Avoid

When analyzing GA4 data via MCP, three technical nuances can lead to wrong conclusions:

1. bounceRate is 0-100, not 0-1: a value of 70 means 70% bounce (high), not 0.70 (low)
2. totalRevenue includes subscriptions and ad revenue: for pure e-commerce, use purchaseRevenue
3. GA4 data takes up to 48 hours to finalize: questions about "yesterday" may be incomplete

Why Use Kondado Instead of Analyzing GA4 Directly?

GA4 has its own analysis interface, but three limitations make it impractical for frequent cross-referencing:

1. Manual export or BigQuery required - no unified API for international ERPs
2. Limited query quota - exploratory analyses via native API deplete quickly
3. No cross-platform JOIN - does not automatically cross-reference with ERPs, ad platforms, and CRMs

Frequently Asked Questions

Does GA4 separate ChatGPT, Claude, and Perplexity traffic automatically?

No. By default, GA4 dumps every AI source into the generic "Referral" channel, mixed with any other site that links to you. You need a Custom Channel Group with a regex over the chatbot domains to isolate that traffic in the sessionDefaultChannelGroup dimension.

Which AI domains should I include in the Custom Channel Group regex?

The main ones are chatgpt.com, perplexity.ai, claude.ai, gemini.google.com, and copilot.microsoft.com. The full regex is chatgpt.com|perplexity.ai|claude.ai|gemini.google.com|copilot.microsoft.com, applied to the Source condition.

Do I need an ERP pipeline to see AI-driven revenue?

No. If the purchase event fires correctly in GA4, purchaseRevenue in ga4_ecommerce_revenue_daily already answers how much AI generated. The ERP comes in to close the loop: comparing GA4 declared revenue against invoiced orders from Bling!, Omie, or Tiny ERP.

How do ChatGPT or Claude query my data via Kondado MCP?

The MCP server URL is https://mcp.kondado.io/mcp. You add this URL in the ChatGPT or Claude MCP server settings, authorise via OAuth at app.kondado.com.br, and the session is bound to your Via Kondado, with read-only queries over the replicated data.

Why doesn't GA4 revenue match my Bling or Omie orders?

Three common reasons: timezone differences between platforms, cancelled or refunded orders that GA4 does not deduct immediately, and PIX or boleto purchases that only confirm after payment. Crossing GA4 with the ERP orders table shows the real percentage gap.

How often is GA4 data refreshed in Kondado?

At the frequency you choose at the pipeline level. Since GA4 itself settles data with up to 48 hours of delay, more frequent syncs add no practical benefit for revenue analysis; typical intervals are hourly to daily.

Which company sizes and markets does Kondado serve?

Kondado serves SMBs, agencies, ecommerce operations, ERP-driven teams, and mid-market to enterprise organisations, in any market. Data sources like Google Analytics 4, Meta Ads, Google Ads, HubSpot, and Salesforce are global.

Kondado replicates data from 80+ sources into a single SQL destination, where the MCP enables unified natural language queries, including AI assistants, ERPs, advertising platforms, and CRMs. Start now with a free trial.