Learn › Episode 2

How to Connect AI Directly to Google Analytics (No Exports)

Ask Google Analytics questions in plain English and get answers in seconds, without exporting a single file. Then build the Episode 1 dashboard straight from live data.

How a connector works

Claude can use connectors: add-ons that let it read data from other services. A Google Analytics connector signs in with your permission, and when you ask a question, Claude requests exactly the numbers it needs. Reading reports doesn't change your Google Analytics, and viewer access to the property is enough. Under the hood, connectors use an open standard called the Model Context Protocol (MCP).

Three ways to connect

1. A ready-made connector (easiest). Supermetrics is listed in Claude's connector directory, and Porter Metrics can be added as a custom connector. Both offer free trials, with paid plans after that. Check their current pricing and terms.
2. Google's Analytics MCP server (free, technical). It's read-only and labeled experimental, and setup means installing Python, creating a Google Cloud project and running commands in a terminal. It works with Claude Desktop and Claude Code.
3. A custom build on the Google Analytics API, for dashboards that refresh on their own. That's real engineering work, and it's what I build for clients.

This lesson isn't sponsored, and the steps are similar for any connector.

Step 1: Add the connector to Claude

1. In Claude, open Customize, then Connectors.
2. If your connector is in the directory, search for it and click Connect. If it gives you a web address instead, click Add, choose Add custom connector, give it a name, paste the address and click Add.
3. Sign in when asked: most connectors have you sign in to their service, connect your Google account and choose which Analytics properties to share.
4. In a chat, click +, choose Connectors, and make sure yours is switched on.

The free Claude plan allows one custom connector; paid plans allow more (Claude Help Center).

Step 2: Ask good questions

• "How many sessions did the website get last month, compared with the month before?"
• "Which landing pages grew the most last month, and did any stall?"
• "Which channel brings the most engaged visitors? Use the last 90 days."
• "Show sessions by week for the last eight weeks, as a table."

Four habits for better answers: name the property and exact date range, ask for a comparison, ask for a table when you want to scan or paste results, and ask follow-ups like "now break that down by device." Watch for a city or page called "(not set)": that's Google Analytics saying it couldn't determine the value.

Step 3: Build the dashboard, no exports

Using my Google Analytics connector, pull data for the [property name] property for the last 12 months:
- Daily sessions, new users and engaged sessions by session default channel group
- Sessions and engaged sessions by landing page, by device category and by city

Then build a website traffic dashboard as a single HTML file with all the data inside it and plain SVG charts. Include four key numbers compared with the previous period, sessions over time, channels, top 10 landing pages, devices and top 10 cities, with date and channel filters. Open on the last 90 days.

When you're done, tell me the total sessions and engaged sessions you pulled, and the exact date range.

If the chat struggles with a full year of daily data, ask for weekly totals or the last 90 days. The Episode 1 prompts still work for filters and design.

Check the numbers

Ask the same thing two ways (total sessions, then sessions by channel added together) and compare with the same date range in Google Analytics. Small gaps can come from time zones, today's partial data or rounding. Big gaps mean you should ask Claude exactly which dates and filters it used.

Privacy and trade-offs

• A third-party connector can read your analytics data. Check its privacy terms and share only the properties you need.
• Some connectors can also change settings. Claude should ask before each change; say no unless you meant it.
• Free plans usually limit how much history you can pull.
• If you manage analytics for a client or employer, check with the account owner first. You can remove access any time in Claude's connector settings and your Google account's security settings.

Sources: Claude Help Center: custom connectors · Google Analytics MCP server · Porter Metrics guide · Supermetrics for Claude