GA4 for AI Search: A Monthly Measurement Routine

#01Before the first run

Set these up once. Each links to the lesson that covers it.

#02The monthly checks

1. AI traffic headline (10 minutes). Open Reports → Acquisition → Traffic acquisition, set last month vs the previous month, and read the AI Assistant row of the default channel group. Then switch the primary dimension to your custom channel group and add Session source as a secondary dimension.

2. Sources you aren't catching (10 minutes). Still in Traffic acquisition, set the dimension to Session source / medium and search for ai, gpt, chat, copilot and assistant. Any assistant-like source sitting in Referral or Unassigned is a candidate for your regex. Edit it in Admin → Data display → Channel groups. Custom groups apply to past data, so the fix is retroactive.

3. Landing pages (15 minutes). Use a free-form exploration (see Free-form explorations and segments):

Save it once and reuse it. The short list it returns is the set of pages AI users actually reach, so keep them accurate and current.

4. Outcomes (10 minutes). In the same exploration, add Event name as a column filtered to your key events. Compare the AI channel's key event rate with Organic Search. With small volumes, report counts and rates together and avoid claiming a precise difference.

5. Search Console and demand (15 minutes). In the Search Console report collection (lesson 15) or Search Console itself, check branded query clicks and impressions month over month. Rising branded demand alongside flat AI referrals is a common pattern when people read an AI answer and then search for you.

#03Optional: the BigQuery check

If export is on, this query counts sessions by assistant-like source for a month. Replace the project and dataset.

SELECT
  collected_traffic_source.manual_source AS utm_source,
  REGEXP_EXTRACT((SELECT value.string_value FROM UNNEST(event_params) WHERE key = 'page_referrer'), r'^https?://([^/]+)') AS referrer_host,
  COUNT(*) AS session_starts
FROM `your-project.analytics_123456789.events_*`
WHERE _TABLE_SUFFIX BETWEEN '20261001' AND '20261031'
  AND event_name = 'session_start'
GROUP BY 1, 2
HAVING REGEXP_CONTAINS(LOWER(CONCAT(IFNULL(utm_source, ''), ' ', IFNULL(referrer_host, ''))), r'chatgpt|openai|perplexity|gemini|copilot|claude|deepseek|grok')
ORDER BY session_starts DESC;

Referrer hosts here are the raw truth your regex should match. Remember that sandbox tables expire after 60 days, so run it before last month's tables disappear.

#04The monthly table

Copy this into a sheet and add one row per month.

Illustrative example. Ledgerline, a fictional finance software company, runs this on the 4th of each month. In one review it finds a new assistant source in Referral, adds it to the regex, and notes the change in the table so the jump in the custom channel isn't misread as growth.

#05What GA4 can't tell you

GA4 measures visits. It can't show how often an assistant mentions or recommends you without a click, or what it says. Pair this routine with prompt tracking, and with "How did you hear about us?" on forms, as covered in self-reported attribution and AI visibility KPIs for marketing teams. For assistant-specific tracking see measuring brand visibility in Claude and Tracking ChatGPT referral traffic and mentions.

Our view: report AI referrals as one leading indicator among several. Treating them as the total effect of AI search undersells it.

You have finished the track. Return to the GA4 for Marketers track.

#06Common questions

How often should I review AI traffic in GA4? Monthly is enough for most teams. Volumes are usually small, so weekly numbers swing too much to act on.

Why does my AI Assistant number differ from my custom channel? They use different rules. The default AI Assistant channel uses Google's own list of assistants; your custom group uses your regex. Report one consistently and note which.

Does GA4 show traffic from AI Overviews? Not separately. Google counts AI Overviews and AI Mode visits as Organic Search, so use Search Console and your own prompt tracking alongside GA4.