track ai search traffic in ga4 — how to build a defensible proxy

Track AI Search traffic in GA4 without giving your CFO bad numbers 👍💡

You can’t “Track AI Search traffic in GA4” out of the box. GA4 rolls Google’s AI Overviews and AI Mode clicks into Organic Search, and it can mis-credit last click if a redirect hop or app webview strips tagging or referrer. Default channel groups • Campaigns and traffic sources

I watched Skale AI Search’s take and I agree. If you want decision-grade clarity, you have to instrument it. No drama, just the work. Here’s the proxy model I run for clients and my own sites: tag likely AI-exposed pages, capture redirect hints, cohort them in BigQuery or Looker, then triangulate with Search Console CTR shifts. Freeze the definitions so finance can read the trend without you apologizing on slide 3, because nothing kills trust faster than changing the yardstick mid-quarter.

1. 01
   1) Classify landing pages

   Create a BigQuery view that tags pages by intent (how-to, compare, local, YMYL). These are most exposed to AI Overviews. Keep the taxonomy small so it survives audits.

2. 02
   2) Catch redirect hints

   In GA4/BigQuery, work with the full page_location and page_referrer. Flag sessions containing known Google redirect patterns or parameters (e.g., google.com/goto, google.com/url, news.url.google.com). Treat as hints, not proof. Register page_location as a custom dimension if you want it in standard GA4 reports.

3. 03
   3) Build an “AI-influenced” cohort

   In Looker/BigQuery, cohort sessions that start on AI-exposed pages AND show redirect hints OR anomalous referrer behavior (e.g., missing referrer on first hit but google.com on second). Compare against a stable pre-period.

4. 04
   4) Triangulate with GSC

   Track CTR deltas and impressions for those pages. Rising impressions + falling CTR + stable rank often means answers on-SERP. Use the new Generative AI report for visibility and standard Web data for CTR.

5. 05
   5) Stabilize reporting

   Freeze your definitions, annotate timeline, and report the same cohort every week. Accuracy beats granularity early. Share the caveats in the deck footnotes, not the headline.

**Stop guessing.** Freeze a cohort definition, report it weekly, and annotate every model change. Finance will forgive coarse if it’s consistent and audited.

My read on the claim

Skale’s core point is right. GA4 is not a reliable lens for AI surfaces. It’s sessionized, query-blind without GSC, and channel grouping rolls most Google inflows into the same bucket. Add Google’s habit of intermediate redirects and parameters and you get attribution smear. Some of that includes the now-common google.com/goto-style hops; early tests show Organic still classifies correctly because the referrer remains google.com, but hops can still alter session starts or drop context in some flows. Google confirms goto • Community test.

Two implications:

// My reaction

Surprised — AI shifts attribution

• You won’t “see” AI Overviews traffic as a distinct source. You’ll see google/organic or even Direct when referrers drop. 📊
• Zero-click lift is invisible in GA4. Your losses show up as lower CTR in Search Console, not missing sessions in GA4.

Why GA4 misses AI search

A few unsexy reasons, all fixable with better instrumentation:

• No query context in GA4. Without GSC, GA4 can’t show which prompts or intents fed a session. And GSC does not label AI Overviews separately.
• Redirects and app surfaces. AI answers, Discover, and app taps often hit Google-controlled hops. Referrer or UTM can get lost, and GA4 may bucket as Direct or generic Organic. 🧪
• Session rules. GA4’s attribution and sessionization can over-credit the last touch that survived the hop, not the AI surface that started it.
• Answer-first UX. AI surfaces resolve intent in-SERP. You measure absence, lost clicks and shorter trails, not presence. That demands triangulation, not a single dashboard.

If you haven’t read my take on telemetry drift, start here. AI Mode is taking over, what to fix before your SEO data lies to you (read it). And if you’re chasing the redirect mess specifically, I broke down the analytics fallout from Google’s "goto" parameter (details).

What I’d implement this week

You don’t need perfect. You need a defensible proxy you can explain on one slide.

1. 01
   1) Classify landing pages

   Create a BigQuery view that tags pages by intent (how-to, compare, local, YMYL). These are most exposed to AI Overviews.

2. 02
   2) Catch redirect hints

   Add a GA4 event param and custom dimension to capture page_location querystrings. Flag sessions containing known Google redirect patterns or parameters. Treat as hints, not proof. 🧰

3. 03
   3) Build an “AI-influenced” cohort

   In Looker/BigQuery, cohort sessions that start on AI-exposed pages AND show redirect hints OR anomalous referrer behavior. Compare against a stable pre-period.

4. 04
   4) Triangulate with GSC

   Track CTR deltas for those pages. Rising impressions + falling CTR + stable rank often means answers on-SERP. 📊

5. 05
   5) Stabilize reporting

   Freeze your definitions, annotate timeline, and report the same cohort every week. Accuracy beats granularity early.

If you need a starting point on the redirect piece, I laid out ways to stabilize analytics before AI surfaces steal intent (stabilize). And if leadership asks “are we even visible in AI?”, run a quick pass with my AI Visibility Checker (tool). It’s not magic, it just gives you a shortlist to validate.

// My reaction

Frustrated — redirects breaking metrics

Proving ROI without pretending

Here’s how I’d brief a CFO without overpromising:

• Define the proxy upfront. “AI-influenced sessions” equals cohort rules you control. Source of truth? No. Decision-grade? Yes.
• Show direction, not decimals. Display three lines: GSC CTR for exposed pages, GA4 sessions for the cohort, and conversions per session. Keep it weekly. 🧵
• Attribute conservatively. Credit assist, not last click. If AI reduces friction earlier, last-click channels will steal credit. Say that plainly.
• Quantify opportunity. Pair the cohort with a bottom-up estimate of lost-click value. My SEO ROI Calculator keeps this honest (calculator).

What I agree and disagree with in the video

• Agree: GA4 won’t hand you AI Search out of the box. You must build an attribution scaffold you trust.
• Agree: Pattern analysis beats detector theater. Don’t waste cycles on “AI detection” gadgets for traffic. Use logs, query patterns, and cohorts.
• Pushback: Treat any single parameter, including “goto”, as a clue, not a fingerprint. The web changes. Your model should, too. 🚦

Bottom line

Skale’s right. GA4 won’t solve AI Search for you. With a few sane proxies and clean cohorts, you can measure enough to decide what to fix next, and stop guessing. If you need a primer on the telemetry gotchas, start here (AI Mode is taking over). 📊

Takeaway: this is classic ops work, define the signals, instrument them, iterate. That’s how you keep your SEO or GEO or AI practice honest and useful, not performative. 🧭