How to Track AI Traffic in Google Search Console

You can now track AI traffic in Google Search Console two ways: through the dedicated Generative AI performance reports Google launched on 3 June 2026, and through the query-level detection method that fills the gap those reports leave open. This is the measurement layer beneath our AI search strategy for South Africa and the practical companion to our first-party traffic analysis.

The short version of why you need both: the official report tells you how often you appeared, but not for which questions. The detection method tells you which questions — which is the half you can actually act on.

What Do Google's New Generative AI Reports Actually Show?

They show impressions inside Google's generative AI features, isolated from classic search results for the first time. Google's Search Central announcement describes dedicated views covering generative AI features in Search — AI Overviews and AI Mode — plus generative AI features in Discover, reporting how often URLs from your site appeared in them.

Three caveats matter before you build reporting on it. The data is not new: Google confirms these impressions were always counted inside your overall performance totals, so your headline numbers haven't changed and historical comparisons remain valid. The reports launched to a subset of websites for testing rather than everyone at once. And the current view is impressions only — no clicks, no click-through rate, no queries.

That last limitation is the important one strategically. Knowing you appeared 40,000 times inside AI answers is useful for proving the surface exists to a sceptical board. Knowing which questions produced those appearances is what tells you what to write next — and that data isn't in the report.

Google shipped one more control alongside the reports: a setting letting sites block their content from appearing in AI features. Google has been explicit that sites opting out receive neither traffic nor impressions from those surfaces — which makes it a decision worth taking deliberately rather than discovering by accident, and one most lead-generation businesses should think hard about before touching.

How to Track AI Traffic When the Report Has No Query Data

Use the standard Performance report and hunt for the machine-reading signature: unusually long, hyper-specific queries sitting at strong average positions with substantial impressions and no clicks at all. Humans don't type forty-character natural-language questions and then decline to click the top result; retrieval systems do exactly that.

Our working detection filter is four conditions applied together — six or more words in the query, average position six or better, fifty or more impressions, and zero clicks. Applied to Growth Pulse Media's own 382-post site over one 28-day window, that pattern isolated eleven queries carrying 11,388 impressions, roughly 7.8% of the site's total.

Building it takes about ten minutes. Open Performance, set the date range to the last 28 days, switch to the Queries tab, then sort by impressions descending and scan the top rows for entries with zero clicks. Export to a sheet if you want the word-count and position conditions applied properly — the eye-scan version catches the biggest offenders, but the filtered version is what you track month to month.

Run it monthly and treat the output as a content brief. The questions appearing in that filter are the questions your site is already trusted to answer, which makes them the cheapest citation wins available — sharpen those pages first, using the properties in our guide to content AI systems quote.

Be clear about what the filter can't prove. Zero-click long-tail queries are strong circumstantial evidence of machine reading, not a labelled confirmation from Google, and some will be genuine human searches that simply didn't convert to a click. Treat the pattern as a well-founded inference and the trend across months as the real signal — that's the honest framing, and it holds up when a client interrogates it.

Why Does the Device Split Matter?

Because machine reading registers overwhelmingly as desktop, so a widening gap between your mobile and desktop click-through rates is corroborating evidence. Humans in South Africa browse and click from phones; automated retrieval doesn't.

In our own 28-day window, mobile delivered 383 of 848 clicks — 45% of clicks from just 25% of impressions — producing a mobile CTR of 1.04% against 0.42% on desktop. Neither number is impressive in isolation. The ratio between them is the signal, and it's free to check in two clicks.

Log both figures monthly alongside the query count. Three numbers — machine-query share, mobile CTR, desktop CTR — take five minutes to pull and give you a trend line that no dashboard sells and most SA agencies aren't producing.

How Do You Track Assistants Outside Google?

Search Console only sees Google. For ChatGPT, Perplexity, Gemini and the rest, the instrument is your analytics referral report — visits arriving from assistant domains show up as referral traffic, and segmenting them reveals both volume and behaviour.

Expect small numbers and unusually good behaviour. Assistant referrals arrive pre-qualified because the shortlisting already happened inside the answer, which is consistent with Semrush data putting AI-referred conversion at roughly 4.4 times organic rates. Judge this channel on conversion quality and pipeline, never on session volume.

Then close the loop where analytics can't reach: the monthly prompt log from our five-layer visibility audit, plus the intake question that costs nothing. "How did you come across us?" answered verbatim catches the referrals that never carried a referrer at all — the buyer who read an answer, remembered the name, and typed it into their browser three days later.

What Should a Monthly AI Reporting Sheet Contain?

Six lines, and it should fit on one page. Anything longer gets skipped by the person who most needs to read it.

Assign an owner and a fixed day. The reason most teams never track AI traffic properly isn't difficulty — it's that the job belongs to nobody in particular, so it happens in bursts of enthusiasm and then stops. First Monday of the month, one person, twenty minutes.

Note what's absent: total sessions. Including it invites the conversation this whole discipline exists to move past, and the six lines above tell a truer story about a channel where impressions rise while clicks stay flat.

What Does Disciplined Tracking Change?

It changes which content gets funded, because you finally know which pages machines already trust. The composite below reflects the typical shape for an SA firm running this measurement stack for three months before acting on it. Figures illustrate the pattern, not a single client's audited results.

One caution on reading that table: the ability to track AI traffic doesn't itself produce citations, and a team that measures diligently while changing nothing will book three months of very well-documented stagnation.

The second row is the mechanism. Measurement doesn't create citations — it redirects effort toward the pages already halfway there, which is why tracking pays before any new content gets written.

The GPM Differentiator: The Method Came From Our Own Console

The detection filter in this guide wasn't borrowed from an overseas blog — we built it looking at Growth Pulse Media's own Search Console, and every number quoted above comes from that account. Our AEO services for South African businesses run the same stack for clients, including the unglamorous parts: the verbatim intake question and the one-page sheet.

We publish it because measurement is the honest half of this industry. Any agency can claim AI visibility work is going well; far fewer will show you the console behind the claim, and fewer still will tell you which of their instruments can't see what.

Who This Is NOT For

Questions SA Marketers Ask About Measuring AI Visibility

Can I track AI traffic in Search Console yet?

Partly. Google launched Generative AI performance reports on 3 June 2026 showing AI Overview and AI Mode impressions by page, country, device and date, rolled out to a subset of sites first. There's no click, CTR or query data, so query-level detection in the standard Performance report remains necessary for knowing which questions drive your appearances.

Why does the Generative AI report show no clicks?

Google's launch announcement describes the reports as impression-focused views of visibility inside generative AI features. Click and query data aren't part of the current version. Pair the impression view with analytics referral segments for the click side, and with the detection filter for the query side.

What exactly is the machine-query detection filter?

Four conditions applied together in the standard Performance report: queries of six or more words, average position six or better, at least fifty impressions, and zero clicks. Queries meeting all four are almost always retrieval systems reading your page rather than people declining to click a top result.

Did my traffic numbers change when the new reports launched?

No. Google confirmed the AI impressions were already included in the overall performance report and continue to be counted there — the June 2026 launch added a separate view rather than new data. Your historical comparisons and aggregate totals remain valid.

How do I see traffic from ChatGPT or Perplexity?

Through your analytics referral report, segmented by the assistant domains. Expect low volume and high quality — AI-referred visitors convert at roughly 4.4 times organic rates per Semrush data. Add a verbatim "how did you find us" intake question to catch the referrals that arrive with no referrer at all.

How often should I run this?

Monthly for everything, with the prompt log on the same schedule. AI answers rebuild from live retrieval, so standings shift between quarters, and the sheet only becomes useful once you have three or four months of trend to read rather than a single snapshot.

Still reporting AI visibility as a hunch? The instruments exist now, they're free, and the first run usually takes about twenty minutes — which makes "we don't have the data" a choice rather than a constraint.

Dirk van Greuning

Founder of Growth Pulse Media and a specialist in South African search dominance. Dirk translates his experience in scaling South African businesses into high-velocity digital strategies for B2B and retail leaders. He writes about SEO, lead generation, and paid media from an operator's perspective — prioritising pipeline value over impressions.

Connect with Dirk on LinkedIn