Search Console and GA4 on AI Traffic: What Google's Own Documentation Says Each One Counts

Search Console counts a click on a link inside an AI Overview or inside AI Mode and files it with every other Google click. GA4 counts the visit that follows and classifies it from the referrer the browser sends. Neither product records the answer itself, which is the part a buyer read. Everything here comes from Google's own help documentation, read on 16 October 2026, and from one file Google publishes for download: the spreadsheet of default-channel-group sources, which has 819 rows, 129 of them categorized as search, and not one of those 129 is an AI assistant.

What Search Console counts when an AI Overview names you

Google's Search Central guide to AI features states that sites appearing in AI features "are included in the overall search traffic in Search Console" and that "they're reported on in the Performance report, within the 'Web' search type". The counting rules sit in the Search Console documentation, and they are specific enough to work with.

Surface	What the documentation says is recorded
AI Overviews	Standard impression rules apply, and clicking out to an external page counts as a click. The Overview occupies a single position, which every link inside it is assigned.
AI Mode	Clicking a link to an external page counts as a click; standard impression rules again. Position is worked out the way it is on a results page.
A follow up in AI Mode	The user is "essentially performing a new query", and all impression, position and click data are counted as coming from that new query.
Search Labs experiments	Not included in Search Console data.

Two of those lines change how a report reads. Because an Overview holds one position and hands it to every link inside, a page cited third shows the same position value as a page cited first. And because a follow up in AI Mode becomes a new query, one conversation can appear as four unrelated queries. Average position was already blended, as our Google Search Console guide sets out, and this is a second reason to watch the trend rather than the digit. How ranking metrics got here is in what replaced the PageRank score and what to measure instead.

The filter the report does not offer, and what to do instead

Count the values in the search type filter and you get five: Web text-based, Web multimodal, Image, Video, News. That was the whole list on 16 October 2026, and AI features sit inside the first entry with the plain blue links. So treat Web text-based as one pooled total and go elsewhere for answer level evidence. Qualifying is ordinary, by Google's own account: the page has to be "indexed and eligible to be shown in Google Search with a snippet" and nothing more, since the guide follows that clause with "there are no additional technical requirements" and rules out new machine readable files and markup by name. One more documented habit is worth planning around. The guide says an Overview appears when the system judges it "additive to classic Search" and so "often don't trigger", which makes a quarter the shortest honest comparison window.

What GA4 does with a click from ChatGPT or Perplexity

GA4 builds its default channel group from rules you can read. Direct is source exactly matching "(direct)" with medium one of "(not set)" or "(none)". Referral is medium being one of "referral", "app" or "link", described as traffic arriving "via non-ad links on other sites/apps". Organic Search is decided from a list Google maintains and publishes as a downloadable spreadsheet. We fetched that spreadsheet on 16 October 2026 and counted the whole file.

Category in Google's file	Rows
Search	129
Social	592
Shopping	52
Video	46
All rows	819

The 129 search rows name the engines you would expect, and we checked for the obvious ones to be sure the file had loaded: Bing, DuckDuckGo, Yahoo with more than twenty country hosts, Baidu, Ecosia, Qwant, Naver, Yandex, Sogou, AOL and Daum are all in there. ChatGPT is not. Neither is chat.openai.com, perplexity.ai, claude.ai, copilot.microsoft.com or gemini.google.com. So a browser click arriving from an assistant with its referrer intact lands in Referral, next to blogs and news sites, rather than in Organic Search. Google publishes an address for requesting an addition to that list and says requests are reviewed at least once a year and often more frequently, so re-read the file each quarter rather than trusting this count forever.

Why a lot of assistant visits arrive with no source at all

Google's help page on (direct) / (none) traffic lists the documented causes, and several describe an assistant click exactly. Links without UTM parameters lose their source information. Redirects strip parameters from the URL. Shorteners strip the referral details. And traffic from "links from offline documents such as PDFs and Word documents" is classified as (direct) / (none). A desktop assistant application sits in that family: it is a program rather than a web page, so there is frequently no referring URL to pass on. The same page adds ad blockers, which matters because the audience most likely to use an assistant is also the audience most likely to run one. The tagging method for links you do control is in the Campaign URL Builder guide, and the limit is simple: nobody can put a UTM tag on a sentence inside somebody else's answer.

Four workarounds that use only what you already have

• Build a custom channel group for assistants. GA4 defines a channel from source and medium conditions, so make one matching the assistant hostnames, one condition per host, kept separate from Organic Search so your search trend stays comparable.
• Segment (direct) / (none) by landing page. Direct traffic to a home page is mostly people typing your name. Direct traffic to a deep page answering one narrow question is the pattern worth reading.
• Export the Performance report monthly and keep the files. The window rolls, so the comparison you will want in a year only exists if you saved it. Export at query level and at page level, with the date in the filename.
• Read your server logs for the assistant crawlers. The operators publish their agent names. On 16 October 2026 OpenAI documented four, GPTBot, OAI-SearchBot, ChatGPT-User and OAI-AdsBot; Anthropic documented ClaudeBot, Claude-User and Claude-SearchBot; Perplexity documented PerplexityBot and Perplexity-User. The method is in how to log and monitor AI crawler visits.

The fifth move is the one the other four point at. If the answer is where the recommendation happens, read the answer. Ask the questions your buyers ask with your own name in none of them, record which companies get named and which domains get cited, and set that count beside your analytics. Our own two rounds of exactly that are public. The first, dated 15 October 2026, kept 597 of 612 attempted readings and logged 10,652 citations: the 597 reading benchmark of ChatGPT and Gemini in three markets. The second, a day later, asked questions built in eight fresh directions and produced 1,260 readings with 20,143 citations over 4,947 domains. Two engines in both rounds, so Perplexity, Claude and Google AI Overviews sit outside what either can speak to.

Questions about AI traffic in Search Console and GA4

Does Search Console show AI Overviews separately from blue links?

One pooled Web figure is what the report hands over, because that is where Google puts the AI features and the overall totals. Read it as a trend line, and run your own question set for anything at the level of a single answer.

Why does ChatGPT traffic appear as Referral instead of Organic Search in GA4?

Because Organic Search is decided from Google's published source list, and that file, read on 16 October 2026, has 129 search rows and no assistant among them. Medium "referral" sends the visit to Referral instead. A custom channel group fixes the reporting in an afternoon.

Can I identify AI assistant visits that arrive as direct traffic?

You can narrow them considerably. Filter (direct) / (none) to deep landing pages, exclude branded entry points, and compare the shape against the questions those pages answer. That is an estimate with a method attached, which beats a round figure with nothing behind it.

Do I need a special file or markup to appear in AI Overviews?

The guide rules that out twice over, once by saying no extra technical requirement exists beyond ordinary indexing and snippet eligibility, and once by naming the things you do not have to build: new machine readable files, AI text files, markup. Put the hours into crawler access and into a page whose first paragraph answers the question.

The row your acquisition report is not going to grow

A buyer who asks an assistant for a shortlist, reads three names and never clicks has made a decision your analytics has no field for. Search Console will tell you about Google's own AI surfaces, accurately and for free. GA4 will tell you about the visits that arrived. The named shortlist in between decided the deal, and it gets recorded only when somebody asks the question and writes down the answer. That is the product we sell: AI Knows Us puts buying questions to ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews without naming your business, prints the sources behind each answer, and writes the pages those answers reach for. Our own score from the 16 October 2026 run was 6 readings out of 1,260 that cited us, published because a vendor who hides its worst number has told you how to read all the others.