Google now measures every AI platform except its own
By: MDG
Google Analytics added an AI Assistants channel in May 2026 that tracks referrals from ChatGPT, Gemini, Copilot, Grok, and DeepSeek. Google’s own documentation says it excludes AI Overviews and AI Mode, which stay filed under Organic Search. We are seeing multi-location groups hold their rankings, watch impressions climb and lose call volume, with no report that explains it.
Key Facts About Measuring AI Search Visibility
- GA4’s AI Assistants channel, released May 13, 2026, buckets referrals from ChatGPT, Gemini, Copilot, Grok and DeepSeek. Google’s help documentation states it “excludes Google’s AI Overviews and AI Mode.”
- Search Console’s generative AI performance report, live for all sites worldwide as of August 31, 2026, reports impressions only. No clicks, no click-through rate, no average position, and no split between AI Overviews and AI Mode.
- When an AI summary appeared, Pew measured clicks on a traditional result falling from 15% of visits to 8%. Clicks on a source link inside the summary happened on 1% of visits.
- 68% of United States Google searches in the first four months of 2026 ended without any click. The share producing at least one click fell 22.9% compared to 2024.
- Referred traffic from AI platforms is still roughly 1% of sessions on most sites. The measurement problem is not the referral you can count. It is the answer that resolved without one.
Does Google Analytics track ChatGPT traffic?
Yes, and as of May 2026, it does so automatically. GA4 now assigns a medium of ai-assistant and files the session under a default channel called AI Assistants when the referrer matches Google’s maintained list of AI platforms. ChatGPT web sessions also arrive carrying a utm_source=chatgpt.com parameter, so the traffic was visible before the channel existed if you knew to look for it.
What matters is the sentence directly underneath it in Google’s own channel definitions. AI Assistants is described as arrivals “from sources like ChatGPT, Gemini, Deepseek, Copilot, or Grok,” and then: “It excludes Google’s AI Overviews and AI Mode.” The Organic Search definition on the same page confirms the other half, describing it as arrivals via non-ad links in organic search results, “including Google’s AI Overviews and AI Mode.”
Read those two sentences together, and the situation is unambiguous. Google built a channel that isolates its competitors’ AI referrals and, by design and in writing, declines to isolate its own. A click from ChatGPT is now a labeled line item. A click from an AI Overview is Organic Search, indistinguishable from a blue link on page one.
For a multi-location healthcare group, this is not a reporting inconvenience. Google AI surfaces reach far more patients than every standalone platform combined, which means the channel carrying the most AI-influenced demand is the one channel that cannot be broken out.
Does Search Console show AI Overview clicks?
No. It shows impressions and nothing else.
Google announced the generative AI performance report in June 2026 and completed the worldwide rollout on August 31. The help documentation describes it as showing “how many times links to your site were shown to a user in a generative AI feature.” AI Overviews and AI Mode are combined with no breakdown, and Search Labs experiments are excluded.
There is no clicks column. There is no click-through rate. There is no average position. Google’s AI features documentation states separately that sites appearing in AI features “are included in the overall search traffic in Search Console” under the Web search type, so the clicks that do occur are already blended into the number you have always looked at.
This is still a meaningful addition, and we would rather have it than not. It is the first time Google has published any site-level signal about AI surface presence. But an impressions figure with no click counterpart is a specific and unusual kind of data. It tells you that you were shown. It tells you nothing about whether being shown produced anything.
The Visibility Ledger
Most measurement conversations stall because participants are arguing about different tiers of a problem without saying so. It helps to separate them explicitly.
The Visibility Ledger is the accounting of AI-era demand in three tiers: what your analytics counts, what it can only imply, and what it cannot see at all.
| Tier | What sits in it | Where it shows up | What to do about it |
|---|---|---|---|
| Counted | Referred clicks from standalone AI platforms. ChatGPT, Perplexity, Claude, Gemini and Copilot all pass their domain as the referrer on web sessions. | GA4 AI Assistants channel, or a custom channel group if you want to extend Google’s list | Report it, trend it, and resist the temptation to celebrate it. It is small. |
| Implied | Presence inside Google AI surfaces. You know you were shown. You do not know what happened next. | Search Console generative AI impressions, trended against total clicks | Watch the gap between the two lines. A widening gap is the signal. |
| Invisible | Answers that resolved with no click at all. Native app sessions that strip the referrer. Phone calls placed after an AI answer named you. | Nowhere, directly | Proxy it: branded and near-me search volume, direct traffic composition, and call intake capture. |
The tiers are not equally sized, and the ordering is the uncomfortable part. Counted is the smallest and the easiest to report. Invisible is the largest and the one that actually moved. Marketing teams end up presenting the smallest tier because it is the one with a chart, which is how a board comes to believe AI search is a rounding error.
How much traffic actually comes from AI platforms?
Around 1% of sessions today, and the growth rate is the more useful number.
Previsible tracked 6.77 million sessions from AI platforms across 166 GA4 properties, held constant from November 2024 through May 2026. AI platform traffic ranged from 0.08% of total sessions for publishers to 1.71% for smaller businesses, with ChatGPT accounting for 92.4% of it. Monthly sessions from AI platforms grew roughly ten-fold over the period. The study explicitly excludes AI Overviews, which is to say it measures only the Counted tier. This is agency research, and we label it as such, but the panel is held constant, which is more methodological discipline than most published figures on this question carry.
Similarweb measured 117.4% year-over-year growth in AI referrals between June 2025 and May 2026 and states plainly that they “remain a low single-digit percentage of most sites’ total traffic.” Adobe Analytics, working from more than a trillion visits to United States sites, reported AI-sourced traffic up 138% in retail and 105% in financial services year over year, with AI-referred visitors showing 36% lower bounce rates and 53% longer time on site in retail. Adobe publishes no healthcare breakout in any quarterly edition, which is worth knowing before anyone applies a retail number to a patient journey.
Two multipliers circulating widely in this category are worth noting because they are being used to set budget and neither one traces to a study.
The claim that AI-referred visitors convert more than twenty times better than organic traffic comes from one software company reporting on its own website, with an audience that self-selects about as heavily as an audience can. The related claim that AI traffic is several times more valuable per visit appears in vendor summaries with no underlying dataset attached.
What is supportable is narrower and still useful. Visitors who arrive from an AI platform tend to arrive further along, having already had part of the question answered, and Adobe’s trillion-visit panel shows that in lower bounce rates and longer sessions rather than in a multiplier. That is the version of the claim worth putting in a budget conversation, because it is the version that survives being asked for the source.
Why do rankings hold while calls drop?
This is the pattern that brings marketing leaders into a room with their agency, and it now has a documented mechanism rather than a theory.
Pew Research Center tracked the actual browsing behavior of 900 United States adults across 68,879 searches in March 2025, of which 12,593 produced an AI summary. When a summary appeared, 8% of visits produced a click on a traditional search result. When one did not, 15% did. Clicks on a source link inside the AI summary occurred on 1% of visits. Sessions ended entirely on 26% of pages carrying an AI summary against 16% of those without one. The study is now eighteen months old, which matters, but it remains the only controlled behavioral panel published on the question.
SparkToro, working from Similarweb clickstream data, found 68.01% of United States Google searches ended without a click between January and April 2026. Searches producing at least one click fell 9.51 percentage points against 2024, a relative decline of 22.9%. The analysis excludes the Google mobile app, where the zero-click share is reportedly higher still.
Put the mechanism in plain terms. Your page still ranks. The AI summary above it answers the question. Search Console counts the impression. No click is generated, so no session appears in GA4, so nothing in your reporting registers a loss. The demand did not disappear. It was satisfied one layer above your website, and the layer that satisfied it is labeled Organic Search.
This is the same dynamic we described in our piece on how patient search behavior has shifted toward AI platforms, viewed from the reporting side rather than the acquisition side.
What AI-era metrics are worth reporting?
Five, and only one of them is new infrastructure.
| Metric | What it actually captures | Where it comes from | Honest limitation |
|---|---|---|---|
| Generative AI impressions against total clicks | Whether AI presence is rising while click yield falls | Search Console, two lines on one chart | Combines AI Overviews and AI Mode. No click attribution at all. |
| Citation share by platform | How often each platform names you against named competitors on your revenue queries | Systematic querying across ChatGPT, Google AI Mode, Perplexity and Gemini on a fixed schedule | Answers vary by session and location. Requires a consistent method to be comparable over time. |
| Branded and near-me search volume | Whether AI answers are creating awareness that resolves as a direct search later | Search Console query data, segmented | A proxy. Other things move branded search too. |
| AI platform referral sessions and outcomes | The Counted tier, with conversion attached | GA4 AI Assistants channel, extended with a custom channel group | Roughly 1% of volume. Do not let its size set the strategy. |
| Call intake source capture | The only direct read on the invisible tier | Call tracking plus a scripted intake question | Depends entirely on front-desk discipline across every location. |
The last one deserves more attention than it gets. In multi-location healthcare, the conversion is usually a phone call, and no analytics platform in existence sees an AI answer that ends with a patient dialing a number. A one-line question at intake, asked consistently across every site, produces better data on AI-sourced demand than any dashboard currently on the market. It is unglamorous, and it works.
Server log analysis is worth adding at the enterprise tier. Crawlers including GPTBot, OAI-SearchBot, PerplexityBot and ClaudeBot fetch pages that will never appear in GA4, and crawl frequency is a reasonable leading indicator of whether your content is even eligible to be cited. That is a lower-cost signal than most of what is sold as AI visibility tracking.
What the numbers do not show about patient behavior
Volume figures understate this shift badly, because in healthcare the highest-value AI interactions are the ones least likely to produce a click.
Pew found in August 2026, surveying 3,488 United States adults, that 34% use AI platforms for at least one health reason. A quarter use them to work out what is causing symptoms, 22% to understand a diagnosis, and 15% to decide whether to see a doctor. KFF found 32% in an independent poll of 1,343 adults fielded in early 2026, with adults under 30 using AI for mental health information at roughly three and a half times the rate of those over 50.
Two independent probability samples landing within two points of each other is about as solid as consumer research gets. And notice which behavior is most common: deciding whether to see a doctor. That is the exact moment a multi-location group wants to be named, and it happens entirely inside a conversation that generates no referral, no session, and no impression you can attribute.
The 1% referral figure is measuring the residue of that conversation, not the conversation itself.
Where to start
- Split the ledger before the next board meeting. Label every AI metric you report as Counted, Implied or Invisible. The exercise is quick, and it stops a 1% referral number from being read as the size of the opportunity.
- Chart generative AI impressions against total clicks. One chart, two lines, updated monthly. If impressions rise while clicks flatten, you are being read and not visited, and that is a content and citation problem rather than a ranking problem.
- Extend the GA4 channel group. Google’s AI Assistants list does not cover every platform. A custom channel group adds the ones it misses and keeps historical comparability when the list changes.
- Start measuring citation share on your revenue-driving queries. Fixed query set, fixed schedule, both major trust models tested separately. As we found looking at where AI platforms actually source information about clinicians, ChatGPT and Google AI Overviews cite very different kinds of sources in healthcare, so a single blended number hides the gap that matters.
- Fix intake capture. One question, asked the same way at every location, logged in the same field. It is the cheapest instrumentation available, and it reads the tier nothing else can.
Frequently asked questions
Does Google Analytics track ChatGPT traffic?
Yes. Since May 2026, GA4 assigns a medium of ai-assistant and files those sessions under a default AI Assistants channel covering ChatGPT, Gemini, Copilot, Grok, and DeepSeek. Google’s documentation states this channel excludes Google’s own AI Overviews and AI Mode, which continue to be reported as Organic Search.
Does Search Console show AI Overview clicks?
No. The generative AI performance report, available worldwide since August 31, 2026, shows impressions only. There is no clicks column, no click-through rate, no average position, and no separation between AI Overviews and AI Mode. Clicks from AI features are blended into overall Web search traffic.
How much website traffic comes from AI platforms?
Roughly 1% of sessions on most sites today. An analysis of 166 GA4 properties held constant from November 2024 to May 2026 found a range from 0.08% for publishers to 1.71% for smaller businesses, with ChatGPT accounting for 92.4% of it. Growth is steep, but the referral figure measures only the traffic AI platforms send, not the demand they absorb.
Why are rankings stable while calls are falling?
Because an AI summary can answer the query above your listing. Pew found that when an AI summary appeared, clicks on traditional results fell from 15% of visits to 8%, and only 1% of visits produced a click on a source inside the summary. Your impression is counted, the click never happens, and no report registers the loss.
How do you measure AI search visibility without referral data?
By combining five signals: generative AI impressions trended against total clicks in Search Console, citation share measured by querying each platform on a fixed schedule, branded and near-me search volume as an awareness proxy, AI platform referral sessions in GA4, and a source question captured at call intake. No single one is sufficient, and together they are workable.
Do visitors from AI platforms convert better than organic search visitors?
They appear to arrive further along, and the widely quoted multipliers are not supportable. Adobe’s panel of more than a trillion visits found 36% lower bounce rates and 53% longer time on site among AI-referred retail visitors. The multipliers circulating on conversion rate and visit value trace to one company’s report on its own website and to vendor summaries with no dataset attached.
Ask for an AI visibility measurement
MDG builds the measurement layer that sits under both paid and organic search: Search Console and GA4 read directly into our own reporting environment, citation share tracked per platform against your named competitors, branded search and intake capture wired into the same view, and an action plan attached to each gap rather than a dashboard handed over. It runs as one in-house methodology alongside the paid, local, and organic work rather than as a separate analytics engagement.
The build starts with a ledger audit: what your current reporting counts, what it only implies, and what it is missing entirely. We support thousands of client locations nationwide. Our approach is all informed by more than 25 years of doing multi-location, high-growth marketing for PE-backed companies.
Start the conversation here, or read more about our reporting and analytics work for multi-location healthcare organizations.