Discussion
  • AI Search

AI Mode clicks arrive in GA4 as google / organic. Has anyone found a defensible way to estimate what share of organic sessions actually came from AI Mode?

Published 1 reply

Discussion starter: an AI search question, posted with an answer from Niraj Raut to open the thread. If you have dealt with this on a site, reply with what you saw, especially where it differs.

Disclosure: Niraj Raut, who posted this answer, runs the SEO consultancy linked at the end of it.

1 reply

  • admin
    Admin
    Original poster

    There is no way to measure AI Mode sessions directly in GA4 today, and I would not trust any setup that claims to. What you can defend is a bounded estimate: take AI Mode impressions from Search Console, apply an explicit and deliberately wide click-through assumption, convert clicks to sessions using your own GA4 to Search Console ratio, and report the result as a range with the assumptions written next to it.

    The qualification is that the weakest input, how AI Mode click-through compares with classic results for your pages, is not published by Google. So the estimate is only as good as the range you put on that assumption. The useful question is not “what is the exact share?” but “does the decision we are making change anywhere across the plausible range?” If it does not, the estimate is good enough. If it does, you need corroborating signals before acting.

    Why GA4 cannot tell AI Mode apart from classic results

    When someone clicks a result on Google, the browser sends only the origin as the referrer: https://www.google.com/. The path and parameters of the results page (including whatever marks it as AI Mode) never reach your site. GA4 therefore classifies classic results, AI Overview links and AI Mode links identically as google / organic.

    A few related points trip people up:

    • GA4’s AI Assistant channel will not catch it. The default channel group added in May 2026 lists assistants such as ChatGPT, Gemini and Copilot as examples, and it excludes AI Overviews and AI Mode, and traffic without a referrer still lands in Direct. AI Mode clicks carry a google.com referrer, so they stay in Organic Search.
    • Historical data has a small gap. In May 2025, AI Mode links briefly carried noreferrer, so some clicks were recorded as Direct. Google fixed it within about a week. It is worth knowing if you are building a long baseline, but it does not affect current data.
    • Text fragments are not a tracking method. AI features often link with #:~:text= fragments, but browsers deliberately hide the fragment directive from page scripts, and fragments are never sent to the server. Custom scripts that claim to detect them should be treated with suspicion.

    What Search Console gives you to work with

    Search Console is where AI Mode is visible, although only partly. Google’s Performance report help documents the counting rules: “Clicking a link to an external page in AI Mode counts as a click”, “Position in AI Mode follows the same methodology as a Google Search results page”, and a follow-up question in AI Mode is treated as a new query, with its impressions, positions and clicks counted against that new query. AI Mode data sits inside the Web search type totals, with no separate click breakdown.

    The Search generative AI performance reports, available to all eligible sites since 31 August 2026, add AI impressions for AI Overviews, AI Mode and generative AI in Discover, filterable by page, country, device and date. They do not give clicks, CTR or queries.

    Data source Sees AI Mode? Clicks? Best use
    GA4 (google / organic) Mixed in, not separable Sessions only, all Google organic Denominator, and the sessions per click ratio
    Search Console Performance (Web) Included in totals Yes, but not split by feature Total clicks per page, query length patterns
    Search generative AI performance reports Yes, impressions No AI Mode exposure per page and date
    Manual or tool-based AI Mode prompt sampling Yes, visibility only No Checking which pages AI Mode actually links

    A bounded estimate you can defend in a meeting

    The approach below keeps every assumption visible. It uses a relative click-through multiplier rather than an absolute CTR, because your pages already have very different CTRs, and it is easier to argue about “AI Mode converts impressions to clicks at somewhere between a quarter and one and a half times the rate of classic results” than to invent a single AI Mode CTR.

    1. Export page-level data for the same period and country: Web clicks and impressions from Search Console, AI Mode impressions from the generative AI report, and organic sessions by landing page from GA4.
    2. Compute an AI Mode exposure share per page: AI Mode impressions divided by Web impressions. Google does not document exactly how the AI report’s counting lines up with the Web totals, so treat this as approximate and cap it below 1.
    3. Pick a range for the relative click-through multiplier (k). No public evidence pins this down, so choose a range wide enough that a sceptical stakeholder cannot call it narrow. The values in the example below (0.25, 0.75 and 1.5) are illustrative assumptions, not findings.
    4. Estimate AI Mode clicks per page: page clicks × (share × k) ÷ (share × k + (1 − share)). This splits the page’s known clicks between AI Mode and everything else in proportion to exposure, adjusted by k.
    5. Convert clicks to sessions using each page’s GA4 organic sessions divided by its Search Console clicks. That ratio absorbs consent mode gaps, bot filtering and session rules, which differ by site.
    6. Sum and divide by total GA4 organic sessions, once for each value of k, and report the low, middle and high results.

    Hypothetical example:

    A how-to page has 20,000 Web impressions and 1,000 clicks in a month. The generative AI report shows 2,000 AI Mode impressions, so exposure share is 10%. With k = 0.25, estimated AI Mode clicks are 1,000 × 0.025 ÷ (0.025 + 0.9), about 27. With k = 0.75, about 77. With k = 1.5, about 143. If GA4 records 0.9 organic sessions per Search Console click for that page, the range is roughly 24 to 129 sessions, or about 2.7% to 14% of the page’s organic sessions. All figures are invented to show the arithmetic.

    The width of that range is the honest answer. Presenting “7.7%” to a leadership team would imply a precision the data does not have.

    Signals that can narrow the range

    None of these is a measurement on its own, but each can tell you whether the low or high end is more plausible.

    • Query shape. AI Mode queries and their follow-ups tend to be longer and more conversational. Growth in long, question-style queries in Search Console on pages with high AI Mode exposure is consistent with AI Mode clicks. This is a practitioner heuristic, not a documented classifier, and anonymised queries hide much of the long tail.
    • Natural experiments. Compare pages with high and low AI Mode exposure before and after a known change, such as AI Mode’s launch in your market or a jump in AI impressions for one section. If clicks on high-exposure pages move with AI Mode impressions while low-exposure pages do not, that supports a higher k.
    • Self-reported attribution. For lead generation sites, adding “Google AI Mode” as an option in a “How did you hear about us?” field gives a rough cross-check on high-value conversions. Recall bias is real, so use it for direction, not size.
    • Prompt sampling. Running your priority topics through AI Mode and recording which of your pages are linked shows where exposure is concentrated. It says nothing about clicks, but it catches pages where the AI report and reality disagree.

    Methods I would not present as measurement

    • Regex channel rules on the Google referrer. There is nothing in the referrer to match.
    • Engagement “fingerprints” (for example, assuming longer sessions on deep pages must be AI Mode visitors). This is circular: you are defining the segment by the behaviour you then report on.
    • Applying a third-party clickstream panel’s market-level AI Mode share to your site. Panels can indicate the market trend, but your query mix is not the market’s.
    • Subtracting AI impressions from Web totals and treating the remainder as clean classic data. Google does not document that the two are directly subtractable.

    Watch for this:

    AI Mode’s share of organic sessions is not the same as its effect on organic sessions. Some AI Mode clicks replace classic clicks the same user would have made anyway. If the business question is “how much traffic did AI Mode cost or add?”, the share estimate does not answer it. That needs a before and after comparison with a control group of low-exposure pages.

    How I would report it

    I would present a single slide or sheet with the low, middle and high estimates, the value of k behind each, the date range and country, and one line stating that Google does not publish AI Mode clicks separately. I would then show which decisions hold across the whole range. For example, “AI Mode is under 15% of organic sessions even at the high end, so it does not justify a separate content programme yet” is defensible. “AI Mode drove 7.7% of organic sessions” is not.

    Rerun it monthly with the same method, because the trend in the range is more informative than any single month. If Google adds clicks to the generative AI report, replace the k assumption with the real figure and retire the model. Until then, a transparent range with named assumptions is the most defensible answer available.

    Need help with this on your own site?

    Niraj Raut works with businesses in Nepal, Australia, the UK, Europe and the US on technical SEO, ecommerce SEO, local SEO and AI search.

    See SEO services

Add a reply

Replies are open to approved members. Log in or apply to join.