Generative engine optimization is being judged on a metric AI was built to eliminate: the click. A chatbot that answers a question completely leaves the user with no reason to visit a website, and analytics quietly records the outcome as failure. GEO \u2014 the discipline of ensuring a brand is named, described accurately, and recommended by AI answer systems \u2014 is a discipline where success may never register as a visit. The latest data shows referral numbers from AI platforms are indeed tiny. The conclusions that follow from them are not.<\/p>\n
Three separate analyses published in late 2026 arrive at strikingly similar raw numbers. Marie Haynes<\/a> examined twelve months of Google Analytics data from 100 websites, more than 145 million sessions in total. ChatGPT delivered just 0.62 percent of the session volume that organic Google search produced; all large language models combined contributed 0.73 percent. The British agency The SEO Works<\/a> studied 95 websites from September 2025 through August 2026 and counted 72,095 AI-referral sessions against 2.83 million organic sessions \u2014 a gap of roughly 39 to 1. An earlier Ahrefs<\/a> study from 2025 estimated AI referral traffic at about 0.1 percent of total website traffic. Even Google’s own AI search surfaces follow the pattern, sending far fewer clicks to publishers than the classic blue-link results that preceded them.<\/p>\n The three analyses agree on the core proportion: direct click volume from AI platforms is a rounding error next to Google organic. Where they diverge is the conclusion. Haynes argues GEO is a distraction. The SEO Works frames it as an essential companion to SEO. Kevin Indig, who spent months hunting for hidden AI influence in analytics data, argues the measurement itself is what deserves scrutiny.<\/p>\n
| Analysis<\/th>\n | Scope<\/th>\n | AI traffic vs. organic<\/th>\n | Conversion comparison<\/th>\n | Verdict on GEO<\/th>\n<\/tr>\n<\/thead>\n |
|---|---|---|---|---|
| Marie Haynes<\/td>\n | 100 sites, 145M+ sessions, 12 months<\/td>\n | ChatGPT 0.62% of Google organic; all LLMs 0.73%<\/td>\n | ChatGPT visitors 13.8% vs. ~3% for organic<\/td>\n | A distraction; prepare for agentic search<\/td>\n<\/tr>\n |
| The SEO Works<\/td>\n | 95 sites, Sep 2025\u2013Aug 2026<\/td>\n | 72,095 AI sessions vs. 2.83M organic sessions<\/td>\n | AI referrals 7.18% vs. 6.08% organic<\/td>\n | SEO and AI visibility reinforce each other<\/td>\n<\/tr>\n |
| Kevin Indig<\/td>\n | 12 GA4 properties<\/td>\n | Direct-channel surge mostly bots; no isolated AI effect<\/td>\n | Correlations up to 0.72, failed holdout tests<\/td>\n | Click attribution understates AI influence<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n The growth curve of AI referrals adds context. Haynes found that the exponential increase in ChatGPT-referred visits ended abruptly in August 2025 with the launch of GPT-5, and ChatGPT has sent fewer visitors since. As answer models become more self-contained, the need to click diminishes. That is not a failure of visibility; it is the product working as designed.<\/p>\n AI Visitors Convert Better \u2014 Until You Look at B2B<\/h2>\nWhen AI-referred visitors do arrive, they tend to be worth more than the average search visitor. In Haynes’s dataset, ChatGPT visitors converted at 13.8 percent, roughly 4.5 times the approximately 3 percent conversion rate of organic Google visitors. The SEO Works measured a smaller but real advantage: 7.18 percent for AI referrals against 6.08 percent for organic search, about 1.18 times. In that sample, for every 100 organic conversions, just 3.2 came through AI referrals. Google still contributed 96.5 percent of all conversions across the 55 sites in Haynes’s sub-sample \u2014 roughly 27 times the combined output of all large language models.<\/p>\n The averages hide a crucial segmentation problem. Among B2B product providers, AI traffic converted at only 0.62 times the organic rate, and on 23 of the 63 sites that tracked conversions, AI visitors performed worse than organic visitors. The pattern from these figures is not “AI traffic converts better” but “AI traffic converts better when the category fits.”<\/p>\n The mechanism is logical. A user who asks a chatbot to compare accounting platforms has already defined the problem, weighed features, and narrowed the options before clicking a single link. By the time they reach the website, a large part of the purchase decision has already been made. The same dynamic explains why long-cycle B2B products see less of a lift: an AI conversation can narrow the field, but it cannot compress a procurement process into a single exchange. Every company should therefore evaluate its own conversion data by segment before deciding what GEO is worth.<\/p>\n How AI Traffic Disappears Inside Google Analytics<\/h2>\nThe more serious problem with low click counts is that they capture only the AI influence that ends in an immediate, referrer-labeled visit. Kevin Indig spent months trying to find the influence that escapes the referral logs. Analyzing twelve GA4 properties, he found that the much-discussed surge in direct traffic was overwhelmingly bot traffic. One site recorded 409,000 direct sessions, of which only 198 showed genuine interaction. Across the dataset, direct traffic grew 140.9 percent between January and September, while the share of sessions with real engagement stayed almost flat. In 632 tests across four brands, Indig measured correlations up to 0.72 between AI visibility and later direct traffic \u2014 but the relationships did not survive time-shifted holdout tests.<\/p>\n Indig’s conclusion was appropriately cautious: the absence of proof in noisy data is not proof of absence. The U.S. company Ramp, for instance, attributes 30 to 40 percent of its direct traffic to answer engines, a figure it validated through engagement patterns and customer surveys rather than through GA4 attribution. No analytics package could have produced that number on its own.<\/p>\n The brand-search blind spot is even harder to dismiss. A consumer who reads a ChatGPT recommendation and then types the brand name into Google appears in GA4 as an ordinary organic search visitor. Similarweb data cited by Indig shows the scale of the phenomenon: users who received an AI recommendation visited the recommended brand’s website more often in the following seven days, and 55.9 percent of those visits arrived through search, against just over 40 percent for other visitors. Part of what traditional analytics credits to Google organic is actually AI-driven demand wearing a different label. Haynes’s framing of 0.73 percent versus 99.27 percent sets two channels against each other when they substantially overlap.<\/p>\n Google’s own AI surfaces blur the lines further. Sessions from AI Overviews and AI Mode are recorded in GA4 as regular organic sessions, which means The SEO Works could not isolate them at all. A brand cited inside an AI Overview is receiving AI visibility while the analytics attribution quietly files the visit under Google. Visits that arrive through in-app browsers or with stripped referrer data are likewise lost, which is why The SEO Works explicitly describes its figures as a lower bound rather than a complete measurement.<\/p>\n AI Visibility Works Even When No One Clicks<\/h2>\nJudging GEO on clicks assumes every valuable AI interaction should end in a visit. Many should not. A chatbot that fully answers a question has accomplished the task the user came with, and success in that context is invisible to web analytics by design. The historical comparison is not paid search but broadcasting: advertisers measured reach and brand impact long before clicks existed. AI answer engines return that model to the foreground.<\/p>\n The absence of a click is not the absence of an outcome. AI answers shape brand perception before the first website visit, and the impression formed in that setting can be more durable than one formed by a search snippet. A user who asks for the best payroll software for small businesses and receives a list of three names will form opinions before visiting any site. The brand missing from that list will not receive a later branded search, and the brand included will begin building trust before the first page loads. The SEO Works searched for this billboard effect \u2014 increased direct traffic attributable to AI<\/a> visibility \u2014 and found none in its data. But its test examined only the direct channel; the more likely path, through branded search, was outside its scope.<\/p>\n |