A domain's ChatGPT citation count can climb for six straight weeks while the referral-traffic line in Google Analytics sits perfectly flat. Two systems are running on different clocks here, not a tracking bug on your end: one counts what a model quoted an hour ago, the other counts a browser event that only fires if a reader clicks through, and lands correctly only if the app bothered to send a referrer at all. Citation share moves first. Referral traffic, when it shows up, moves weeks later, usually undercounted, and often filed under a header that has nothing to do with AI.
01The dashboard that's still asleep
In May 2025, that undercount had a name and a culprit. Google's AI Mode results were built with a noreferrer attribute on every outbound link, so a click through an AI Mode answer arrived at the destination site carrying zero referral data. Ahrefs' Patrick Stox confirmed the pattern directly: clicks from AI Mode don't show up in Search Console, and analytics tools file the visit as Direct or Unknown instead. Amsive's Lily Ray called it "Not Provided 2.0," a callback to the day Google stopped passing search-term data into Analytics, and argued the pattern kept anyone from seeing how much of AI Mode's traffic actually reached real domains. Google's John Mueller called the behavior unexpected, and the company shipped a fix within days.
The underlying mechanism, an AI answer surface that can simply choose not to identify itself to your analytics, hasn't gone away. ChatGPT, Perplexity, and Gemini's own apps still strip referrer headers often enough that a meaningful share of AI-driven visits keeps landing in the same Direct bucket as someone who typed your URL from memory. A rank tracker was never built to notice that. Citation tracking was.
02Citation share moves first, revenue moves second
Two real datasets show the same lag from different angles. Similarweb's May 2026 tracking of ChatGPT found the share of its answers carrying any citation at all rose from 0.6% in January 2025 to 2.8% by August 2025, nearly a fivefold climb. Total AI-sourced web visits across the engines Similarweb tracks grew more than threefold between September 2024 and September 2025, and the growth wasn't smooth. It arrived in the same stair-step pattern the citation rate did: flat, then a jump, then flat again at a new, higher floor.
Adobe Analytics ran the same story through retail. Generative-AI-sourced traffic to U.S. retail sites had been doubling every two months since September 2024, and by February 2025 sat 1,200% above its July 2024 level, with the conversion gap behind other channels narrowing from 43 points to just 9 over that same seven-month stretch. By the 2025 holiday season, that same channel was up 693.4% year over year, 670% on Cyber Monday alone, even as Adobe's own analysts cautioned that the base of shoppers using it "remains modest." None of those milestones surprised anyone who'd been watching citation behavior instead of the quarterly traffic report. The citation climb shows up first. The revenue line just takes longer to catch up to whatever the model already decided to quote.
For a meaningful share of AI-driven visits, citation share is the only report that will ever get filed. The referral line was never built to write one.
03The four-engine race, ninety days in
Caposeo watches that race article by article, on every domain in the workspace. Pull ninety days of citation share for a single tracked domain across all four engines and the shapes rarely agree. ChatGPT tends to move first and hardest, since its retrieval pass churns through its index more often than the other three refresh theirs. Gemini tracks whatever's already winning Google's organic results, because it grounds directly in Search rather than running an independent retrieval pass of its own. Claude only enters the count when its search tool gets invoked for a given answer, so its line can sit flat for weeks and then jump the moment an update makes tool use the default. Perplexity moves the most day to day, footnoting several sources per answer instead of settling on one, which keeps its share noisy even on a domain that's gaining ground.
Citation share by engine, one tracked domain over 90 days
Share of answers citing the domain, sampled every 10 days
Illustrative: a composite pattern typical of what Caposeo's daily citation tracking sees across the four engines, not one client's literal series. Sources: Similarweb, Gen AI Stats, May 2026 and Search Engine Land, May 2025.
Read the shape and the real takeaway sits past who's "winning" this week: three of the four lines can be climbing while a rank tracker, watching only Google, reports no meaningful change at all. The race has already started by the time a single-engine view notices anything happened.
04Reading the divergence, engine by engine
None of the four engines route a citation into your analytics the same way, which is exactly why watching one blended number instead of four hides the story rather than simplifying it.
What moves each engine's line, and where it lands downstream
05What we track instead of a single number
Keeping a rank tracker running still makes sense. Google's organic results send more traffic than the other four surfaces combined, and that isn't changing this quarter. But watch the four-engine citation race too, the way you'd watch a real race: per lane, updated daily, without waiting for the referral-traffic line to catch up and confirm what already happened. Caposeo runs 12.4K keywords through Google plus ChatGPT, Perplexity, Claude, and Gemini every day and rolls citation share, alongside crawl coverage, Core Web Vitals, and structured data, into a single 87 health score. The score moves when the underlying inputs move. It doesn't wait for a Direct-traffic spike to give it permission.
Watch your own citation race, not just your rank.
Caposeo tracks the same keywords on Google, ChatGPT, Perplexity, Claude, and Gemini in one workspace, every day, and rolls it into a single health score.