LLM referrals amount to about 1% of website traffic. That figure comes from Conductor, which measured 3 billion sessions across its customer base. The same company describes a C-suite stampede: one Conductor executive has “spoken to more C-level executives in the last six months than in the last six years,” all asking what AI says about their brand. A 1% channel is generating 100% of the panic.
The mechanics of that gap are on the record. Ahrefs’ Patrick Stox, on the GEO Blueprint podcast, watched SEO budgets sit small for years and then swell the moment “people are scared.” The buying trigger is a board asking a question the CMO can’t answer. Fear moved the budget before any dataset did.
How much traffic has AI search taken?
About 2.5%. Graphite’s Ethan Smith pulled SimilarWeb data for the top 40,000 US sites and compared 2024 against 2025: “Total traffic is down, but it’s down 2.5% overall and not 25%” (Ahrefs Podcast). The distribution matters more than the average. The top 10 sites gained traffic, and the smallest sites gained the most. The middle fell 6 to 7%. Graphite cross-checked the result against its own client data and published the raw dataset at graphite.io/5%, with a stated reason: “everything is reproducible. So you don’t need to trust us.” An agency publishing raw data that shrinks its own scare story is a rare enough event to reward with a click.
Ahrefs runs a parallel measurement across about 70,000 sites connected to its tools. Tim Soulo’s read on Google’s traffic share since the chatbot era began: “it has moved single digit percentages… it was like six or 7%.” Two independent panels, tens of thousands of sites each, same single-digit answer.
The 25% number that headlines the doom decks comes from surveys. Smith investigated the studies behind the collapse narrative and found each one premised on self-reported recall: “Asking you to quantify your prior behavior is a terrible research method.” His background is human-computer interaction research, where that sentence is the first week of class. He also names the instinct underneath the panic: zero-sum bias, the assumption that if AI usage rises, the web must fall by the same amount. The same logic said mobile apps would kill the web in 2010. Apps blew up. The web grew. “The pie is getting bigger.”
Why does it feel worse than 2.5%?
Because the damage lands where marketers stare all day: informational clicks. Sistrix measured position-1 CTR falling from 27% to 11% under AI Overviews, with 265 million monthly clicks lost in Germany alone. Ahrefs data cited on the same episode: AI Overviews appear on about 30% of queries, weighted toward informational ones, and cut CTR by 34% when they show. Smith’s caveat carries the whole resolution: “The conversions are not going away. The transactional queries are not seeing AI overviews everywhere.”
HubSpot is the famous casualty. It lost around 80% of its blog traffic, and the assessment from a Conductor guest was blunt: “call their CEO, the business is doing just fine.” The pattern holds across the panels: informational queries migrate into AI answers, commercial queries keep clicking. Traffic graphs measure the first kind. Revenue measures the second. Most of the panic in the market is people reading the first graph and pricing it like the second.
What should you measure instead of borrowing panic?
Your own funnel, with your own buyers’ questions. The borrowed statistics run from 1% to 25% depending on who is selling what, and a number that moves 25x with the seller is an instrument problem. The question that decides your quarter is narrower: when a buyer asks an AI engine the questions that lead to your product, do you appear, and who appears instead.
That is a checkable fact, and checking it is cheap. Our free scan asks 5 questions your buyers would ask and shows you the answers with their citations, verbatim. The paid audit runs each question 7 times per engine, because single runs lie, and reports mention rates as ranges tied to captured evidence. No rank number appears anywhere in the report. The data above explains why.
Pick five buyer questions tonight and run them yourself, twice. If your brand shows up in the answers, the 1% figure is your ceiling of exposure and you can stop paying for panic. If your brand is absent from all ten, the 1% figure stops being comforting, and you have found real work worth doing. Run the free scan and see which case you are in.
Sources
- Ethan Smith on the Ahrefs Podcast: the 40,000-site SimilarWeb study, survey-method critique, AI Overviews CTR data
- There is no AI rank. Here’s the math: why single-run results are snapshots