AI Search Statistics: Which Numbers to Trust (2026)
Studies say AI sends 0.1% of traffic, or 1%, or more, and converts worse than search, or 4x better. Here is what each one actually measured.
- AI assistants send roughly 0.1% to 1% of website traffic, depending on who measured it and when. Numbers much above that usually count something other than clicks.
- AI referrals often convert better than organic search, but the measured gap ranges from about 1.3x to over 4x. One large independent study found the opposite for ecommerce.
- AI Overviews cut clicks to the top organic result. Ahrefs measured a 34.5% drop in April 2025 and 58% in February 2026.
- AI brand recommendation lists almost never repeat. Track how often you appear across many runs, not your position in a single answer.
- Before you cite a stat, check four things: sample size, method, who paid for it, and the data window.
Why AI search statistics disagree
Ask how much traffic AI sends and you will find answers from 0.1% to double digits. Ask how well it converts and you will find anything from "worse than search" to "more than 10x better". Marketers on Reddit have started calling the whole category made up. The numbers are not made up, but most of them measure different things.
Four differences explain most of the spread. What is counted: referral clicks, estimated "dark" visits, or phone calls. Who is in the sample: enterprise customers of one vendor, a panel, or a random set of sites. When: AI traffic changed fast in 2025, so a March 2025 number and a 2026 number are not comparable. Who paid: many studies come from companies selling a fix for the problem they measured.
How we graded each study
- A: large sample, method published, independent or vendor-neutral, data from the last 12 months.
- B: large sample and clear method, but vendor-run, customer-only, or data older than 12 months.
- C: useful signal, but small, narrow, or the method is only partly disclosed.
A grade is about how far you can generalize a number, not whether the authors were honest. A B-grade study can be exactly right for the segment it measured.
How much AI traffic is hidden as "Direct"?
Loamly analyzed 446,405 visits and found that 14,413 of 20,428 identified AI visits (about 70.6%) arrived with no referrer, so GA4 would file them as Direct. Loamly sells software that detects exactly this traffic, so the reporting outlet itself describes the figure as "directionally strong rather than settled". Grade: C.
The practical takeaway holds even if the exact share is lower: your GA4 AI-referral number is a floor, not the total. Source: Ecommerce Fastlane on the Loamly dataset.
Does AI traffic convert better?
| Study | Finding | Sample | Data window | Grade |
|---|---|---|---|---|
| Kaiser & Schulze (Univ. Hamburg / Frankfurt School) | Organic LLM traffic underperformed every traditional channel except paid social on conversion, AOV and revenue per session; gap narrowing | 973 ecommerce sites, $20B revenue, 50K+ ChatGPT transactions | Aug 2024–Jul 2025 | A |
| Visibility Labs (via Search Engine Land) | ChatGPT converted at 1.81% vs 1.39% for non-branded organic (+31%); AOV $204 vs $238 | 94 ecommerce sites, GA4 | Jan–Dec 2025 | B |
| Similarweb | AI referrals converted at 11.4% vs 5.3% organic | Similarweb panel | Jun 2025 | B |
| Invoca Lead Conversion Benchmarks 2026 | Calls referred by ChatGPT had a 49% lead rate, highest of any channel | 70M+ calls, 10 industries (Invoca customers) | Published Jul 2026 | B (calls only) |
| Loamly | Dark AI traffic converted at 10.21% vs 2.46% non-AI | 446,405 visits | Early 2026 | C |
Defensible claim: AI referrals are small in volume and, for considered purchases and lead generation, usually convert at or above organic search. For ecommerce the evidence is split, and the only academic study in the set found AI traffic converting worse. Do not put a single multiplier in a board deck.
Sources: Digital Commerce 360 on Kaiser & Schulze, Search Engine Land on Visibility Labs, Blog du Modérateur on Similarweb, PPC Land on Invoca.
How much do AI Overviews reduce clicks?
Ahrefs compared 300,000 keywords. For keywords that later showed an AI Overview, position-1 desktop CTR fell from 0.073 in December 2023 to 0.016 in December 2025. Seer Interactive tracked 3,119 informational queries across 42 organizations from June 2024 to September 2025. Grade: A for both, with the caveat that Seer's sample is informational queries only.
The finding that matters most is Seer's: being cited inside the AI Overview recovers part of the loss. Sources: Ahrefs 2026 update, Ahrefs April 2025, Seer Interactive.
Can AI visibility trackers be trusted?
Rand Fishkin (SparkToro) and Patrick O'Donnell (Gumshoe.ai) had hundreds of volunteers run 2,961 prompts 60 to 100 times each across ChatGPT, Claude and Google's AI in November and December 2025. There was less than a 1% chance that ChatGPT or Google returned the same brand list twice. How often a brand appeared across many runs was far more stable than its position. Grade: A.
So a tracker that reports "you rank #2 in ChatGPT" from one run is reporting noise. A tracker that runs each prompt repeatedly, stores the full answer with a timestamp, and reports the share of answers that mention you is measuring something real. Ask any vendor, including us, which one they do. Source: Search Engine Journal on the SparkToro study.
llms.txt and AI crawlers
Does llms.txt increase AI citations?
No measurable effect so far. SE Ranking analyzed nearly 300,000 domains; 10.13% had an llms.txt file, and adding the variable made their citation model less accurate (November 2025). Grade: B (correlational). Source: SE Ranking.
How much do AI crawlers take versus send back?
Cloudflare measured pages crawled per visitor referred between January and July 2025: OpenAI went from about 1,217 to 1,091, Anthropic from about 286,930 to 38,066, and Perplexity from about 55 to 195. Grade: A for its network, though these ratios move quickly. Source: Cloudflare.
Stats to stop citing
- Any single conversion multiplier ("AI converts 4.4x", "23x") presented without the sample and industry.
- Double-digit "AI traffic share" figures with no click-based source.
- 2025 traffic-share numbers presented as current. AI referral volume changed materially through 2025.
- Any AI "ranking position" taken from a single prompt run.
Frequently asked questions
What percentage of website traffic comes from AI?
Roughly 0.1% to 1% for most sites. Conductor measured 1.08% across 13,770 enterprise domains (May to September 2025), with 87.4% of AI referrals from ChatGPT. Ahrefs measured 0.1% across about 35,000 sites in March 2025, and 0.3% for sites under 10,000 visits.
Does AI search traffic convert better than Google?
Often, but not always. Visibility Labs found ChatGPT traffic converted at 1.81% vs 1.39% for non-branded organic across 94 ecommerce brands in 2025, and Similarweb measured 11.4% vs 5.3% in June 2025. An academic study of 973 ecommerce sites (Aug 2024 to Jul 2025) found AI traffic underperformed every traditional channel except paid social.
How much do AI Overviews reduce clicks?
Ahrefs found AI Overviews correlate with a 58% lower click-through rate for the top-ranking page (February 2026, 300,000 keywords), up from 34.5% in its April 2025 study. Seer Interactive found brands cited inside the AI Overview earned 35% more organic clicks than those not cited.
Are AI visibility tracking tools accurate?
Single runs are not. A SparkToro and Gumshoe study found less than a 1% chance that ChatGPT or Google's AI returned the same brand list twice. How often a brand appears across many runs was much more stable, so trustworthy trackers repeat prompts and report share of answers, not a single position.
We build Measure, a self-serve platform that tracks how AI engines describe and recommend brands, and connects that visibility to real traffic and revenue. Everything here is written from what we see in the data every day.