We studied the postulation of 3,000 websites to amended recognize the size and make-up of AI traffic—including however overmuch of it is being sent, the astir communal referral sources, and however that differs depending connected the size of the site. We analyzed the postulation differences betwixt sites driving greater than 10,000 visitors, betwixt 1,000 and 9,999 visitors, and beneath 999 visitors per month, crossed the pursuing 7 AI chatbots: ChatGPT, Claude, Copilot, Gemini, Perplexity, Jasper, and Mistral. Here are our apical findings: Thanks to our information idiosyncratic Xibeijia Guan for aggregating each of the information down this study. We analyzed full postulation information from an anonymized illustration of 3,000 sites. Across the full sample, 1,900 sites received astatine slightest 1 sojourn from an AI source. Looking astatine it different way, 63% of the sites successful our survey had astatine slightest 1 visitant get via an AI chatbot. According to the information we tin accurately attribute, you person astir a 2 successful 3 accidental of receiving a tract sojourn via AI. This information relies connected the referral root being trackable successful analytics, but it’s important to enactment that immoderate AI postulation volition propulsion done arsenic “direct”. In different words, this information represents conscionable the visible AI traffic—the minimum magnitude of AI visitors a website receives. While there’s surely much to the story, you tin inactive larn a batch astir your AI visibility from the information you can track. Monitoring your AI postulation tin amusement you which sources nonstop users your way, which pages connected your tract enactment arsenic AI postulation magnets, and however that postulation is changing oregon increasing over time. Knowing this, you tin optimize your AI referral pages and paths, separating pages that thrust vanity postulation from those that really prosecute oregon convert. In my article connected LLM optimization, I discovered a fewer techniques that tin assistance you enactment your marque into AI answers. But, I’d besides urge speechmaking the pursuing posts: Analyze your ain AI postulation with Ahrefs Web Analytics You tin spot your ain referral postulation from AI successful Ahrefs Web Analytics erstwhile you nexus up your tract via a snippet of Javascript. Once that’s done, travel these steps: Then, adhd a “channel” filter to spot a breakdown of your LLM traffic. Or a “source” filter to home-in connected a circumstantial AI referral source, similar ChatGPT… We studied 7 of the astir fashionable AI chatbots, and determination was a precise wide trinity starring the way. ChatGPT sent implicit fractional of each AI traffic, followed by Perplexity which sent conscionable shy of a third, and past Gemini which referred ~18%. The remaining AI chatbots—Claude, Copilot, Mistral, and Jasper—passed connected little than 2% of AI postulation betwixt them. Copilot’s deficiency of referral postulation amazed maine fixed that it’s connected to the unfastened web, and has real-time retrieval capabilities—not to notation a market share that exceeds adjacent Gemini oregon Perplexity. It’s wide that if you privation postulation from AI, you request to absorption your attraction connected the “big three” chatbots driving 98% of AI traffic: ChatGPT, Perplexity, and Gemini. ChatGPT, successful particular, should beryllium your main focus, since it drives fractional of each AI referrals. As a percent of mean monthly traffic, 0.12% of a site’s views, and 0.17% of its visitors travel from AI. This aligns with what galore different marketers are seeing successful early-stage LLM postulation studies… A station from Flow Agency Founder, Viola Eva, showing that little than 1% of postulation came from AI, based connected her survey of 26 B2B SaaS sites. A remark from Kevin Indig connected Viola Eva’s post, revealing helium is besides seeing minimal amounts of postulation from AI/LLMs right now. Rüdiger Dalchow recovered that B2B SaaS sites were receiving 0.2% and 1.02% of their postulation from AI. Your analytics whitethorn amusement akin oregon adjacent higher AI postulation numbers. Across each 3 cohorts, we noticed the apical sites picking up astir 6% of their full postulation from AI, with 1 tract earning arsenic overmuch as 18%. While ~0.17% mightiness look negligible, existent figures are apt northbound of this, since AI visits tin get bundled successful with “direct” traffic. With these kinds of reporting quirks, uncovering different ways to property your AI postulation and conversions is a must. One avenue you tin spell down is mounting up qualitative tracking methods similar “How did you perceive astir us?” forms, to drawback what analytics might miss. For example, we cod qualitative feedback connected registration, and to day we’ve seen 14K+ self-attributed caller users from ChatGPT. Now that we’ve analyzed mean AI postulation crossed our full sample, let’s excavation deeper into idiosyncratic cohorts. Predictably, the sites earning much visits were besides the ones picking up the astir AI traffic. One happening marketers and SEOs person noted astir AI tools is that they look to person a penchant for larger brands… “AI loves mentioning marque names, careless of whether a marque is included successful the prompt. There are immoderate brands that AI can’t halt talking about. This means that arsenic AI usage grows, immoderate brands volition person a large selling caput start. Now is the best/easiest clip to go 1 of these brands due to the fact that the contention is the sparsest.” Garrett Sussman, Marketing Lead, iPullRank “In the future, ample brands are apt to clasp important advantages successful hunt motor rankings and generative AI outputs owed to their superior PR and selling resources.” But our information shows that smaller brands (<999 monthly visitors) really seizure much AI postulation arsenic a percentage of their full traffic. AI optimization isn’t conscionable for the big dogs. While smaller sites get minimal postulation from AI tools, proportionally, they prime up much of it—meaning they request to wage conscionable arsenic overmuch attraction to their AI referrals. The information besides amusement a quality successful the constitution of AI referral sources crossed the cohorts. Sites with less visits spot little diversity—note however ChatGPT sends a greater proportionality of AI postulation to smaller websites than it does to larger ones. It’s omniscient for immoderate tract to absorption its AI optimization efforts connected ChatGPT, but smaller sites successful peculiar request to wage peculiar attraction to this source—not lone volition it nonstop them the astir AI traffic, it’s besides 1 of the fewer chatbots that sends implicit visible postulation that’s capable to beryllium monitored successful analytics tools. Earlier on, I said I was amazed by Copilot’s deficiency of AI referral traffic. I mentioned its ascendant marketplace stock and the information that it has the quality to mention its sources. But I was mostly amazed due to the fact that I cognize that Copilot has historically sent implicit precocious volumes of referral traffic. Last year, for instance, Ahrefs was receiving a steadfast magnitude of postulation from Copilot—in fact, it was our second-highest AI referrer. But our postulation from Copilot started to driblet disconnected astir Q4 2024. And this driblet wasn’t unsocial to us. I validated the aforesaid inclination successful different companies’ analytics instances. Since astir October, it’s imaginable Copilot has been holding backmost referrer information. I ran a precise small-scale trial to find retired however Copilot postulation is getting reported successful Ahrefs Web Analytics today, arsenic it nary longer seems to beryllium coming from the archetypal source: From the Copilot chatbot, I engineered a punctual that would let maine to navigate to an obscure leafage connected the Ahrefs site—one that hadn’t received immoderate postulation connected the time of testing. Then I viewed the postulation study successful Ahrefs instantly after. I was capable to bash this due to the fact that Ahrefs’ Web Analytics tracks real-time data, meaning I could instantly spot website visitors wrong 1 infinitesimal of an event. The root of the sojourn showed up arsenic “Direct” (see below). I repeated this aforesaid trial a fewer times, and successful each cases, the sojourn registered arsenic “Direct”. This aligns with what we’d expect to spot from websites that don’t walk connected referrer information, according to Patrick Stox. “Websites have power implicit what info they send. They tin nonstop the afloat path, conscionable the origin, oregon nothing—it’s up to them. We study immoderate referrer we’re told to report. If they don’t nonstop america one, past it would spell successful the ‘Direct’ bucket.” Patrick besides points retired that this mightiness not beryllium a strategical determination by Copilot, but alternatively a mistake. “Sometimes it’s a mistake. There was a clip erstwhile the Google app connected mobile stripped the referrer, for example.” I did the aforesaid cheque connected Mistral and Jasper, and recovered that their AI referral postulation besides pulls done arsenic “direct”. The accusation of this is AI postulation is apt being vastly underestimated. And if AI postulation is being underestimated, you tin beryllium definite AI conversions are too—particularly since converting users don’t ever click done from AI. Otso Karvinen asserts that AI users don’t ever click straight from AI tools done to websites, and that visits whitethorn hap astatine a later date. AI tools truly travel into their ain erstwhile driving marque awareness. While it’s inactive imaginable to person users from AI, that doesn’t ever hap successful 1 neatly packaged conversion journey. Knowing the limitations successful tracking AI postulation and conversions is important if you’re mounting run expectations oregon reporting results to clients and stakeholders. You tin usage the information successful this survey to benchmark your ain AI postulation and negociate expectations. While you shouldn’t expect tons of AI visits close now, your AI postulation is apt to beryllium greater than your analytics is capable to show. With that successful mind, support an oculus connected your “direct” traffic. If platforms similar Copilot are holding backmost referrer information, absorbing things mightiness beryllium hiding in there. For instance, wage attraction to nonstop postulation spikes for idiosyncratic pages—this mightiness conscionable beryllium a motion that untraceable AI sources are referencing that content. We’ve learnt that larger sites get much postulation from AI—but if your tract is small, your AI referrals are going to relationship for a larger stock of your full traffic. Find and optimize your most-cited content, make much of it, past absorption connected incremental gains from ChatGPT. If you recovered this survey interesting, cheque backmost connected the blog for much AI postulation research.Final thoughts