Very little: in about two weeks of server logs from three course websites we work with, exactly one human visitor arrived from an AI assistant (via Copilot), and ChatGPT, Perplexity, Gemini, Claude and Meta AI sent none. Businesses still win customers from AI answers, though, because the value sits in being named in the answer, not in the click, and our clients report installs, B2B customers and leads that came from ChatGPT, Claude and Gemini.
The rest of this page covers what our logs show, what outside studies found, how a business gets customers from an assistant that sends no clicks, and what to measure instead.
How many visitors do AI assistants actually send?
Almost none. The two busiest of those course-site logs held roughly 1.2 million requests. Among them we found one human visitor referred by Copilot and zero from ChatGPT, Perplexity, Gemini, Claude, Meta AI or any other assistant. On websites, assistants mostly answer in place rather than send someone over.
The machines, on the other hand, visit constantly. Across the full retained nginx logs of an edtech platform we work with, pickaclass.com, from 28 July to 11 August 2026 we counted:
- ClaudeBot: 19,272,125 requests
- Meta-ExternalAgent: 3,854,076
- GPTBot: 2,744,985
- Applebot: 2,655,555
- Amazonbot: 1,170,268
- Googlebot: 52,292
That is a strange picture if you are used to search engines. The AI companies read your site far more than Google does, and the people using their assistants almost never follow a link back.
There is a middle category worth watching. Live retrieval agents such as ChatGPT-User, OAI-SearchBot and Perplexity-User fetch a page while an assistant is answering a real person. In the same fortnight the edtech platform received 1,918 of these fetches on content pages, and a client blog of a few dozen articles received 56 across 24 different articles. Each one means a person asked a question and an assistant read that page to answer it. That person never shows up in your analytics, but your page shaped what they were told. Our guide on how to tell if AI crawlers visit your site shows how to find these lines in your own logs.
What do independent studies say about AI clicks?
They point the same way. In a March 2025 analysis of browsing data from 900 U.S. adults, Pew Research Center found that Google users who saw an AI Overview clicked a traditional result in 8% of visits, against 15% when no summary appeared, and clicked a link inside the summary itself in just 1% of visits.
Google disputed the reading. In its response, reported by TechCrunch, Google's VP of Search Liz Reid said total organic clicks from Search had been relatively stable year over year, while acknowledging that user trends are shifting traffic to different sites, so some gain and some lose.
For ChatGPT specifically, a comparative analysis summarised by PPC Land estimated that ChatGPT handles about 12% of Google's search volume but sends roughly 190 times less referral traffic, with an estimated click-through rate of 1.3% against Google's 29.2%.
Our logs are smaller than any of these, but they agree: people read the answer and stop there.
If nobody clicks, how do businesses get customers from AI?
They get them through the name, not the link. Here is what our clients reported to us in October 2026:
- FreightSizer, a B2B SaaS for freight dimensioning, got 2 new B2B customers in one month, both from ChatGPT.
- MagiScan, an iOS 3D-scanning app, gets hundreds of installs a month from ChatGPT, Claude and Gemini.
- appslift, an app store optimisation service, gets 2 to 5 leads a month from AI assistants.
These are the clients' own figures, not ours. The mechanism is simple once you stop looking for a referral. Someone asks an assistant which app scans objects in 3D, reads a name, opens the App Store and types it in. Someone asks which tool calculates freight dimensions, then searches the company name or types the domain. In your reports that shows up as an app store search, a branded Google search or a direct visit. Nothing in the referrer says chatgpt.com.
So referral reports undercount what assistants do for you, often to zero. The practical fix is low-tech: ask new customers where they first heard of you, and add the assistants as answer options on your signup or contact form.
Why does being named matter more than the click?
Because for many buyers the answer is the shortlist. If an assistant names three firms, the buyer usually looks at those three. A business that is not named is not on the list, however good its website.
Not every question produces a shortlist. In a test on 23 September 2026 with 20 German business questions, the model answered 10 without searching at all and cited nobody. Where it did search, it cited primary sources such as laws, ministries, the OECD and large accounting firms on 9 of 10. Firms were named only on 'who should I hire' or 'how do I choose' questions. A local example shows the same split: on 20 October 2026, for an orthopaedic practice in Bonn, the four assistants named the practice in 12 of 80 answers to 20 patient questions (Claude 5 of 20, Gemini 4, ChatGPT 2, Perplexity 1). The directory jameda.de was cited in 12 answers, competing practices' own websites were the most cited sources, and only the 'who to choose' questions named any practice. Questions about costs, risks and treatments named none.
Two lessons follow. First, the questions that name businesses are the buying questions, which is exactly where you want to appear. Second, the sources an assistant reads for those questions are often your competitors' sites and directories, so the content on your own domain matters. The assistants are built to lean on the web for this: Wikipedia notes that ChatGPT's web search was added to make answers more accurate and up to date.
Answers also vary. In the same September test, two runs of the same question named two different sets of firms. Being named is a probability per question, not a fixed ranking.
What should you measure instead of referral traffic?
Measure the answers themselves, plus the few traces assistants leave on your server.
| Signal | What it tells you | Where you find it |
|---|---|---|
| Mentions in answers | Whether assistants name you for your buyers' questions | Ask the same questions repeatedly and record which businesses are named |
| Citations | Which pages the assistant used as sources, yours or others | The cited links in answers; via the API, citation annotations |
| Live retrievals | That an assistant read a specific page while answering a real person | Server logs: ChatGPT-User, OAI-SearchBot, Perplexity-User |
| Customer-reported source | Whether being named turned into revenue | Signup forms, sales calls, asking new customers |
| Referral clicks | The small share of people who did click through | Analytics referrers, expect very low numbers |
We ask each buyer question three times per assistant and count a business as named if any of the three answers names it. We compare weeks only on the same questions with the same number of asks, otherwise the variation between runs swamps any real change.
Citations are machine-readable. OpenAI's web search documentation says that with web search enabled, the response includes inline citations with a url_citation annotation holding the URL, title and location of each cited source. That is how you can count which sites an assistant leans on.
Measuring is cheap. At API prices of October 2026, one question asked once with web search costs us about $0.069 with Claude, $0.010 with ChatGPT, $0.009 with Gemini and $0.0013 with Perplexity. A report of 20 questions across four assistants, three asks each, costs about $5.40. For a manual method, see how to check whether AI assistants mention your business.
Why might assistants not name your business at all?
Most often because they cannot read or find your pages. We learned this on our own site. A robots.txt rule meant to hide one redirect page, /b, was written as a prefix. Because robots.txt rules match prefixes, from 11 August to 21 October 2026 it also closed /blog, /business and every article in our business blog to all crawlers. Bing Webmaster Tools listed our business page as 'discovered but not crawled' from 11 August, and Google had none of the business pages indexed. We changed the rule to /b$ on 21 October 2026. The same day we put 20 English and 20 German questions about getting a business into AI answers to ChatGPT, Claude, Gemini and Perplexity, three times each, and were named in 0 of 160 answers, with none of our pages cited. We wrote up that day-0 measurement in full; if you suspect a similar problem, check how to allow GPTBot and other AI crawlers in robots.txt.
Being indexed by Google is not enough either. On 19 October 2026 we asked assistants for the exact titles of 18 articles, each older than three weeks, on three client sites. ChatGPT found 1 of 18 and Claude found 7 of 18, while Google had indexed 29 of 30 articles of one of those sites.
And markup alone does not carry you. A site with server rendering, Article, FAQPage and Organization JSON-LD and a sitemap, but only 75 URLs, was absent from answers on its own article's exact topic. A site with about 190,000 crawlable URLs was crawled constantly. Surface, meaning many pages that answer real questions, beats tidy markup.
The good news is that change does show up. In our weekly measurement of 20 October 2026, Perplexity cited an article we had published on a client's site, a German company-formation advisory.
What should you do next?
Stop judging AI search by referral clicks; they will stay close to zero. Instead, check whether crawlers can reach your pages, publish answers to the questions your buyers ask on your own domain, measure mentions on those questions every week, and ask customers where they heard of you.
To see where you stand today, the free check at itgrows.ai generates five buyer questions from your homepage and shows in about 30 seconds whether ChatGPT and Perplexity name you. If you want the work done, ItGrows publishes articles on your domain, fixes crawler access and indexing, and measures ChatGPT, Claude, Gemini and Perplexity weekly for $499 a month (plus a one-time $499 setup, free with 3 months or a year). Nobody can guarantee that an assistant will name a business, and we do not.
Frequently Asked Questions
Can I judge my AI visibility from referral traffic in analytics?
No. Assistants rarely send clicks, and people who act on an answer usually arrive as direct visits, branded searches or app store searches. Measure mentions in answers and ask customers where they found you.
Why does ChatGPT send so few clicks even when it cites my page?
The answer is complete enough that most people stop reading there. One analysis estimated ChatGPT's click-through rate at 1.3% against Google's 29.2%, and our own logs showed no ChatGPT referrals at all in about two weeks.
Should I block AI crawlers if they send no traffic?
Blocking them removes the chance of being read, cited and named, which is where the value is. Live retrieval agents fetch your pages while answering real people, even though those people never visit.
How can I find out whether AI assistants name my business today?
Run the free check at itgrows.ai: it builds five buyer questions from your homepage and shows whether ChatGPT and Perplexity name you, in about 30 seconds. For a fuller picture, ask your buyers' questions several times per assistant, because answers vary from run to run.