Blog · 25 September 2026 · By HeardOf

How much traffic does AI send? Published figures, with what each measured

How much traffic does ChatGPT send to websites?

No page we read on 25 September 2026 prints ChatGPT's own share of all website traffic; what the pages print is AI's share of all traffic, which is 1.08% or less in the three studies we read that measure it, and ChatGPT's share of the AI part, which each source counts on a different set of sites. Conductor's 2026 AEO / GEO Benchmarks Report, first published 12 November 2025 per its structured data and shown as last updated 6 July 2026, puts AI referral traffic at 1.08% of all website traffic across 1,215 of its enterprise customer domains in ten industries in the U.S., as monthly averages over May to September 2025, historical, and says 87.4% of that AI referral traffic came from ChatGPT on average. WebFX (18 March 2026, historical) gives generative AI 0.18% of total sessions in 2025, out of 2.3 billion sessions whose sites the page does not name. Ahrefs (16 June 2025, historical) measured its own site: "AI search traffic accounts for just 0.5% of our traffic for the last 30 days. 0.3% for YTD."

The newest page we read on how much traffic does ChatGPT send to websites is BrightEdge's press release of 24 September 2026, and it answers with a growth rate and a share of AI referrals: ChatGPT referral traffic "increased 101% from January to August this year", and its share of AI referral traffic reached 95.1% in August, after a 2026 low of 89.5% in April. The release names no sample, site count or network on the page as served, and prints no share of all traffic; it says AI "still represents a small portion of website referral traffic compared with traditional search". Its AI Market Pulse page, which carries no publication date, gives AI search 1.9% of total search referrals, up from 1.5% over a window its text does not state, with Google at 86.9%. That is a share of clicks from search engines and AI search, not of all traffic, so it does not sit on the same scale as Conductor's 1.08%.

Why do the sources give ChatGPT different shares of AI referrals?

Each one counts a different set of sites and adds them up differently, and none of the pages we read reconciles its ChatGPT share of AI referrals with another source's; Goodie sets its referral shares beside Similarweb's visits to the platforms, which is a different unit. For the same month, August 2026, BrightEdge's release gives ChatGPT 95.1% of AI referral traffic, while Statcounter's chart, as our post on which AI assistant people use prints it, gives 79.4% of AI chatbot referrals to websites worldwide; Statcounter's count rests on page views to its member sites and leaves Grok out, and BrightEdge's release names no sample at all.

The older figures differ in method as well as window. Goodie's 2026 report of 21 May 2026, historical, gives ChatGPT 62.6% of "measurable B2B AI referrals" for March and April 2026 as an average of each brand's own share, across a panel of mostly B2B brands whose number the page does not give, and its appendix notes that the five engine shares computed this way "sum to 103.0%". Conductor's 87.4% averages 1,215 enterprise domains in the U.S. over May to September 2025, and WebFX's 82.6% covers 2024 and 2025. BrightEdge's AI Market Pulse page says that "In July 2026, we updated historical data" after analytics platforms changed how they report AI referrals. A ChatGPT share quoted without the panel behind it has lost its subject.

Is AI referral traffic growing?

Yes, on the four pages in the table that print a growth rate for AI referral traffic, and each rate is on its own base and window, so none of them says how large the channel is. BrightEdge's release gives ChatGPT referral traffic 101% growth from January to August 2026 and 36% from July to August alone, and total referral traffic from AI platforms "roughly 73% in less than a year", without a start month. Adobe's figures, as Digital Commerce 360 relays them on 19 August 2026, put AI-referral traffic to U.S. retail sites 62% higher in July 2026 than a year earlier and 1,219% higher than in October 2024, on "more than 1 trillion visits to U.S. retail sites"; the relay does not link Adobe's own page and we did not reach it. WebFX's 796% is dated two ways on the same page, "in two years" and "YoY" for 2024 to 2025, historical, and Conductor's report says AI referral traffic "is growing around 1% month-over-month" on average over May to September 2025, historical. A growth rate says how fast a number moved; the share rows say how large it is.

Does traffic from AI show up as direct traffic instead?

Some of it may, say Goodie's and Similarweb's pages, and neither puts a number on how much. Goodie's report lists links copied out of a chat, native AI apps and privacy browsers as ways a visit loses its referrer, puts forward "a conservative estimate (5% of direct is misattributed AI)" without measuring it, and says "We can’t precisely size this from the data here." Goodie leaves Google's AI Mode and AI Overviews out, and Conductor leaves AI Mode out, both because Google Analytics files those clicks with Google organic. Similarweb's AI Search Stats page of 29 July 2026 says "roughly six in ten" ChatGPT referrals now land on a homepage, up from "roughly 26–29%" before a May 2026 change to how ChatGPT shows brand links, and that some traffic filed as direct "may increasingly be AI-driven discovery". Where the visits that do carry a referrer land is covered in how to see ChatGPT and Perplexity visits in your analytics, which starts from the AI Assistant channel in Google Analytics 4.

The figure we found on visits that follow an AI mention, by whatever route they arrive, comes second-hand. SparkToro's post of 28 June 2026 relays Similarweb's study The Downstream Impact of AI Visibility, on three consumer industries, and the transcript printed on the post, whose author box names Rand Fishkin and whose speaker says he contributed to the study, says that when an AI answer to which credit card company to choose recommended American Express, visitors were "seven point two percent more likely" to visit its website, and for Capital One "fourteen point two percent", over the following seven days, on Similarweb's clickstream panel. That is a lift in visits after a mention, not traffic an assistant referred. We could not read the study: its page as served carries navigation only, the Internet Archive's capture of 2 August 2026 is the same, and a headless render answered with a CloudFront 403, which we did not retry.

Are visits to ChatGPT the same as traffic it sends?

No. Similarweb's AI Search Stats page counts 9.5 billion average monthly web visits to generative AI platforms worldwide between June 2025 and May 2026, with ChatGPT's share of those visits falling from about 76% to roughly 53%; that counts people going to the assistants, not the clicks the assistants send on. Statcounter counts the other direction, clicks from a chatbot to a website, which is why its 79.4% and Similarweb's 53% are not two readings of one share.

How often do people click the links in an AI answer?

Rarely, in the two studies we read that ask, and the two count different things. Pew Research Center, using the browsing data of 900 U.S. adults in March 2025 and publishing on 22 July 2025, historical, found a click on a link inside Google's AI summary in "just 1% of all visits to pages with such a summary". The Reuters Institute's Digital News Report 2026, in a chapter dated 16 June 2026, historical, asked people in 27 markets: 4% of all respondents say they always or often click through to news sources from AI, against 19% from search and 17% from social media, and the chapter notes the figures are "self-reported behaviour". One is observed clicks on one Google feature, the other what people say about news.

Does AI traffic convert better once it arrives?

That is a separate compilation. Does AI traffic convert better answers "Higher in the headline of ten of eleven studies that print a baseline, each in its own unit, so sizes do not compare", and carries the figures and their units; this post prints none of them.

Does HeardOf measure how much traffic AI sends?

No. Our audit records which brands an answer names and which pages it cites, from our own systems, and not whether anyone clicked. The section "So does ChatGPT or Perplexity drive more traffic?" in who the AI actually cites answers "We do not know" for that reason, about our research run of 4 September 2026. The crawling side of the same exchange, how many pages AI crawlers fetch for each visitor they send back, is compiled in our post on crawl-to-referral ratios.

How was this read?

Every page named here was fetched on 25 September 2026 between 04:42 and 04:58 UTC with curl and a Chrome user-agent and read in full. BrightEdge's AI Market Pulse page was also rendered in Chrome headless at 04:43 UTC, because its chart labels are drawn by script. Dates are the ones each page shows, except Conductor's first publication and Similarweb's study page, which are dated from structured data only; historical means published, or measured, more than 90 days before the reading day, and the ages run from one day for BrightEdge's release to 466 for Ahrefs' post.

What we could not do: Similarweb's study page carried navigation only, as served and in the Internet Archive's capture of 2 August 2026, and our headless render answered a CloudFront 403, so its figures are here only as the transcript on SparkToro's post relays them; Adobe's own page was not linked from the relay and we did not reach it; and no page we read states the sample behind BrightEdge's figures, or over what window its 1.5% became 1.9%. What each publisher sells, from its own page: BrightEdge's release describes its AI Catalyst product, Conductor calls itself an "enterprise AEO platform", Goodie's and Similarweb's pages offer AI-visibility products, and WebFX is an agency offering "SEO and AI search optimization". Similarweb's page carries a "Summarize with AI" menu and WebFX's a "Summarize in ChatGPT" link, each pre-filling a prompt for the reader's assistant; both were read as data and not followed. Everything this post says about the pages named carries a link and the day it was read; what it says about our own audit comes from our own systems and cannot be checked from here.

What nineteen AI-visibility tools charge, and the ten things we could not verify