Source: bklink research, 27 October 2026
``` 97% of the 793 sites have a non-zero AI-assistant share, and 67% get 1% or more of their visits that way. Among publishers it is 38%. ## Is the share moving? No: in the one month we can measure, it fell. The 425 sites, of any kind, with an estimate in both August 2026 and September 2026 are compared with themselves, so a change in which sites are in the data cannot pass for a change in traffic: | Channel | Aug 2026 | Sep 2026 | Change in points | 95% interval | Read | |---|---:|---:|---:|---|---| | Organic search | 33.1% | 32.4% | −0.68 | −1.21 to −0.15 | fell | | Direct | 40.0% | 41.0% | +0.96 | +0.37 to +1.54 | rose | | Referrals | 8.8% | 8.4% | −0.40 | −0.62 to −0.17 | fell | | Social | 5.7% | 6.1% | +0.40 | −0.00 to +0.79 | no clear move | | AI assistants | 1.7% | 1.5% | −0.11 | −0.18 to −0.04 | fell | | Email | 3.0% | 2.9% | −0.14 | −0.28 to +0.00 | no clear move | | Paid | 6.4% | 6.3% | −0.12 | −0.36 to +0.12 | no clear move | | Affiliate and other | 1.3% | 1.4% | +0.09 | −0.00 to +0.19 | no clear move | Across those sites, the AI-assistant share fell by 0.11 points: higher at 98 sites and lower at 272 (55 within 0.05 points). Organic search fell by 0.68 points, and direct rose by 0.96 points. Among the 108 publishers in both months, the AI share did not move by more than the estimate's noise (−0.09 points). That is one pair of months. Estimates like these carry model error, and part of a month's change can be the model's own revision, so read it as one month's change, not a trend. The channels that did move, in points: organic search (−0.68), direct (+0.96), referrals (−0.40), AI assistants (−0.11). Ahrefs' tracker, in its September 2026 data, had Google's share of visits fall 0.28 points from the month before while AI assistants' share rose 0.02 points: the two AI changes differ in sign, and both are about a tenth of a point or less, so neither shows a surge. ## Who gets AI-assistant visits **Smaller publishers get more of their visits from assistants, as a share.** The publisher sites with fewer than 10,000 estimated visits a month average 1.4% from AI assistants; the ones with a million or more average 0.4%. The model has the least to go on for the smallest sites, so treat that end of the table as indicative. | Publisher size (estimated visits a month) | Sites | AI share, average site | Middle site | Sites where AI sends 1% or more | |---|---:|---:|---:|---:| | Under 10,000 visits a month | 50 | 1.4% | 1.1% | 68% | | 10,000 to 99,999 | 26 | 1.3% | 0.9% | 42% | | 100,000 to 999,999 | 38 | 0.6% | 0.4% | 21% | | 1 million or more | 40 | 0.4% | 0.2% | 13% | **It also depends on what the site is.** Counting every site with a report, the kind of site matters more than its size. Every other kind in the table gets a larger share than publishers do: | Kind of site | Sites | AI share, average site | Middle site | Sites where AI sends 1% or more | |---|---:|---:|---:|---:| | Publishers (news, magazines, blogs) | 154 | 0.9% | 0.8% | 38% | | Company sites | 299 | 2.1% | 1.8% | 80% | | Directories and listing sites | 200 | 1.6% | 1.6% | 79% | | User platforms | 27 | 1.3% | 1.2% | 59% | | Not yet classified | 97 | 1.4% | 1.2% | 54% | A plausible reason is that people ask assistants about products, tools and lists, and assistants link to those sites; the data does not say why. Kind of site is mxverify's reading of each site's homepage (what it is mainly for), and a site not yet read is shown as not classified. News and magazine sites read lowest here, which fits the Chartbeat data on news publishers cited below. ## Which assistants send the visits On the 241 sites with an assistant split in each of the last three months, the average split of a site's AI-assistant visits was: | Share of a site's AI-assistant visits | Jul 2026 | Aug 2026 | Sep 2026 | |---|---:|---:|---:| | ChatGPT | 49.6% | 60.4% | 55.2% | | Gemini | 9.9% | 10.1% | 8.7% | | Perplexity | 3.4% | 2.4% | 3.0% | | Claude | 3.7% | 3.5% | 2.8% | | Copilot | 0.0% | 0.0% | 0.3% | | Grok, DeepSeek | 0.7% | 0.5% | 0.4% | | No assistant named in the estimate | 32.8% | 23.0% | 29.6% | **ChatGPT is the largest named source, at 55% in Sep 2026; Gemini is second at 9%.** The estimate leaves 30% of a typical site's AI visits with no assistant named, so these are shares of all AI visits, not of the named ones; leave the unattributed out and ChatGPT is 78% of the named visits. That is close to what other studies find: 4.3 million of the 5.3 million AI-assistant visits in Ahrefs' September 2026 data (81%) came from ChatGPT, and SE Ranking has 74.78% for ChatGPT and 11.56% for Gemini in its 2026 sample. The estimate names no Meta AI, which Ahrefs' tracker puts second by visits that month, so some of the unattributed share may be assistants it does not list. ## What other studies show | Study | Sample and period | What it found | |---|---|---| | [Ahrefs, AI vs Search Traffic](https://ahrefs.com/benchmarks/ai-vs-search) | 115,254 sites on Ahrefs Web Analytics, to September 2026 | AI assistants sent 0.43% of visits; search sent 26.82% (Google 23.96%) | | [SE Ranking](https://seranking.com/blog/ai-traffic-research-study/) | 101,574 sites with Google Analytics, January 2025 to April 2026 | AI platforms sent 0.32% of traffic in 2026, up from 0.24% in 2025 and 0.02% in 2024 | | [Chartbeat 2026 Publisher Playbook](https://www.searchenginejournal.com/google-search-decline-publishers-chartbeat/590681/), as reported by Search Engine Journal | Chartbeat's publisher clients, July 2025 to July 2026 | Google Search referrals fell 40.2% in a year; AI referrals were 0.01% of publisher traffic and fell in 2026 across the assistants | | [Reuters Institute, Trends and Predictions 2026](https://reutersinstitute.politics.ox.ac.uk/sites/default/files/2026-01/Newman%20-%20Trends%20and%20Predictions%202026%20FINAL.pdf) | Chartbeat data, over 2,500 sites, November 2024 to November 2025 | Google organic search traffic fell 33% worldwide and 38% in the US; Google delivers 500 times as many referrals as ChatGPT from search alone | | [Pew Research Center](https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears-in-the-results/) | 900 US adults, 68,879 Google searches, March 2025 | With an AI summary on the page, people clicked a result in 8% of visits, against 15% without one; 1% clicked a link inside the summary | | [Similarweb](https://aisearch.similarweb.com/blog/gen-ai-stats/) | Visits to AI platforms themselves, June 2025 to May 2026 | ChatGPT's share of generative-AI web visits fell from about 76% to about 53%; Gemini rose from under 9% to about 27–28% | Three things here line up with the studies. The pooled AI share of publisher visits (0.34%) compares with the 0.32% and 0.43% the two large trackers find across all the traffic they see. Publishers read low in both: this data, and Chartbeat's news publishers at 0.01% of traffic. And where search is losing visits, the studies do not show assistants making up the difference: Chartbeat has Google Search referrals to publishers down 40.2% in a year and AI referrals at 0.01% of their traffic. Pew's result is the mechanism people point to: an AI answer at the top of the results page means fewer clicks to any site. It measures clicks on Google, not visits from assistants, so it explains the search side only. ## What this means for link building - **Search is still where the visits are.** An average publisher gets about 41 organic-search visits for each AI-assistant visit, so a link that helps a page rank still reaches far more readers than a link an assistant might surface. - **A link's value to rankings does not count AI-assistant visits.** mxverify's Backlink Value takes a site's organic-search visits as one input, beside its authority rating and the kind of link it gives; visits from AI assistants do not move it. See [how it is calculated](https://mxverify.uk/backlink-value), and the [publisher sites ChatGPT sends visitors to](https://mxverify.uk/publications/ai/chatgpt). - **Assistant visits are worth a look when you choose where a link goes.** A site an assistant sends visitors to has an audience that asks questions the assistant answers with that site. That is a reason to read the column, not a price for it. - **This study does not measure citations.** It counts visits that assistants send. Whether a placed link makes an assistant name a page is a separate question, and nothing here answers it. Browse the publisher sites in this data, with their traffic breakdowns, in mxverify's [publisher directory](https://mxverify.uk/publications/type/publisher), or [check a site's backlinks](https://mxverify.uk/) yourself. ## Method and limits - **Modelled, not measured.** Every figure is a third-party model's estimate for a site's latest month, not the site's analytics. Small sites have the least data behind them, and the model's channel split for one site can be far off; an average over hundreds of sites is steadier than any one row. - **Not a sample of the web.** The sites are those mxverify has a report on: 793 in September 2026, from some of the web's largest to a long tail of small ones. They are not drawn at random, and "all sites" counts companies and directories, not only publishers. - **The set changes by month.** 425 of the 465 sites measured in August 2026 have an estimate for September 2026 (91%). Levels in different months describe different sets of sites; only the matched comparison reads a change. - **A change is called a move only when its 95% interval leaves zero.** The interval is the mean paired change ± 1.96 standard errors. - **What counts as an AI assistant is the model's.** It does not say how it treats a click from Google's own AI answers, so this post does not assume. - **Assistant shares are partial.** The named assistants rarely add to 100%; the rest is shown as unattributed, not spread across the named. - **One traffic model.** The outside studies use other panels and measure other things; the table says which. - **Outside figures read 27 October 2026.** Each was read on its source's page that day. ## Edition log This post is rebuilt each month, once mxverify has re-read its sites for the new month. Earlier months stay in the log. | Month | All sites | AI share, average site | Change on matched sites, points | Publishers | AI share, average publisher | Change on matched publishers, points | |---|---:|---:|---|---:|---:|---| | August 2026 | 465 | 1.68% | — | 128 | 1.00% | — | | September 2026 | 793 | 1.61% | −0.11 (fell) | 154 | 0.90% | −0.09 (no clear move) | The newest data, September 2026, was read on 27 October 2026. ## Key Takeaways - AI assistants sent 0.9% of the average publisher's visits in September 2026; organic search sent 37%. - On 425 sites measured in August 2026 and September 2026, the AI-assistant share fell by 0.11 points. - Smaller publishers and company sites get a larger share of their visits from assistants than large news publishers do. - ChatGPT is the largest named source of assistant visits (55%), then Gemini (9%). ## Frequently Asked Questions ### What share of website traffic comes from AI assistants? In mxverify's data, 0.9% of an average publisher's visits in September 2026, and 1.6% for the average of all 793 sites with a report. Pooled across all the visits they track, the large trackers find 0.32% (SE Ranking) and 0.43% (Ahrefs). ### Is traffic moving from search to AI assistants? Not in this data. Between August 2026 and September 2026, matched sites' organic-search share fell by 0.68 points and their AI-assistant share fell by 0.11 points. Chartbeat reports Google Search referrals to publishers down 40.2% in a year with AI referrals at 0.01% of publisher traffic, so what search loses has not shown up as assistant visits. ### Which AI assistant sends the most traffic? ChatGPT, at 55% of a typical site's AI-assistant visits in Sep 2026 in mxverify's data, ahead of Gemini at 9%. About 30% of those visits name no assistant in the estimate. ### Do AI assistants send more traffic to small or large sites? A larger share to small ones. Publishers with under 10,000 estimated visits a month average 1.4% from assistants; those with a million or more average 0.4%. ### Where does this data come from, and how reliable is it? A third-party traffic model's monthly estimate for each site mxverify has a report on, not the sites' own analytics. It is steadier as an average over many sites than for any one, and a change is called a move only when its 95% interval leaves zero. ### Does an AI-assistant visit count towards a link's value? No. mxverify's Backlink Value takes a site's organic-search visits as one input, beside its authority rating and the kind of link; visits from AI assistants do not move it. ### How often is this updated? Monthly. mxverify re-reads its sites when the traffic model publishes a new month, and this post is rebuilt from that data. The edition log at the bottom keeps earlier months. ## Sources 1. [Ahrefs – AI vs Search Traffic Analysis](https://ahrefs.com/benchmarks/ai-vs-search) 2. [SE Ranking – AI traffic research study (June 2026)](https://seranking.com/blog/ai-traffic-research-study/) 3. [Search Engine Journal – Google search decline for publishers (Chartbeat 2026 Publisher Playbook)](https://www.searchenginejournal.com/google-search-decline-publishers-chartbeat/590681/) 4. [Reuters Institute – Journalism, media and technology: trends and predictions 2026](https://reutersinstitute.politics.ox.ac.uk/sites/default/files/2026-01/Newman%20-%20Trends%20and%20Predictions%202026%20FINAL.pdf) 5. [Pew Research Center – Google users are less likely to click on links when an AI summary appears](https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears-in-the-results/) 6. [Similarweb – Generative AI statistics](https://aisearch.similarweb.com/blog/gen-ai-stats/) 7. [mxverify.uk – How links are read and audited](https://mxverify.uk/methodology)