ChatGPT now sends a steady stream of visitors to small business websites, and most owners picture them arriving at the front door: the homepage. We checked where they actually land.
LovedByAI's WordPress plugin logs every visit that arrives from ChatGPT, along with the page it lands on. We took the 23,740 ChatGPT visits it logged between 14 September and 13 October 2026 on 323 small business websites that got at least 10, sorted each landing page by type, and worked out each site's split. The homepage is the single biggest landing page on a service business website, but on most sites it gets less than half of ChatGPT's referral traffic. On online stores, product pages beat it. That changes which pages are worth your time first.
ChatGPT referral traffic: the key numbers
- Service businesses: the homepage drew 42% of ChatGPT's visitors on the average site, service pages 18% and blog posts 18%.
- E-commerce: product pages drew 37%, the homepage 23% and category pages 13%. Product pages beat the homepage on 77 of 148 stores.
- Most visitors start off the homepage. That was true on 109 of 175 service business sites and 129 of 148 online stores.
- Blog posts are all or little. On 47% of service business sites no ChatGPT visitor landed on a blog post; on 14%, blog posts took half or more.
- Pricing, contact and about pages barely register: under 5% each on both kinds of site. Only 23 of 175 service sites had any ChatGPT visitor land on a pricing page.
All figures: LovedByAI plugin data, 14 September to 13 October 2026, sites with 10 or more ChatGPT visits, each site counted once.
Where does ChatGPT referral traffic land on a service business website?
On a service business website, ChatGPT sends the biggest single group of visitors to the homepage: 42% on the average site in our data. The rest spread out, and two kinds of page take most of it: service pages and blog posts, at 18% each.
Averages hide how much sites differ, so we also counted which kind of page came first on each site. The homepage led on 80 of the 175 service business sites. On the other 95, something else did: service pages on 28 and blog posts on 32. On 109 of the 175 sites, most ChatGPT visitors never started on the homepage.
That fits how people use ChatGPT. Nobody asks it for "a plumber's website". They ask who can fix a burst pipe tonight, and a page about burst pipe repair answers that better than a homepage does.
Where does ChatGPT referral traffic land on an e-commerce site?
On online stores the order flips: product pages are the biggest landing page, with 37% of ChatGPT's visitors on the average store against 23% for the homepage. Category pages add 13% and blog posts 11%.
Product pages beat the homepage on 77 of the 148 stores, and the homepage was the top landing page on just 32. On 129 stores, most ChatGPT visitors started somewhere other than the homepage. If you sell online, the product page is where a ChatGPT shopper meets you first.
Does ChatGPT send traffic to blog posts?
On some sites it sends most of its visitors to blog posts; on most it sends few or none. Blog posts average 18% of ChatGPT visitors on service business websites, the same as service pages, but that average is made of two very different groups.
On 82 of the 175 service business sites (47%), not one ChatGPT visitor landed on a blog post. On 25 (14%), blog posts took half or more of all ChatGPT visitors, and it was rarely one lucky article: on the typical one of those sites, ChatGPT visitors landed on 10 different posts. If one of your articles shows up in your own landing page list, it has become a front door, and it deserves the same care as a service page.
Does ChatGPT send visitors to pricing, contact or about pages?
Rarely. Pricing pages drew 1.4% of ChatGPT's visitors on the average service business website and 0.3% on the average store. Contact and booking pages drew 3% and 1%, about pages 4% and 2%.
Only 23 of 175 service business sites and 8 of 148 stores had any ChatGPT visitor land on a pricing page in the month. These pages may still matter once a visitor is interested; our data shows where visitors arrive, not where they go next. But they are not where ChatGPT sends people first, so they are not where to start.
How to track ChatGPT referral traffic in Google Analytics 4
Your own list matters more than our averages, and Google Analytics 4 can show it. ChatGPT adds utm_source=chatgpt.com to the links it gives, so tagged visits carry ChatGPT as their source.
- In Google Analytics 4, open Reports, then Engagement, then Landing page.
- Click Add filter, choose the dimension Session source, set it to exactly matches and enter
chatgpt.com. - Set the date range to the last 90 days and sort by sessions.
One caution: since 8 October 2026, part of ChatGPT's traffic arrives with no tag and no referrer, and Google Analytics files it under Direct. Treat the report as a sample of your ChatGPT visitors rather than the whole count. Our guide to UTM tracking for AI traffic covers the rest of the setup, and if any of the pages on your list return a 404, see why ChatGPT sends visitors to broken links.
Which pages to optimize for ChatGPT, and how
Once you know which pages ChatGPT sends people to, the work on each one is the same: make the top of the page say exactly what you do, who it is for and where, in the words a customer would use when asking.
Two of our own studies back this up. When we asked ChatGPT 50 local questions on 13 October 2026, 117 of the 135 businesses it recommended for a specific need were described with exactly that need, such as "24/7 emergency burst pipe repair" for a burst pipe question. And in our 108,834-page study of AI search optimization, optimized pages earned 11 times the AI referral visits of untouched pages on the same websites over the same 90 days.
LovedByAI is a generative engine optimization platform for small businesses, built around WordPress. It automates that on-page work across the site and reports AI mentions and AI referral traffic back, which is where the data in this study comes from.
What this ChatGPT referral traffic data can't show
- What visitors did next. We see the page a ChatGPT visitor landed on, not whether they called, booked or bought. Nothing here shows that a page type converts better.
- Every ChatGPT visit. Since 8 October 2026 some ChatGPT visits arrive untagged and look like Direct traffic, so the last six days of the window undercount. We have no reason to think they land on different pages, but we can't check.
- Every page type perfectly. Pages were typed from their URL by an AI model, so a few service pages filed as "other" or a blog post read as a service page will be in the mix.
- Every kind of website. These are WordPress sites running LovedByAI, most of them small businesses, and only sites with 10 or more ChatGPT visits in the month.
How we measured ChatGPT referral traffic (methodology)
Data. Every visit the LovedByAI WordPress plugin logged with ChatGPT as its source (utm_source=chatgpt.com or a chatgpt.com referrer) between 14 September and 13 October 2026, excluding links tagged as paid. That is 25,533 visits to 747 websites.
Sites. We kept the 333 sites with at least 10 ChatGPT visits in the month, so a single visit can't decide a site's split. Of those, 175 are service business sites (no ChatGPT visitor landed on a product, category or listing page), 148 are online stores (at least one landed on a product or category page) and 10 are directories, which we left out. That leaves 323 sites and 23,740 visits.
Page types. Each landing page was typed from its URL into one of eleven kinds: homepage, product, category, service, location, listing, blog post, about, contact or booking, pricing, other. Gemini did the typing, from the domain and the path.
Averages. For each site we worked out the share of its ChatGPT visitors that landed on each kind of page, then averaged those shares across sites. Every site counts once. Pooling the raw visits instead would let a handful of large sites decide the result: the five biggest of the 333 sites hold 26% of their visits, and one directory alone would make listing pages look like 15% of ChatGPT traffic.
Why not medians. A typical site gets ChatGPT visitors on only three or four kinds of page, so for most page types the median share is zero and can't rank them. The counts of sites in the text ("109 of 175") carry the spread instead.
