On an online specialty grocery store we run in Northern Europe, ChatGPT sent 5.3% of all sessions between January and late October 2026. Those sessions produced 25.5% of the store's tracked revenue. Google organic search, by comparison, sent about half of all sessions and produced 58% of revenue.
Read per visit, the gap is wide. Over the four months where purchase tracking was complete, July to October, a visitor arriving from ChatGPT completed one of the store's tracked key events in roughly 24% of sessions. A visitor from Google organic did so in about 15%. Revenue per session from ChatGPT was about 2.8 times Google's.
AI search does not replace a website. It shrinks the audience the website sees, and the people who still arrive are further along, so the website's job moves from attracting visitors to closing the ones an AI already convinced.
The volume is small, and it is growing
The concession first, because it frames everything else: AI referral traffic is still a small share of the web. Across more than a hundred thousand websites, SE Ranking measured ChatGPT at about a third of one percent of all referral traffic in May 2026, and other trackers put all AI assistants together well under 1% of visits, against roughly a quarter from search. If your plan depends on AI sending you crowds, it will disappoint you.
The trend is the part worth watching. On the same store, ChatGPT sent roughly six to ten sessions for every hundred from Google in the first quarter of 2026. Over the summer it sent fifteen to eighteen. Almost all of it was ChatGPT: 143 of 144 AI-assistant sessions in the last 28 days came from chatgpt.com, which matches industry trackers that put ChatGPT's share of AI referrals well above three quarters.
Why AI visitors behave differently
The mechanism is selection, not magic. Someone who asks an assistant "where can I buy this in Sweden" or "who delivers this to Norway" has already done the research step inside the chat. By the time they click, they have a shortlist of one or two, and the click is a decision rather than a question. A search click often comes earlier, when the person is still comparing ten blue links.
We see the same pattern on very different sites. On our own agency site, visitors from AI assistants were engaged in 61% of sessions over the last 28 days, against 31% for Google organic. On PrepMyPost, our free creator-tools site, AI assistants sent 29% of all sessions in its first tracked week while Google organic sent under 4%, because a young site has no rankings yet but can already be recommended.
Industry studies agree on the direction and disagree wildly on the size. Published comparisons of AI-referred and organic conversion range from about 1.3 times to over 20 times. We would not plan around any multiplier, including ours. We would plan around the direction.
Where AI visitors land
This was the most useful finding in the store's data. Only 16% of ChatGPT sessions this year landed on the homepage. The rest arrived on specific product pages, on the page explaining delivery across Europe, and on a seasonal guide to a religious holiday. Each of those pages answers one narrow question completely.
That is the shape of AI referral traffic for a store: the assistant answers "can I get this, here, by then" and links to the page that proves it. A product page with a vague description, or a delivery policy hidden in a PDF, gives the assistant nothing to cite and the visitor nothing to act on.
Service businesses see the opposite. Industry data shows ChatGPT referrals shifting toward homepages after a product change in May 2026, and for an agency or a clinic the homepage is often the page an assistant links when it recommends the company by name. Which pages matter depends on what people ask about you.
What the website's job becomes
Three jobs replace "get more traffic", and each has a cost.
Be the source. Publish the facts an assistant needs to recommend you: what you sell, where you deliver, what it costs, how long it takes. Plain sentences, on the page, in the HTML. The cost is that you give away answers that used to require a visit.
Close on arrival. The visitor who comes from an AI answer has skipped your homepage tour. The page they land on must carry price, availability, delivery and a clear next step without making them hunt. On the store, the delivery page converts because it answers the question that blocks the order.
Keep the long tail honest. Product and policy pages that were never worth optimizing for Google are exactly the pages AI cites. Thin descriptions copied from a supplier are now a measurable loss.
Where this stops applying
Three limits. First, all of these numbers are per visit; total revenue from AI is still a minority on every site we run. Second, AI visitors are self-selected, so the conversion gap partly reflects who they are rather than anything the website did. Third, many AI-driven visits arrive with no referrer at all and are counted as Direct, so every figure here is likely an undercount. We cover how to measure around that in our piece on measuring a website when clicks fall.
Questions to answer about your own site
- What share of your revenue, not your sessions, came from AI assistants this quarter?
- Which pages do AI visitors land on, and does each one answer its question completely?
- Are your price, delivery area and lead time written in plain text an assistant can quote?
- If an assistant recommended you by name today, would the page it links close the sale?
If you want this read on your own analytics, our SEO team runs exactly this channel-and-landing-page analysis as the first step of every engagement.