I used Neil Patel’s keyword research tool for the first time in years yesterday. And I forgot how good it was - I also became a victim of his very slick funnel straight afterwards and immediately signed up to a webinar he is giving on LLM’s and SEO on Tuesday. And, this got me thinking. People who do this stuff every day have been talking about AI eating SEO for about 18 months now. And I thought I’d check if this has happened yet, especially in a B2B eCommerce context (which I like to think we’re rather good at).

So, I thought I would use an (anonymised) data set from a current client to see what I could find. N.B. This article was produced by me and not ChatGPT, so please excuse the occasional long sentence, lack of emojis and ‘here’s the thing’. 

For the purposes of this exercise, I used a data set from the last six months. This particular client has Server Side Tracking implemented on their site, so we’re pretty confident in the attribution GA-4 can give us. We wanted to see: 

• What referrals were being driven by LLM’s
• How many actual sales were being driven by LLM’s (it has to be last click in this context, but this client, though B2B, sells consumer-like products where purchases don’t require a long decision process) - spoiler alert there were none
• What’s changed in the last six months (so we have benchmarked) 

Traffic patterns (17 Oct 2025 – 16 Apr 2026)

• The main AI referrer was ChatGPT. GA‑4 recorded roughly 106 sessions from chatgpt.com (about 0.47 % of all sessions). These visitors had 87 engaged sessions and an 82 % engagement rate, spending around 47 seconds on average

• No direct purchases were attributed, suggesting research‑stage intent.
• Perplexity delivered 14 sessions (combined perplexity and perplexity.ai) with nine engaged sessions

• Gemini (gemini.google.com) sent 4 sessions, half of which were engaged

• Traffic from Microsoft Copilot and other LLMs was negligible, and there was no measurable traffic from Bing Chat, Claude, You.com or Bard

• Monthly plots for ChatGPT show a steady rise: around 12 sessions in late October 2025, 17 sessions in November and December, ~20 sessions in January 2026 and a peak of ≈22 sessions in February. March was slightly lower (~21 sessions) and the first half of April shows about 11 sessions. Overall, ChatGPT referrals more than doubled from October to February and then levelled off.

What about the previous six months?

‍

What we learned

Traffic from LLM’s to this B2B eCommerce website actually declined. And, overall, the data indicate that generative‑AI referrals surged mid‑2025 and have tapered slightly since, though the quality of interactions (engagement rate and time) has improved. Now, we’re no soothsayers and wouldn’t like to make any predictions (we’ll keep this one rooted in facts), but we can see, empirically, that for this B2B audience, LLM referrals were nascent and now appear to be in decline. Whether this pattern mirrors in B2C and other B2B niches remains to be seen, and we’d love to know. We’ll run this exercise on some of our B2C clients and keep you posted.