Your traffic is falling. Your visibility probably isn't.
Falling organic traffic looks like a problem. It's often AI systems answering the question without sending a click, which isn't the same thing.
Ben Rees - 1 May 2026

Most B2B marketers are watching their organic traffic decline and concluding something has gone wrong. In most cases, the diagnosis is incorrect.
What has changed is not whether people are finding your content. It is what they do after they find it. When ChatGPT, Perplexity, or Google's AI Overview answers a question, it pulls from your content, summarises it, and hands the user an answer. The user doesn't click. Your analytics record nothing. You assume you've lost ground.
You may have gained it.
This is the core measurement problem with AI search. The tools most marketers use - third-party SEO platforms, GA4 session counts - were built for a world where visibility and clicks were roughly the same thing. They no longer are. Research suggests those tools underreport actual impressions by a factor of five to ten compared to Search Console. Add AI-generated answers into that, and the gap widens further.
There are now three distinct things worth measuring, and they require different methods:
SEO : does your page appear in traditional search results? Measurable via Search Console.
AEO (Answer Engine Optimisation) : does an AI system lift a specific fact or sentence from your content in response to a direct question? Measurable by running test queries in Perplexity, Bing Copilot, and Google AI Overviews and checking for citations.
GEO (Generative Engine Optimisation) : is your brand or thinking woven into synthesised answers about your field, without necessarily being cited? This is harder to measure, and most companies aren't trying.
The practical implication: if your impressions in Search Console are stable or rising while clicks fall, your content is probably being used by AI systems, just without the click. That is not failure. It may be that users arrive on your site having already been informed about your value. There is some evidence that traffic originating from AI platforms converts at significantly higher rates than standard organic traffic, precisely because the AI has done the mid-funnel work before the visit.
None of this means ignoring traffic drops. But it does mean not treating a click decline as the whole story before you have checked what is happening to impressions, citations, and AI-surface appearances.
The measurement framework needs updating. The conclusion that everything is broken usually doesn't.
Citation Attention in Language Models: A Nadaraya-Watson Analysis of Learned vs. Surface-Level Kernels
When AI models cite a source, is it because the source is genuinely relevant, or just because the text is similar to the query? This paper quantifies the answer using Nadaraya-Watson kernel analysis.
My Mac Mini is going back to the shop
Three separate experiments on the same Mac Mini all pointed at the same conclusion: free and private don't matter if the answers are wrong or don't arrive at all.
Testing whether I should reuse old content
I fit a hierarchical Bayesian model to test whether retitling old content for AI citation works, and a robustness check collapsed the one clean result it found.
Who actually owns the stack I depend on
Six companies own or fund almost everything bjrees.com and Klara run on, including the AI that helped audit my blind spot.