AI search and assisted discovery

A growing share of research now happens inside an AI assistant or an AI answer surface rather than on a results page. The user asks a question, gets a synthesised answer that compares options, and clicks through - if they click through at all - already holding a shortlist.

Most of the writing about this is SEO writing, concerned with how to be cited. The CRO question is different and more immediate: the traffic that does arrive has different properties, and the pages it lands on were built for someone else.

What changes on arrival

The visitor is further down the journey. They’ve had the category explained, seen a comparison, and formed a view. In awareness stage terms they’re arriving solution-aware or product-aware where previously a chunk of that traffic arrived problem-aware. Warmer, in traffic temperature terms, and expecting to be treated that way.

Most landing pages are built to educate first. Hero explains the category, three benefit blocks explain why the category matters, proof and CTA arrive somewhere below. For a visitor who has already been sold the category by an assistant, that’s a page arguing a case they’ve accepted, and delaying the specific evidence they came for - your actual specification, price, availability, and why you rather than the two alternatives the assistant listed alongside you.

The pre-click comparison is invisible to you. You never see the comparison set. The visitor may arrive having been told something about you that isn’t current, or having been shown a competitor’s strongest claim next to yours. Objections you’d normally handle on-page have already been formed elsewhere.

What changes in the numbers

This is where teams misread it. If assistant-mediated discovery replaces some informational search traffic, the pattern is fewer sessions with a higher conversion rate. Both halves are true simultaneously and neither is the story on its own.

Read that on session volume and it’s a crisis. Read it on conversion rate and it’s a triumph. Read it on absolute conversions and revenue and you find out what actually happened. This is a good moment to remember that a rate is a ratio with a denominator that can move independently.

Attribution also degrades further. Assistant referrals frequently land as direct traffic or with unhelpful referrer data, so the channel under-reports, which biases every attribution model against it. The practical counter is self-reported attribution - a “how did you hear about us” field becomes more valuable as machine-readable attribution gets worse, not less.

What’s actually testable

The reason this belongs in a CRO reference rather than an SEO one is that there are real experiments here, and they don’t require believing any particular forecast about how big this gets.

• Segment first, theorise second. Isolate assistant referrers in your analytics and compare conversion rate, AOV, and repeat rate against organic search. Segment analysis rules apply - specify it before you look, and treat it as one segment rather than a licence to slice everything. You may find the segment behaves indistinguishably, in which case you’ve saved yourself a workstream.
• Size it honestly. For most sites this traffic is currently small. A segment at 2% of sessions cannot be tested on directly with any power, and a “personalised experience for AI traffic” built on it is a project with no measurement attached. Knowing the size tells you whether the next two bullets are worth doing at all.
• Test the arrival experience for warm traffic generally. Rather than building something AI-specific, this is usually better framed as a test of front-loading specifics - price, spec, availability, differentiation - against the education-first default. That hypothesis is testable on your whole warm segment today, and it happens to serve assistant traffic well.
• Make the substance retrievable. Specifications in text rather than baked into images, prices on the page rather than behind an enquiry form, comparison content that states its own conclusion. This is partly an SEO play and partly not: the same changes serve a returning human who wants a fact quickly. If your pricing is “contact us”, you are absent from every comparison an assistant generates, and that’s a commercial decision worth making deliberately rather than by default.

The honest caveat

This is early, the volume is small for most businesses, and there is a large amount of confident nonsense being sold about it. I’d treat it the same way as any emerging channel: instrument it, measure the segment, size it against everything else competing for the roadmap, and let that decide how much attention it gets. The discipline doesn’t change just because the channel is new.