Marketers noticed it first in their own dashboards: a thin stream of visitors arriving from ChatGPT and Perplexity that converted at rates their Google traffic had never reached. What looked at first like a rounding error is now one of the better-evidenced patterns in digital marketing.

The finding matters because Google organic traffic is shrinking for many B2B brands as AI-generated answers absorb the top-of-funnel queries that used to feed it. If the visitors AI does send convert several times better than the search traffic replacing them, the channel marketers fear losing may be worth more per visit than the one taking its place — provided teams can see it at all.

Andy Crestodina, co-founder and chief marketing officer of Orbit Media Studios, set out the mechanics in a nine-step breakdown, “Prompt Reverse Engineering: Deconstructing an AI Lead in 9 Steps,” published on the agency’s blog and amplified again this month by the trade outlet PPC Land. It traces a lead from first prompt to completed contact form, and argues that marketing teams treating AI referrals like ordinary search traffic will misread their own reporting.

Why do AI-referred visitors convert so much better than Google’s?

Crestodina attributes the gap to a shift from options to recommendations. In 2022 a buyer clicked several results and assembled their own answer; by 2025 the same buyer types a full sentence into a chatbot and receives a synthesised shortlist. That reframes these as commercial-intent searches rather than the informational “how-to” queries that fill a lot of Google traffic.

The persuasion that once happened across competing web pages now happens inside the AI’s response, before the click. By the time a visitor lands, the model has effectively pre-qualified them: they arrive late in the buying journey, already holding a shortlist, closer to a decision. That is why they complete forms at higher rates. Crestodina’s own GA4 shows ChatGPT.com converting at 2.93%, which he describes as significantly higher than his other sources — a single-site figure, but one many teams will recognise from their own reports.

Prompts also carry far more context than keywords. Citing click-stream research, Crestodina notes that some run roughly 20 times longer than a typical query, as buyers unconsciously state their role, their task, the context behind the request and the output they want. That extra detail lets the model infer where a buyer sits in their journey and filter toward a high-fit answer, which is what a late-stage prospect looks like when they reach a site.

How large is the gap, and how firm is the evidence?

The size of the gap depends heavily on who measured it. Ahrefs, in research published in June 2025, found AI search visitors converting 23 times higher than conventional organic visitors; that traffic was just 0.5% of visits but produced 12.1% of signups across the sites it studied. Microsoft Clarity, analysing more than 1,200 publisher and news sites, reported a more modest threefold gap in November 2025, alongside 155.6% growth in AI-referred traffic over an eight-month window. Semrush earlier valued AI search visitors at 4.4 times traditional organic by conversion rate.

The three-to-23-times spread is wide enough to warn against leaning on any single number, but every study points the same way. Two caveats keep the finding in proportion. Crestodina’s piece is a practitioner’s synthesis, not primary research: he presents no controlled study of his own and attributes the prompt-length and other figures to prior reports rather than to data Orbit gathered. And the raw conversion rates flatter AI partly because so little of its traffic is captured at all. Similarweb found that 55.9% of AI-influenced visits arrive as branded organic search rather than a trackable referral, which means the channel is at once under-counted and, on the visits it is credited with, unusually high-intent.

That measurement gap is the part most likely to trip up a marketing team. A visitor who reads a recommendation in ChatGPT, then types a brand name into Google or straight into the address bar, lands as branded search or direct traffic with no sign an AI sent them. Judged on referral traffic alone, AI looks like a trickle; judged on the leads it quietly seeds, it looks like something closer to a pipeline.

What can marketing teams do to earn and measure AI-driven leads?

Crestodina’s practical answer is to write for the prompt, not the keyword. If buyers frame requests around role, task, context and output, key pages should mirror that language so a model recognises the business as a match — a technique he calls prompt reverse engineering. Taglines, he writes, do not train the AI, and crawlers read text rather than images or logos, so proof points have to be spelled out in words to count.

That leads to a sharper distinction than most “AI visibility” advice allows: being listed among options is not the same as being named the best one. Closing that gap, in Crestodina’s account, means supplying evidence a model can weigh — case studies, testimonials, review counts, concrete delivery figures — that both persuades a human and gives the system a reason to prefer one supplier over an unproven rival. The emphasis on evidence has some independent support: an NP Digital survey in May 2026 found original research the most-cited content format in AI answers, at 82%, with comparison content second at 76%.

On measurement, the honest position is that attribution will stay partial. Crestodina suggests adding an “AI sent me” option to contact forms, while conceding the data will be rough, since a single lead can span several visits, people and channels before anyone fills anything in. The steadier move is to stop judging AI referrals on referral traffic alone and to watch for their signature instead: shorter paths to conversion, and a rise in branded and direct visits that tracks brand mentions inside AI answers.

Google’s own search leadership argued in June 2026 that good SEO remains good AI optimisation, a sign the discipline may be shifting faster than the tools built to measure it. For now, the clearest indication a brand is being recommended is a quiet one: unusually high-converting visitors its analytics still file under “direct.”