AI Traffic Converts 3x Better. It's 0.5% of Your Traffic.

Andy Crestodina spent a year watching 29 million visits arrive at 97 B2B websites so he could answer one question: do the people who show up from ChatGPT ever buy anything. They do, at roughly three times the rate of everyone else. There were 140,000 of them.

That’s the whole study in two sentences, and the second one is the one nobody is quoting.

Crestodina, co-founder and CMO of the Chicago web agency Orbit Media Studios, published the numbers on July 22 covering July 1, 2025 through June 30, 2026. Visitors arriving from AI assistants converted into leads at 1.91%. Visitors from organic search converted at 0.71%. Broken out by platform, ChatGPT hit 2.1% against Google Search at 0.5%.

Then he did the thing most people publishing this kind of research don’t:

“Across 29 million visits, just 140,000 were from AI sources. Just 0.5% of all traffic was from AI sources.”

A rate isn’t a volume. 0.5% of your traffic converting three times better is a statistic about a rounding error.

Run the numbers the headlines skip

Crestodina states the comparison himself, in the same post that produced the 3x number everyone screenshotted:

“But remember, Google is driving 100x more traffic. Organic search is still a more powerful source for lead generation.”

That’s the author of the study telling you not to over-read the study. It’s worth pausing on how rare that is in vendor research.

Run it through your own numbers and the shape is obvious. On a site taking 100,000 sessions a year, this dataset says roughly 500 of them come from AI assistants and about 10 of those turn into a lead. The same site’s organic search traffic, converting at less than half the rate, produces vastly more of them because it’s a vastly larger number to start with. The multiplier is real. It’s also being applied to a base too small to change a quarter.

None of which means the trend is fake. Similarweb, measuring a different population with a different method, reported on July 29 that monthly web visits across generative AI platforms worldwide grew 70% YoY to 9.5 billion between June 2025 and May 2026. Something is growing quickly. It’s growing from a base that’s currently a rounding error on a B2B lead count, and both of those things are true at once.

Why the visitors are better

The quality difference isn’t mysterious, and it isn’t about the technology. Gaetano Nino DiNardi, a growth advisor who consults for B2B software companies on demand capture, gave the explanation quoted in the research:

“Website visitors coming from AI search platforms are generally better informed due to the hyper-personalized and conversational nature of the search journey.”

Somebody who arrives from a chat has already had the comparison conversation. They asked what tools do this, got three names, asked which one fits a 20-person team, and clicked through with the shortlist already made. A Google result catches people at every stage, including the ones three months from caring. Crestodina’s own read is the same, stated flatly: “The most important factor is the motivation of the visitor.”

The consequence is a measurement problem more than a strategy one. AI referrals also showed visitors twice as likely to subscribe to a newsletter, download a guide, or register for a webinar, which means the channel looks even better on the soft metrics most dashboards default to. A channel that converts well and reports well on 0.5% of sessions is exactly the kind of thing that gets a budget line it can’t yet justify. We’ve covered how AI visibility numbers get quoted past what they can support and how the split between search and AI reporting hides which is which.

The honest limits

This is 97 sites, all B2B, weighted toward the kind of company that hires a Chicago web agency. AI-referred sessions were identified by matching against known assistant domains, which undercounts anyone who reads an answer, doesn’t click, and types your name into a browser three days later. That path shows up as direct traffic and gets credited to nothing. If it’s large, this study understates AI’s contribution, and nobody currently has a clean way to size it.

The pattern held across about two-thirds of individual sites, which is a real result and not a universal one. A third of the sites in the sample did something else.

Here’s what this licenses you to do. Keep publishing the pages that answer comparison questions well, because those are what get recommended and the visitors they send are your best ones. Don’t move budget out of organic search on the strength of a conversion rate attached to 0.5% of your traffic. And when a vendor shows you a multiplier this year, ask what the denominator was before you ask anything else. Crestodina published his. Most won’t.