Obility Editorial · · 4 min read
The famous multipliers come from very different measurements
A common line in AI search pitches is that one visitor from ChatGPT is worth several from Google. The figures usually quoted are 4.4 times, from Semrush, and 23 times, from Ahrefs. Both companies did publish them. Neither is a number you can drop into a business case for your own site, and the largest independent study so far points the other way.
Semrush’s AI search study says the average visitor from a non-Google AI source such as ChatGPT is 4.4 times as valuable as the average traditional organic search visit, based on conversion rate. The page introduces the figure with “we have seen” and gives no sample size, no description of the sites and no definition of a conversion. The rest of that study is built on digital marketing and SEO topics.
The Ahrefs figure comes from Ahrefs’ own website. In a post by Patrick Stox, Ahrefs reported that AI search sent 0.5 percent of its traffic over the previous 30 days and 12.1 percent of its signups, most of them from ChatGPT. That is where the 23 times comes from. The post is clear that the result is for Ahrefs and that a study across many sites was still in progress. It describes one software company, one month and one kind of conversion, a signup.
The largest peer-reviewed study put ChatGPT below most channels
In April 2026, Marketing Science published a study by Maximilian Kaiser of the University of Hamburg and Christian Schulze of Frankfurt School of Finance & Management. They used Google Analytics data from 973 e-commerce websites with about $20 billion in combined revenue, from August 2024 to July 2025, covering more than 50,000 transactions from ChatGPT referrals and 164 million from other channels. ChatGPT referrals converted better than paid social and worse than every other established channel in the study, including organic and paid search, email and affiliate links. Revenue per session followed the same pattern. Visitors bounced less often than those from most channels, but viewed fewer pages and left sooner.
Two details in the paper matter more than the headline. The results were stronger in complex product categories, where buyers need help comparing options, and weaker for simple purchases. Conversion rates also rose over the year while order values fell. The authors conclude that this traffic currently serves the information needs of a small group of practised users and is not yet a broad conversion channel.
A conversion ratio depends on who sits in the denominator
These results conflict less than they appear to. A multiplier compares two groups of visitors, and the groups differ before they reach your site. Someone who clicks through from an AI answer has usually read a summary and compared options there, and clicks because they are ready to act. Organic search traffic to a company with a large blog includes many people reading a how-to article with no plan to buy. Divide one rate by the other and you are measuring the blog as much as the channel. A software company with a large, popular blog sits near the extreme of that effect.
The kind of purchase matters too. A trial of business software is the sort of complex decision where the Marketing Science study found ChatGPT referrals did best. A phone case is not. Both sets of numbers can be accurate and still tell you little about your own site.
One limit cuts the other way. The paper’s authors point out that last-click attribution understates channels that mainly help people discover options. A buyer who sees your name in an answer and later types it into Google is credited to branded search, not to the answer that introduced you.
Compare like with like on your own site
A fairer test uses your own analytics. Compare AI referrals with non-branded organic search only, and only for sessions that land on the same kinds of pages: product, pricing, comparison and use case pages. Measure the outcome the business cares about, such as a demo request, an activated trial or revenue per session, rather than any form submission.
Then read the counts before the ratio. In an illustrative quarter, 400 AI-referred sessions produce 12 demo requests, a 3 percent rate, and 20,000 non-branded organic sessions produce 300, or 1.5 percent. That looks like double. Yet three requests more or fewer in the AI group would put its rate anywhere from 2.25 to 3.75 percent. With a dozen conversions the ratio is a hint, not a finding, so repeat the comparison next quarter before acting on it.
Obility’s Agent Analytics is built around that separation. It looks at the landing pages and actions of visitors arriving from answer engines and keeps those visits apart from automated traffic and from citation observations, so a referral conversion rate is not mixed up with crawler hits or with answers nobody clicked.
Clicks are the smaller part of what an answer does
Do not put any vendor multiplier into a forecast, and do not read the e-commerce study as proof that AI answers do not matter to you. Both mistakes treat the click as the whole effect. An answer can shape a decision without producing a visit: the buyer reads the shortlist, closes the tab and searches for one of the names a week later.
The conversion rate of AI referrals measures the small group who clicked. Whether you are named in the answer, and how you are described, covers everyone who read it. Track both, and report them separately.