The claim that AI traffic converts several times better than organic is everywhere, with figures from 2x to 27x depending on who is selling. In April 2026 Marketing Science published the largest study yet: 973 e-commerce sites, $20 billion of revenue, over 50,000 ChatGPT-referred transactions against 164 million from traditional channels. It found ChatGPT referral traffic converted better than paid social and worse than every other traditional channel, with organic search ahead of it. That is close to the opposite of the industry consensus. Here is why the numbers disagree, and what a small business should take from it.
If you have been sold anything to do with AI search in the last year, you have heard some version of this: traffic from AI assistants is small but mighty. Fewer visitors, far better ones. The numbers quoted run from twice as good to twenty-seven times as good, and the reasoning is intuitive enough that most people, including me, accepted it without much of a fight. The assistant has already done the comparing, so the person who clicks through has decided.
It is a good story. It also now has a large peer-reviewed study pointing the other way, and I think anyone writing about this owes you the update rather than the version that sells better.
What the claims actually say
To be fair to them, here are the main figures in circulation, with what each one is measuring.
- Semrush, 2026: AI-driven visitors convert at about 4.4 times the rate of standard organic, across industries.
- Visibility Labs, 94 e-commerce brands, 2025: ChatGPT referral traffic converted at 1.81% against 1.39% for non-branded organic, which is 31% better. Note the size of that gap compared with the headline multiples elsewhere.
- Seer Interactive, 2025: ChatGPT referrals at 15.9%, Perplexity at 10.5%, Google organic at 1.76%, over seven months. Roughly nine times. This is a single site.
I have quoted the Seer figure myself, in the piece on AI Overviews and falling traffic, with the caveat that one site is not the market. I still think the caveat was the right one. I now think it was not strong enough.
The study that says otherwise
In April 2026, Marketing Science published a paper by Maximilian Kaiser at the University of Hamburg and Christian Schulze at the Frankfurt School of Finance and Management, titled "Frontiers: ChatGPT Referrals to E-Commerce Websites: How Do LLMs Compare Against Traditional Channels?". It appeared in volume 45, issue 4, pages 699 to 715.
The dataset is the reason to take it seriously.
The headline finding: one year after launch, organic traffic from large language models showed conversion rates and revenue per session above paid social but below every other traditional channel. Trade coverage of the study reported affiliate links converting around 86% better than ChatGPT referrals, and organic search ahead of ChatGPT by roughly 13%.
A few of the secondary findings are more interesting than the headline.
It depends heavily on the product. The effect is moderated by product complexity: ChatGPT referrals performed better in complex categories. That fits the mechanism everyone has been describing. Where a purchase involves real comparison, the assistant does useful work. Where it does not, it is just another link.
The direction of travel is upward. Conversion improved steadily across the study period. But average order value fell at the same time, so revenue per session gained only moderately. That is a genuinely subtle result and it gets lost in both the optimistic and the pessimistic retellings.
The engagement pattern is odd. Favourable bounce rates, but lower session duration and fewer page views than traditional channels. People arrive, do not immediately leave, and do not look around much either.
The authors' own framing is that this traffic currently serves niche informational needs of proficient consumers rather than working as a broad conversion channel, and they raise consumer trust as one possible reason conversion trails.
Why the numbers disagree so wildly
A range from "27 times better" to "worse than everything except paid social" is not a small disagreement. It is not really a disagreement about facts either. It is mostly four measurement problems.
1. The comparison is usually rigged, not deliberately
Several of the friendly figures compare AI referral traffic against non-branded organic search. That is not like for like. Someone who arrives from an assistant has often already been given your name, seen it alongside two competitors, and chosen. The right comparison for that person is branded organic search, which converts far better than non-branded for exactly the same reason. Compare AI traffic to non-branded organic and you have compared people who know who you are against people who do not.
2. Who is doing the counting
Most of the impressive multiples come from vendors selling AI visibility products or from agencies selling AI visibility work. That does not make them dishonest. It does mean nobody has an incentive to publish the case where the numbers came out flat, and a study of the cases that got published is a study of the cases worth publishing.
3. Dark traffic biases the survivors
A share of AI-originated visits arrive with no referrer and get filed as Direct. Which visits survive with their referrer intact is not random: it correlates with browser, device and how the link was opened. So the sessions your analytics labels as AI traffic are a filtered subset, not a sample. I have written about what you can and cannot measure here in more detail.
4. The denominators are tiny
This is the big one for a small business. If you had eleven AI sessions last month and one enquiry, your AI conversion rate is 9%, against maybe 2% for organic. That is four and a half times better, it will look wonderful in a report, and it means nothing whatsoever. Two enquiries would have made it 18%. Zero would have made it 0%. Nothing about your website changed between those three worlds.
The Marketing Science study is persuasive largely because 50,000 transactions is enough that this stops mattering. Almost every number quoted at you is not.
So which is right?
Both, in the sense that they are measuring different things and neither is lying.
Individual sites, particularly in complex considered-purchase categories, genuinely do see AI referrals converting well, and the study agrees that complexity helps. Across a large sample of e-commerce, at market level, the advantage disappears and reverses.
What I would take from it is narrower and more useful than either headline: the rate was never the point.
| Scenario | Sessions | Conversion | Enquiries |
|---|---|---|---|
| Organic search | 900 | 2% | 18 |
| AI referrals, at the pessimistic rate | 14 | 1.8% | 0.25 |
| AI referrals, at the optimistic rate | 14 | 16% | 2.2 |
The difference between the pessimistic and optimistic worlds, for a typical small UK business today, is about two enquiries a month. Worth having. Not worth restructuring your marketing around, and certainly not worth a premium retainer sold on the strength of a multiple.
What would change that arithmetic is volume, not rate. And volume is a function of whether these systems can find, read and correctly describe your business, which is the unglamorous technical work rather than anything to do with conversion optimisation.
The work that changes the volume, not the multiple
Stagg Studios handles the technical side of being findable and readable, checked every month. £100 a month, first month free, no lock-in.
Where this leaves what I have written before
I said earlier that I quoted the Seer Interactive numbers on this site, describing AI visitors as converting far better, with a note that one site is not the whole market. Given what has been published since, that section is now more confident than the evidence supports, and I would write it differently today.
What I would say instead is this. AI referral traffic is small, growing, better than paid social, worse than most other channels at market level, better than that in complex considered purchases, and improving over time. The reason to care about being visible in AI answers is not that the clicks are magic. It is that a growing share of people are forming a shortlist inside a chat window, and you would rather be on it than not.
That is a less exciting argument. It is also one I can defend with a peer-reviewed study rather than a case study, and I would rather be the person who updates than the person who keeps quoting the number that sells.
What I would actually do
- Stop reporting AI conversion rate below about a hundred sessions. Report the count instead. Eleven sessions and one enquiry is a fact. A 9% conversion rate is a fiction built on it.
- Compare like with like. If you want to know whether AI traffic is good, compare it to branded organic search, not to non-branded. Most of the flattering gap lives in that choice.
- Ask new customers where they heard about you. Free text, on the enquiry form. It is the only method that catches the person who asked an assistant, thought about it, and typed your name into Google two days later. No analytics will ever attribute that.
- Judge the work on visibility, not on conversion rate. Whether your pages appear in AI answers is now measurable in Search Console. Whether the resulting handful of clicks converted at 9% or 4% this month is noise.
- Be more suspicious of multiples than of small percentages. The 31% figure from 94 brands is far more credible than the nine-times figure from one site, and the ordering of those two in most sales decks is exactly backwards.
Questions people ask
Short sourced answers. The working is above.
Does AI search traffic convert better than organic search?
At market level, the largest study available says no. Kaiser and Schulze, published in Marketing Science in April 2026, analysed twelve months of first-party data from 973 e-commerce websites covering $20 billion in revenue, comparing over 50,000 ChatGPT-referred transactions against 164 million from traditional channels. They found conversion rates and revenue per session above paid social but below every other traditional channel, including organic search. Individual sites, particularly in complex considered-purchase categories, do report better results, and the study found product complexity moderates the effect.
Why do so many sources say AI traffic converts several times better?
Four reasons, mostly measurement rather than dishonesty. Many comparisons pit AI referrals against non-branded organic search, which is not like for like, because someone arriving from an assistant has often already been given your name. Most impressive figures come from vendors or agencies with a product to sell, and nobody publishes the flat result. A share of AI visits lose their referrer and land in Direct, so the sessions that get counted are a filtered subset. And most reported rates rest on denominators far too small to mean anything.
Is the Seer Interactive 15.9% figure wrong?
Not wrong, but narrow. It describes one website over seven months from October 2024 to April 2025, and single-site case studies cannot tell you what happens across a market. The Marketing Science study covers 973 sites and reaches a different conclusion. Both can be accurate about what they measured. If you are deciding where to put money, the larger sample should carry more weight.
Does the type of product make a difference?
Yes, and this is one of the more useful findings. Kaiser and Schulze found that product complexity moderates the effect: ChatGPT referrals performed better in complex product categories. That fits the mechanism people describe, where the assistant does genuine comparison work before the click. For simple, low-consideration purchases, an AI referral behaves much like any other link.
Should I still care about AI visibility if the traffic converts worse?
Yes, but for the right reason. The case for AI visibility is not that the clicks are unusually valuable. It is that a growing share of people now form a shortlist inside a chat window before they visit anything, and being absent from that shortlist costs you the visit entirely. The study also found conversion improving steadily across its period, so the picture is moving. What it does not support is paying a premium on the basis of a conversion multiple.
How many AI sessions do I need before the conversion rate means anything?
As a rule of thumb, do not read a conversion rate off a denominator under about a hundred sessions, and be cautious well above that. Below that threshold a single extra enquiry can double or halve the number, so the figure is telling you about chance rather than about your website. Report the raw count of sessions and enquiries instead, and watch the trend across quarters rather than months.
Sources
Everything below is linked. None of it describes a Stagg Studios client. The Marketing Science study covers e-commerce sites, so its findings do not transfer automatically to a service business, and the vendor figures come from panels and case studies with the limits described above.
- Maximilian Kaiser (University of Hamburg) and Christian Schulze (Frankfurt School of Finance and Management), Frontiers: ChatGPT Referrals to E-Commerce Websites: How Do LLMs Compare Against Traditional Channels?, Marketing Science, volume 45 issue 4, pages 699 to 715 (2026)
- Digital Commerce 360, Study: ChatGPT traffic, sales referrals still trail traditional digital channels (2025)
- Search Engine Land, ChatGPT ecommerce traffic converts 31% higher than non-branded organic search (2026)
- Semrush, ChatGPT traffic analysis: insights from 17 months of clickstream data (2026)
What I would do about it, and what it costs
Here is the part worth sitting with. A conversion multiple is the easiest number in the world to sell with and one of the hardest to earn honestly. It needs a fair comparison, a large enough sample, and clean attribution, and almost no small business has any of the three. Which is why the figure quoted at you is usually the most flattering of several possible readings of a handful of sessions.
The thing that actually decides whether AI search is worth anything to you is not the rate, it is whether you are in the answer at all. That comes down to whether the crawlers can reach you, whether your pages are retrievable for the questions your customers ask, and whether the facts about your business agree across every place a machine might read them. None of that is glamorous and all of it is checkable.
That is the job I do, handled as one thing rather than a pile of separate products, and re-checked every month because sites drift and engines change. It gives you a chance of ranking better and makes it more likely that what these systems say about you is accurate. I cannot promise you a ranking, a citation, or a conversion rate, and after reading the study above I would be even more careful about anyone who does.
£100 a month, first month free, no contract, and you deal with me rather than an account manager. Start here, or send me a note and I will tell you what your site is currently missing.