What AI visibility is actually worth.

The post is titled "0.5% of Visitors Drove 12.1% of Signups." Ahrefs published it in June 2025. Sample size: one website, ahrefs.com. Visitor counts: not disclosed. Signup counts: not disclosed. And the traffic in question was largely people asking ChatGPT to recommend an SEO tool and then arriving at the website of an SEO tool. Branded intent, in the one category where the brand was the answer.

Ahrefs was candid about all of this at the time. The post's author, Patrick Stox, doubted the effect would scale: "If it did scale, we'd see more than 5x more conversions based on the higher conversion rate, and my gut tells me it won't be quite that good." Ryan Law, the company's director of content marketing, added that "these CTRs are probably the highest they'll ever be as the novelty of this format wears off."

Eight days later, Ahrefs published the multi-site study the first post had promised: roughly 82,000 websites. It measured pages per visit, time on site, and bounce rate. It did not measure conversions, and it said why: "We've only just released conversion tracking in Ahrefs Web Analytics, but as soon as we have enough data, I'm hoping to dig into AI conversions."

Thirteen months later, that conversion study has not been published. The number has.

It now circulates across the category as a robust, multi-site finding. It is neither. Every independent multi-site estimate is dramatically smaller: Semrush puts the premium at 4.4x, RankScience at 5x, Visibility Labs at 31% across 94 ecommerce brands. Same claimed effect, an order of magnitude apart, and the version that traveled is the one with a sample size of one.

That is not a footnote. That is the foundation of an industry.

What the measured answer looks like

In April 2026, Marketing Science published the first peer-reviewed study of the question. Kaiser and Schulze analyzed first-party analytics from 973 e-commerce sites with $20 billion in combined annual revenue, across twelve months of data.

Two findings matter.

The first is volume.

That figure holds up under the most demanding test available: parties with opposite financial incentives report the same thing. Similarweb puts AI referrals at 0.15% to 0.25% of global traffic. Semrush, which sells AI visibility tooling and would benefit from a larger number, reports under 0.15% of visits against organic's 16%. Conductor, a generative engine optimization vendor whose entire business depends on this being big, reports 1.08% of sessions against organic's 25%. Cloudflare's network telemetry agrees. Everyone who would profit from inflating this number reports it as tiny anyway.

The second finding should stop the room. Those referrals converted worse than organic search, not better. Across every channel measured, LLM referrals beat only paid social. They lost to organic search, to email, to affiliate, to direct.

The study did something else worth noting. It reproduced the industry's favorite figures, the ones showing AI traffic converting at 6.7% against organic's 3.9%, and then demonstrated that the ranking inverts once you control for which website, which device, and which season. The gap was an artifact of the comparison, not a property of the channel. The authors' own summary: the results "contradict widespread expectations of oLLM superiority."

The effect that is real is subtraction

Here is the part that does not make it into the deck.

The measurable business impact of AI search today is not that it sends you traffic. It is that it takes traffic you already had.

Pew Research tracked the actual browsing behavior of 900 American adults across 68,879 Google searches. Not a survey. Passive clickstream data, watching what people really did.

Pew is a nonprofit. It does not sell SEO tools, AI optimization software, or anything else that benefits from this number pointing in either direction. That is exactly why it is worth having.

The other number, and what it leaves out

The second statistic you will see quoted is Adobe's: AI-referred shoppers convert 42% better, with 37% higher revenue per visit. In Adobe's May 2026 data both climbed again, to 54% and 53%.

Read the actual reports and three things surface.

The headline is retail. In travel, AI traffic still converts worse than non-AI, by 28% in May 2026. Here I have to argue against myself, because the trend is not on my side: that gap has closed hard, from 86% in October 2024 to 14% in March 2026, before widening back to 28% in May. Adobe's reading is that travel is converging, and on the evidence that is the better reading. One quarter of movement is not a reversal.

Across fifty-three pages, Adobe never once discloses what share of traffic is AI. Every figure is a ratio against an unlabeled baseline: 393% growth, 194% year over year, 54% better, 53% higher. Better than what, on how much, is not in the document.

And Adobe sells an AI optimization product, linked from inside the report.

None of that makes the numbers false. Adobe's data is probably real, and I will take its strongest version rather than its weakest. But apply the arithmetic Adobe leaves out: a 53% revenue premium on roughly 0.2% of your traffic is not a growth channel. It is a rounding error with an excellent conversion rate.

Now the part nobody is honest about

Not one primary source in any of this measures a law firm. Or a medical practice. Or an electrician, a mediator, a consultant, a doula, or any business whose customers live within twenty miles of its front door.

Every usable dataset is enterprise e-commerce, retail, travel, or SEO-savvy software companies. Their core metrics, revenue per visit, average order value, cart conversion, do not exist for a business whose conversion event is a phone call.

The question "does AI visibility produce clients for a one-person professional practice" has not been answered. It has not been asked with data.

That cuts both ways, and I want to be precise about it. It means the optimistic case is unsupported. It also means the pessimistic case is unsupported. Everyone extrapolating from Adobe's enterprise retail panel to your practice is guessing, including the people quoting the bearish numbers. Including, until we measure it, me.

The one thread that points somewhere

There is a finding in the peer-reviewed work that deserves more attention than it has received.

Product complexity changes the entire picture. In high-complexity categories, AI referrals take roughly 4.6 times more traffic share, and their conversion rate exceeds five traditional channels at once: paid social, referral, email, direct, and organic search. The same organic search that beats them everywhere else. The authors' guidance splits cleanly: for simple, routine purchases, treat AI as discovery value and do not move budget away from proven channels. For complex ones, it "already deserves attention," warranting optimization "despite modest current volumes."

Now look at how they drew the high-complexity line. Their categories: heavy industry and engineering, business and consumer services, vehicles, law and government, finance, and jobs and career.

Law and government. Business and consumer services. Finance. The one segment where the peer-reviewed evidence says AI referrals already out-convert organic search is, by the authors' own classification, professional services.

The instrument cannot see the mechanism

Underneath all of this sits a deeper problem, and it may undercut the entire argument in both directions.

Every conversion finding here is last-click.

Consider what actually happens. Someone asks ChatGPT for a family mediator in their county. The model names three. They recognize one, search that name, and arrive through Google. Your analytics records organic search. Or they read the name, pick up the phone, and dial. Your analytics records nothing at all.

The AI did the entire consideration step. It received no credit, because credit goes to the last click, and the AI's contribution never happened on your website.

The study's own authors concede this: early-stage trust verification may shift credit to traditional channels. Which means referral volume, the metric this entire industry is arguing about, may simply be the wrong thing to count.

If AI does the consideration off-site, the honest measurement is not referral clicks. It is branded search lift, direct traffic, and calls from people who already know your name. Nobody has sized that gap. Not the optimists, not the skeptics.

For a local business, the win may never produce a click at all. Being named in an answer, with a phone number the customer dials directly, is invisible to every study cited in this essay.

So does earning an AI citation get results?

Today, for most businesses, the honest answer is that there is no evidence it does, and there is reasonable evidence that AI search is quietly taking clicks you used to get for free. That is the finding. I would rather publish it than sell around it.

But three things are simultaneously true, and holding all three is the only defensible position available right now:

AI referral traffic is currently too small to build a business case on. That is measured, and it is confirmed by sources whose commercial incentives run in opposite directions.

The category where it plausibly works first is high-consideration, trust-gated, comparison-heavy professional services. That is the academic finding, and it is the one segment nobody has tested.

The instrument everyone is using cannot see the mechanism that would matter most. Last-click attribution is structurally blind to a customer who was persuaded off-site and arrived by name.

None of that is an argument for doing nothing. Being clearly described, correctly structured, and consistently corroborated across the web has been the right move for twenty years, for reasons that predate AI entirely. It made you findable. It made you legible to Google. It made you understandable to a human reading a review at eleven at night. AI does not change that logic. It raises the cost of being illegible.

But that is a very different sentence than the one the industry is selling. "Be legible, because legibility compounds" is defensible and always has been. "Earn AI citations and watch the leads arrive" is a claim with one blog post behind it.

What we are going to do about it

We manage a roster of exactly the businesses nobody has studied: a law firm, mediators, a doula, an electrician, a physical therapist. Over the coming months we are going to instrument AI referral, branded search lift, and inbound calls across all of them, and publish what we find.

Including if what we find is nothing.

That would be the only dataset of its kind in existence. And it will be real, which in this category is apparently a differentiator.

Until then, when someone quotes you 23x, ask them where it came from.