ChatGPT sends 0.62% of website traffic. All large language models combined send 0.73%. That is less than DuckDuckGo.
Those numbers come from a twelve-month study of 100 websites by Marie Haynes, one of three studies Rand Fishkin pulled together in a recent piece on SparkToro asking whether AI visibility is a marketing boondoggle. It is a fair question, and I want to answer it honestly, even though my answer has a conflict of interest attached.
The numbers are small, and they are real
The Haynes data is blunt. ChatGPT visitors converted at a higher rate, but Google still drove 97% of total conversions across the sample, 27 times more than all LLMs combined.
The second study, from SEO Works, looked at 95 websites from September 2025 to August 2026. Organic search sent 39 times more traffic than AI assistants. On 23 of the 63 sites where conversions could be compared, AI traffic converted worse than organic.
The third study is the most interesting. Kevin Indig tested the argument everyone in my industry uses, that AI influences buyers who later search the brand or type the domain directly. Across 12 sites in different sectors, he could not find a reliable link between AI visibility and branded search or direct traffic. Not yet, at least.
I have a bias, so here it is
I am CMO and co-founder of GEOflux.ai, which measures how brands appear in AI answers. So you should read the rest of this knowing I have a stake in the outcome.
Here is the part where I agree with Rand completely. If anyone is telling marketers to move most of their budget into AI visibility because it will replace Google traffic, the data says they are wrong. Google remains the dominant traffic and conversion channel for almost every business I see, and the clients we work with at difrnt. are no exception.
Any vendor in this category that sells AI visibility as a traffic play is setting customers up for a painful quarterly review. I would rather say that out loud than defend it.
Wrong channel, wrong ruler
Where I disagree is the measurement frame. Mike King made the point that AI behaves more like a brand channel, a digital billboard, and I think that is the right model.
Nobody measures a billboard by clicks. A billboard on a highway sends zero visitors to your website, and nobody concludes it had no effect. You measure it by reach, by recall, by whether the brand shows up when the buyer builds a shortlist.
AI answers work the same way. A procurement manager asks Claude for five vendors in a category. Your brand is on the list or it is not.
If it is, the manager probably does not click anything. They copy the names into an email, ask colleagues and search later, or call someone they already know.
That influence will not show up as AI referral traffic, and it may not show up cleanly in branded search either, which is exactly what Indig found.
The honest conclusion is not that AI visibility has no value. It is that we do not yet have clean attribution for it, which is also true of almost every brand channel ever invented.
How I would measure a billboard you cannot see
If AI is a brand channel, then the metrics should be brand metrics. I would track four things, monthly, against a fixed baseline.
Share of answer: across the 30 to 50 questions your buyers actually ask, how often does your brand appear, and in what position?
Accuracy: when the models describe you, are the facts right? Wrong prices or outdated products cost more than absence.
Shortlist company: which competitors appear beside you, and who replaces you when you drop out? And in your own sales process, ask every new customer where they first heard of you, with AI assistants as an explicit option.
That last one is cheap and underused. It is the same self-reported attribution that has always caught what pixels miss, and I wrote about why AI mentions are not the same as AI trust, so the qualitative answer matters as much as the count.
This is how we built GEOflux.ai to report it: presence, position and the sources behind each answer, as a visibility baseline rather than a traffic forecast. If you want to measure where your brand actually stands in AI-generated answers, that is the problem it is designed for.
How much budget it actually deserves
Here is my practical answer. Treat AI visibility as a share of your brand and content budget, not your performance budget. For most mid-market companies, that means a measured slice, not 80% of the team's attention.
The good news is that most of what improves AI visibility also improves everything else. Clear product pages, accurate third-party listings, real reviews, being cited by credible publications. I made the case earlier that AI visibility starts with human receipts, and that work pays off in Google too.
What does not deserve budget is anything sold as a shortcut. Prompt manipulation, fake mentions, content written only for models. That is the same mistake the industry made with link farms, and it will end the same way.
The studies are right about traffic. They are silent about influence. Measure the billboard like a billboard, and you will stop being disappointed that it never sends clicks.