On 6 November 2026 Greg Jarboe published a piece in Search Engine Journal that pulls together four datasets on AI visibility. The largest, from Marie Haynes, covered 100 sites and more than 145 million sessions over 12 months. It found ChatGPT sent 0.62% of the volume Google organic sent, and all large language models together sent 0.73%.
The other three point the same way. Kevin Indig ran 632 correlation tests across four brands and found no reliable link between AI visibility and later direct traffic or branded search. SEO Works looked at 95 GA4 properties and found organic sent 39 times more traffic than AI assistants. Ahrefs, across 75,000 brands, found branded search volume correlated only 0.392 with AI Overview brand visibility. Jarboe’s conclusion: “we’ve been grading a brand channel with traffic-channel math, then filling the gap with scores nobody has validated against behavior.” He wants AI treated as a brand channel and measured on exposure, recall and consideration.
The claim worth testing
The first reading of these numbers is that AI search is tiny and can wait. Jarboe doesn’t say that, and I don’t either. He says the measuring is broken, and that AI visibility becomes a boondoggle “the moment we report a number we haven’t checked against what customers actually do.”
That matches what I see. Traffic was always the wrong yardstick for this. On stridec.com, non-branded rankings in positions 1 to 3 got a 0.08% click-through rate between June and August 2026. The AI answer sits on top, the customer reads it and decides, and very little of that decision shows up as a session in GA4. So a dataset that counts sessions will always find AI sent almost nothing. That measures the yardstick more than the channel.
The AI visibility vendors know this, which is why they sell a score instead. The score blends mentions, citations and sentiment across engines, with weights the vendor chose. When it goes up, nobody asks what moved it. Indig’s 632 tests looked for a link between that kind of score and what customers later did, and didn’t find one.
What a score hides
Take a clinic with three branches and a marketing executive who has been handed a dashboard. Last quarter’s visibility score was 41. This quarter it is 58. What changed? Maybe the clinic got named in more answers. Maybe a competitor’s bad review was weighted as sentiment. Maybe the vendor added Gemini to the index. Nobody in the clinic can tell, and the owner is being asked to approve the renewal on it.
Now ask two plain questions instead. When a patient types “dental implant clinic near Tampines” into ChatGPT or Google’s AI Mode, does the answer name us? And of the patients who booked this month, how many said an AI told them about us? Both can be answered this week, with no vendor.
I learnt the first question the hard way. In August 2026 AeroChat ranked third on Google for “best Shopify chatbot” and was not named in the AI Overview. The rank looked fine. The outcome was missing. A month later I went through the sources behind that answer and found the reason. One article named us first, but none of the other sources mentioned us at all, so the AI didn’t either. Ranking and being named are separate. A blended score would have averaged that gap away.
Named, dated, and said out loud
Here is how I measure it for my own brands. Write down the questions a real customer asks when they are choosing. Run them on Google’s AI Overview, AI Mode, Gemini and ChatGPT. For each one, record whether the answer names you, where in the list, and the date. Not whether you were cited as a source. Being used as a source is a leading indicator. Being named is the commercial outcome, because that is what the customer reads.
Date everything. AI answers are regenerated on every search and change week to week. A result from September proves nothing about November. Repeat the sweep on a schedule so the comparison is like for like. When I did this for Stridec on 11 October 2026, 24 checks across four engines, the “consultant” questions did best, and that told me where to put the next month’s work.
The second measure costs nothing. Add “an AI recommended you” as an answer to the question you already ask new customers about how they found you. Put it on the intake form, at the counter, in the first reply to an enquiry. When my own coffee store got an enquiry for 150 custom-labelled bottles on day ten, after being listed first on three of five commercial queries, that was the number I cared about. No dashboard reported it. The customer told us.
What to do
If you are paying for an AI visibility dashboard, keep it for now but stop reporting its score upward until you have tested it. Run your own question set this month and record who got named, with dates. Add the “AI sent me” option to your how-did-you-hear question and count it for a quarter. Then put the vendor score next to those two columns. If the score moves and your two columns don’t, you have your answer about the budget. If all three move together, the vendor has earned the renewal, and you now know what the number means.
Source: Search Engine Journal: 4 Datasets On AI Visibility: ChatGPT Sends 0.62% Of Google’s Organic Traffic