Two Numbers That Do Not Fit Together

Two numbers, and the space between them

I keep two facts in my head that refuse to sit together, and I have come to think the gap between them is the interesting part.

The first: in a survey published in March 2026, forty-five percent of people said they had asked an AI tool for a local business recommendation in the past year. That was a thousand or so adults in the United States, asked directly.

The second: in a study of real search behaviour published in July 2025, clicks on links inside an AI summary accounted for around one percent of visits to pages where such a summary appeared. That one watched what people actually did across tens of thousands of searches, rather than asking them.

So: people are asking AI tools about local businesses. People are almost never clicking the links inside the answers.

Before anything else, I want to be careful about what I just did, because the careless version of this is everywhere.

Those are two different studies, with different populations, different years, and different methods. One asked people what they remember doing. One measured what people did. The forty-five percent and the one percent are not two halves of the same measurement, and you cannot say "forty-five percent ask, and only one percent of them click." That sentence has no referent and it is not supported by either source. I wrote something close to it myself in a recording earlier this year and had to go back and pull it apart.

Kept separate, they are both interesting. Merged, they are a statistic that does not exist.

What I think is happening, labelled as a guess

Here is where I stop reporting and start speculating, and I will mark the line clearly because plenty of people do not.

My working read is that the AI answer is not functioning as a traffic channel. It is functioning as a filter. Somebody asks a tool for a recommendation, gets a few names, does not click anything — and then goes and looks up one of those names later, in an ordinary search, or a map, or by typing the business name directly.

If that is right, the click never gets attributed to the AI tool at all. It looks like a normal search visit, because that is exactly what it is.

I want to be honest that this is a guess consistent with two numbers, not a finding. I have not measured it. As far as I can tell nobody has measured it cleanly yet, which is why anyone selling certainty about AI search is selling something they do not have.

What it changes about the boring work

If the filter idea holds, then the useful question is not how to rank inside an AI answer. It is whether the information about your business is accurate and readable enough to be picked up in the first place.

Which turns out to be the same unglamorous list it always was. Is your address the same everywhere it appears. Do your opening hours match across your own site and the map listing. Is there a plain sentence somewhere saying what you actually do and where you actually are. Does the page load and can a machine read it without guessing.

That is not a new discipline and it is not an AI strategy. It is the same last mile — small, specific, boring things, close to the door.

Why I build my own tools

There is a reason this topic sits in a campaign about accessibility and small websites, and it is not only that the fixes overlap.

Both of these numbers describe systems that decide what people see, that I have no visibility into and no way to check. I cannot read the criteria. I cannot test whether a change made a difference. I cannot even reliably tell when a visit came through one.

My response has been to build my own things and keep my own records, which is less impressive than it sounds — it mostly means owning the site rather than renting the audience, and keeping analytics I can actually inspect. But it means that when a recommendation system changes its mind about me, I can see what happened on my end rather than guessing.

Where to read the sources

I am deliberately not going to be your source for these numbers.

The forty-five percent comes from BrightLocal's Local Consumer Review Survey, published 10 March 2026: Half of consumers are asking AI for business recommendations. It surveyed 1,002 adults in the United States, and the question asked about the past twelve months. The year before, the same question returned six percent.

The one percent comes from Pew Research Center, published 22 July 2025: Google users are less likely to click on links when an AI summary appears in the results. It followed the actual browsing of 900 adults in the United States through March 2025, covering 68,879 searches, of which 12,593 produced an AI summary.

Both publish their methodology, both state their sample sizes, and both are more careful about their own findings than most of the articles written about them. If you only take one thing from this post, take the habit of reading the method section, because that is where the merged version of these two numbers falls apart.

If you read them and want to talk through what it means for your own site, text me at 503-664-0546 or email eog@ernestofgaia.xyz. I will tell you what I think and I will tell you which part of it is a guess.

Nobody has this figured out yet. I would rather say that than pretend otherwise.

#ernestGoesToAI