There is a specific feeling that arrives when you open your analytics and the line is pointing down. Not catastrophically, just steadily, month after month, in a way that no single decision explains. Traffic was 4,000 a month last year. It is 3,500 now. Nothing broke. Nobody did anything wrong.
A lot of small business owners have been quietly living with that chart for about a year, assuming it is their fault.
It is mostly not. And the more interesting finding, from agency data covering 54 brands over fourteen months, is that the visitors who stopped arriving were largely the ones who were never going to buy anything. The ones replacing them are worth considerably more per head, which is a different story from the one the chart appears to be telling.
What the data actually says
The dataset comes from Brainlabs and spans 54 clients across roughly fourteen months, from January 2025 through April 2026, which is long enough to cover the period when Google AI Overviews went from occasional to routine.
Organic search sessions fell by an average of 10.5% after AI Overviews reached around 30% penetration of search results in September 2025. That is the number everyone quotes and it is real. If your organic traffic dropped by roughly a tenth over the past year and you have been hunting for the cause, this is very likely most of it.
In the same window, referral traffic from AI platforms rose 163%. Visitors arriving from ChatGPT, Copilot, Gemini, and Perplexity are now a measurable channel rather than a rounding error. More importantly, Brainlabs tracked what those visitors did rather than only counting them, and AI-driven key events, meaning purchases, sign-ups, and other meaningful completions, rose 335%.
The number that reframes everything is the conversion comparison. Visitors arriving from AI assistants completed key actions at roughly 1.5 times the rate of visitors arriving from traditional organic search. Not marginally better. Half again as likely to do the thing you built the page for.
Why AI visitors convert better
The mechanism is not mysterious once you think about what has actually been removed from the funnel, and understanding it tells you what to do next.
Consider who used to arrive from a Google search for something like whether a particular tool integrates with your accounting software. Some of those people were genuinely evaluating a purchase. Many were students, competitors, the merely curious, and people who wanted one fact and left the moment they had it. Your page served all of them equally and converted almost none of them, which is why average conversion rates on informational content have always been dismal.
AI Overviews and chat assistants now answer that question directly. The person who wanted one fact gets the fact and never visits. That is the 10.5%, and the uncomfortable truth is that most of it was traffic that cost you bandwidth and returned nothing. What it did do was flatter your numbers, which is a genuine loss if you have been reporting traffic to yourself as a measure of health.
The person who clicks through from an AI answer is doing something different. They have already had the summary and it was not enough. They want to see the actual pricing, verify a specific claim, or look at the thing itself before deciding. That is a person moving toward a decision, and they arrive pre-qualified in a way search traffic never was. The assistant did your filtering for free.
It is not the same for every industry
The averages hide a wide spread, and where you sit in that spread should change how urgently you act.
Fitness, fintech, insurance, and consumer packaged goods brands experienced some of the largest organic declines in the dataset. The common thread is that these are categories where a great deal of the search volume is genuinely informational, meaning questions about how something works, what a term means, or whether a thing is worth doing. That is precisely the content AI answers completely, so the click never happens.
Retail, beauty, and entertainment saw smaller drops and stronger gains from AI referrals. These categories involve seeing the actual thing. An AI can tell you which moisturiser suits dry skin and it cannot show you the one you will buy, so the summary becomes a step toward the visit rather than a replacement for it.
The useful question for your own business is what proportion of your traffic was arriving to learn something versus arriving to evaluate something. If most of your content answers questions, expect the larger decline and expect the surviving traffic to be better. If most of your content shows a product, a portfolio, or a price, you are in the more protected category. This is the same divide we described in zero-click search and what it means for small business, now with conversion data attached to it.
The metric that is now misleading you
Here is the practical consequence, and it is the reason this data matters more than most SEO reporting.
Sessions have quietly stopped being a useful measure of whether your website is working. A business whose traffic fell 10% while its conversions rose is doing better than it was, and every dashboard it looks at will report the opposite. That is not a small reporting inconvenience, it is a mechanism for making bad decisions: cutting content that works, chasing volume that does not convert, or concluding that SEO has stopped paying when it has actually got more efficient.
The replacement is not complicated. Track conversions by source rather than sessions by source, and specifically separate AI referrals from organic search rather than letting them blur into one channel or, worse, land in direct traffic where they become invisible. Most analytics setups do not do this by default, and a business that has not looked will genuinely not know whether its AI channel is worth anything.
The second measurement worth adding is which pages AI assistants actually cite. If ChatGPT recommends your comparison page repeatedly and never mentions your service pages, that tells you something concrete about what to write next, and it is a signal that did not exist two years ago. Watching your own referrer data is the cheapest version of this and it costs nothing but attention.
What to actually do about it
The strategic response is smaller than the volume of writing on this topic suggests, and most of it is about writing for citation rather than for clicks.
Be the source an assistant quotes. AI systems preferentially cite content with specific, checkable claims, named sources, clear structure, and direct answers near the top of a section. That is not a new discipline, it is the same thing that has always made content credible, applied with the knowledge that the first reader is now often a machine deciding whether you are worth mentioning. Original data and genuine comparison tables perform particularly well because there is nothing else to substitute for them.
Lean into the content types the assistants cannot absorb. A comparison with an actual verdict, real pricing, and a page showing the thing itself all survive, because the summary is not a substitute for them. Purely definitional content, meaning the article explaining what a term means, is the category that has structurally lost and is not coming back. Writing more of it is the most common wasted effort in small business content right now.
And do not chase the lost traffic. The instinct when a chart falls is to produce more of what used to fill it, which in this case means more informational content competing for clicks that AI Overviews now absorb. The businesses doing well in this data are not the ones that recovered their session count. They are the ones that noticed the remaining visitors were better and built for those visitors instead.
The honest limits of this data
Fifty-four brands is a real dataset and it is not the whole internet. These are agency clients, meaning businesses investing enough in marketing to hire an agency, which is a different population from a two-person company with a website somebody built in 2021. Treat the direction as reliable and the exact percentages as indicative.
The conversion advantage also deserves one caveat. AI referral traffic is currently small in absolute terms for most businesses, and small channels frequently show flattering conversion rates precisely because they are small and self-selected. A 1.5x rate on a channel that is 3% of your traffic is genuinely interesting and it is not yet a reason to reallocate your entire strategy. Watch whether the rate holds as the channel grows.
What is not in doubt is the structural point, and it is worth ending on because it is the part that will still be true in a year. The era where traffic volume was a reasonable proxy for business value has ended, and it ended for a good reason rather than a punitive one. The people who left were the ones getting their answer elsewhere, which they were always going to do the moment something made it easy.
The owner watching that line drop from 4,000 to 3,500 has been reading it as a decline. What the data suggests is that she is now getting the same number of people who might actually buy something, from a smaller and less flattering number, and that the chart was never measuring what she thought it was.