You Can Lose Most of Your Traffic and Be Fine

The number everybody quotes, and the one nobody does

In March 2026, Chartbeat gave Axios two years of data covering thousands of sites. Small publishers, meaning those between a thousand and ten thousand daily page views, lost sixty percent of their search referral traffic. Medium sites lost forty seven percent. Large sites lost twenty two percent.

The pattern is the part worth attention. The smaller you are, the harder you were hit, and the reason is not that large sites optimised better. It is that they already had audiences who arrived without a search engine in the middle. Newsletters, apps, people who typed the name directly. When the algorithmic layer thinned, the businesses with a direct relationship kept their audience and the ones renting attention lost it.

That is a description of most first year businesses. You have no direct audience yet. Everything arrives through an intermediary you do not control.

The replacement did not arrive on schedule either. Referrals from AI assistants grew roughly two hundred percent through 2025, which sounds decisive until you learn they still account for under one percent of referral traffic. A large percentage of a small number remains a small number.

So the honest position, as of the middle of 2026, is that a channel most small businesses relied on has contracted sharply, and the thing replacing it sends almost nobody. If your plan is to wait for AI referrals to fill the gap, the data does not currently support the wait.

Why the clicks stopped

The mechanism is not mysterious. Similarweb measured zero click searches rising from fifty six percent to sixty nine percent between May 2024 and May 2025. Where an AI summary appears at the top of the results, that figure reaches into the eighties.

Pew Research put a finer point on it. When an AI summary was present, users clicked a result eight percent of the time. Without one, fifteen percent. Nearly half the clicks disappear from the presence of the summary alone.

Ahrefs tracked the trend rather than a single moment, which makes it more useful. In April 2025, AI Overviews reduced clicks to the first organic result by about thirty four percent. By December, that figure had reached fifty eight percent. It accelerated rather than settling.

The searcher is not behaving badly. They asked a question, they got an answer, and they had no reason to click. From their side the system improved. From yours, a page that answers a question well now answers it on somebody else's surface.

The finding that changes the conclusion

Here is where most coverage of this stops, and where it becomes misleading.

Adobe examined the 2025 holiday period across a very large volume of US retail traffic. Referrals arriving from AI tools converted about thirty one percent better than traffic from other sources. Separate analysis found those visitors stayed longer, viewed more pages, and bounced less.

The explanation is straightforward once stated. Somebody who arrives after a conversation with an assistant has already narrowed their options. They have described their situation, received a comparison, and clicked through to one of a handful of names that came up. They are not browsing. They are checking.

A second finding matters as much. Businesses cited in AI answers earn substantially more organic clicks than businesses that are not, according to analysis published this year. Being named in the answer does not only prevent loss. It compounds whatever visibility you already had.

Put those together and a specific picture emerges. You can lose most of your traffic and gain business at the same time. Fewer people arrive, each arrival is worth more, and the mechanism that decides who gets recommended sits upstream of the click entirely.

Your analytics will tell you this is going badly

This is the practical problem, and it is the reason a business can respond to this moment by doing exactly the wrong thing.

Analytics measures sessions. Sessions are falling. Every report you open says the same thing, month after month, and the natural response is to conclude that whatever you are doing has stopped working.

But the interaction that determined whether you were recommended never touched your site. Somebody asked an assistant for a supplier in your city, your business was named alongside two others, and the mention did not appear anywhere in your data. It is not a tracking gap you can close by configuring something correctly. The event happened somewhere you have no visibility.

The businesses making bad decisions right now are the ones treating a falling session count as a verdict. They cut the content that is being cited, because the page it lives on shows fewer visits. They shift budget to paid, because paid reports cleanly. Both responses are rational given the numbers on screen, and both make the underlying position worse.

What to measure instead

Three things, and none of them are complicated.

First, enquiries. Not sessions, not rankings, not impressions. How many people contacted you this month, and how many became customers. If enquiries hold steady while traffic falls forty percent, nothing is wrong and you have learned something valuable about the quality of what remains.

Second, how people say they found you. Add one question to your enquiry form or ask it in the first conversation, and write the answer down. This is the only reliable way to capture the arrivals your analytics cannot see. When somebody says they asked an assistant and you came up, that is a data point no dashboard will ever produce.

Third, whether you are cited at all. Ask an assistant the questions your customers ask. What is a good bakery in Henderson. Who does commercial plumbing in the east valley. Do it monthly, note whether you appear and who does, and treat that as a ranking check. It is imprecise and it is directionally correct, which is more than a session count currently offers.

What actually earns the citation

The uncomfortable answer is that it looks almost exactly like good content practice, which means most of the advice you have already heard remains correct. The difference is which parts matter more.

Structure matters more. A page that states a clear answer near the top, uses honest headings, and covers one question properly is easier to extract from than a page that buries the answer under six hundred words of preamble. Assistants read the way a hurried person does.

Specificity matters more. Ranges, numbers, conditions, named exceptions. Generic advice is available in a thousand places and gets attributed to none of them. A page that says what something costs in your city, or what changes at a particular threshold, contains something worth citing.

Third party signals matter more. Reviews, directory listings, mentions on sites you do not own. When an assistant assembles an answer about local businesses, it is drawing on more than your own pages, and a business that exists only on its own website is thinly evidenced.

And your own site matters more, not less, which is the part people find counterintuitive. It is the source everything else reads from. Losing search traffic does not reduce the value of having a clear, complete, well structured explanation of what you do. It changes who is reading it.

The direction of travel

The Reuters Institute surveyed two hundred and eighty media executives across fifty one countries in early 2026. They expect search referrals to fall a further forty three percent over three years. One in five expects losses above seventy five percent.

Those are predictions and predictions are frequently wrong. The useful part is not the figure. It is that the people whose businesses depend entirely on this channel are planning for it to keep shrinking rather than recover.

For a first year business the implication is not despair. It is sequencing. Anything that creates a direct relationship is worth more than it was two years ago, and anything that depends on an intermediary is worth less. An email list you own outranks a follower count. A customer who returns without searching outranks ten who found you through a link. Reviews compound and cannot be removed by an algorithm change.

None of this is new advice. What changed is the margin. Building a direct audience used to be the prudent second priority behind capturing search demand. The numbers now suggest the order has reversed.

What to do this month

Stop reporting traffic as your headline number. Replace it with enquiries and conversations. Traffic is still worth watching as a diagnostic, and it is no longer a measure of whether the business is working.

Start recording how people found you, from the next enquiry onward. Six months of that record is worth more than any analytics configuration, because it captures the arrivals that never appear in a dashboard.

Check whether you are cited, monthly, on the three or four questions that matter most to your business. Note who appears instead of you and look at what those pages do differently.

Then keep publishing, and publish more specifically than feels comfortable. The pages being cited are the ones that answer a real question with real detail. That was always the better approach. It is now the one with a measurable advantage, even if the measurement has to come from somewhere other than your traffic report.