Two years ago your website got more visitors than it does now. Nothing obviously broke. You did not stop publishing, your rankings did not collapse, your customers did not vanish. The line on the chart simply bent downward and stayed there.
That experience is close to universal, it has real measurement behind it, and it is not the same thing as your business getting worse.
Where did all the search traffic go?
Nowhere. People are searching more than they used to, but a steadily shrinking share of those searches ends with a visit to an independent website. Amanda Natividad, VP of marketing at SparkToro, gave the number its plainest form in her MicroConf talk on zero click marketing: for every 1,000 US Google searches, only about 360 clicks reach the open web (MicroConf).
Key takeaways
- SparkToro and Datos clickstream data puts 58.5 percent of US Google searches in the zero click bucket, meaning no click on anything at all.
- Of the clicks that do happen, close to three in ten go to properties Google itself owns rather than to independent sites.
- Search did not shrink. It scattered across search engines, commerce platforms, social networks and AI tools, and the zero click habit applies on every one of them.
- Social platforms actively suppress posts that contain outbound links, so this is not only a Google story.
- Traffic and revenue are different measurements. HubSpot lost the large majority of its organic search traffic and posted record revenue in the same period.
What does zero click actually mean?
It means the search happened and no page got a visitor. Natividad's figures, from SparkToro's clickstream work with Datos covering US Google searches in 2024, split every hundred searches three ways: about 41.5 percent produce one or more clicks, about 37.1 percent produce nothing at all because the browsing session simply ends, and the remaining 21.4 percent produce another search instead. Add the last two together and 58.5 percent of Google searches end without a single click.
The clicks that survive get divided again. Roughly 70.5 percent go to organic results, about 28.5 percent go to Google owned properties such as YouTube, Maps and Images, and around 1 percent go to paid ads. Run those proportions through a thousand searches and you arrive at her headline number of 360 clicks reaching the open web. Nearly two thirds of the demand passing through Google never becomes a visit to anybody's website.
Has search shrunk, or has it scattered?
It scattered. Natividad divides all search behaviour into four buckets: traditional search engines take about 81 percent of searches with Google alone around 74 percent, commerce platforms such as Amazon and eBay take about 10 percent, social networks including YouTube, Facebook and Instagram take about 5 percent, and AI tools take about 3 percent while growing faster than any of the others.
Roughly one in five searches now happens somewhere that is not a traditional search engine. The tempting misreading is to look at the 3 percent and relax, and it inverts her actual argument, which is that the zero click dynamic is universal across all four buckets rather than a Google peculiarity. Amazon has no interest in sending you a visitor. YouTube is engineered to keep people on YouTube. An AI assistant answers in place rather than handing off.
Her State of Search work with Datos through late 2025 found traditional search, AI tools and commerce search had all grown. Search is not dying. As she puts it, the pie is getting bigger while the slice that reaches your website gets smaller.
Why are my impressions rising while my clicks fall?
Because more people are seeing you and fewer of them are clicking. Natividad calls the widening gap between those two lines the alligator graph, with impressions climbing, clicks declining, and the jaws steadily opening. Her framing of the change is worth holding onto: the problem is no longer that nobody sees your marketing.
She backs it with panels she does not own. Ahrefs, looking across roughly 75,000 websites, measured traffic down about 5 percent over eighteen months, and her observation about that sample is the sharp part: those are sites with marketing teams actively trying to grow. A separate Bloom analysis found searches up 15 percent, AI Overviews doubled and traffic down 10 percent, and the Financial Times reported the chief executive of Condé Nast describing the same pattern across major publishers.
Why do social platforms punish links?
Because a link is an exit, and no platform is paid for exits. Meta's own most widely viewed content reports covering 2021 to 2025 show that 97.3 percent of US post views on Facebook land on posts containing no outbound link, and Meta Business Suite now advises business accounts to put their link in the comments instead. SparkToro's own measurement across Facebook, LinkedIn, X and Threads found roughly ten times the reach on posts with no link in them.
This matters because the standard reaction to falling website traffic is to post more links to the website, which is the one behaviour the distribution systems are tuned to bury. Natividad's alternative is to publish value that stands on its own where people already are, and to accept that most of the people who come to know you will never see your homepage.
Which is her reframe, and it is the part worth taking away. The web is not dying. Your homepage moved. It is now a search result, an AI answer, and whatever your customers and former employees say about you in public. We covered the mechanics of that last surface in what is answer engine optimization.
Why can't my analytics see any of this?
Because the measurement layer broke at the same time as the traffic. Only around 30 percent of users accept cookies, Safari rejects third party cookies by default, and between 20 and 60 percent of browsers block analytics outright, with the higher end applying to technically literate audiences.
SparkToro tested this directly, sending more than 1,100 visits from eleven social networks and comparing the real source against what Google Analytics reported. Traffic from TikTok, Slack, Discord, WhatsApp and Mastodon was logged as direct 100 percent of the time, with no referral data at all. Three quarters of Facebook Messenger traffic was hidden, 30 percent of Instagram direct message traffic was hidden, and even LinkedIn concealed the source 14 percent of the time.
Natividad also cites a peer reviewed Dropbox study that ran month long channel blackouts to find out what the spending actually caused. Mobile advertising showed an attributed return on ad spend of 1.53 against a causal 0.7, meaning it was destroying value, and search engine marketing showed roughly 2 against a causal 0.92. Her conclusion from it is that attributed outcomes can overstate causal impact by two to ten times.
The dashboard, in other words, is systematically blind to the surfaces where demand is now created and systematically generous to the last channel before the sale.
If my traffic fell, why might my business be fine?
Because traffic was always a proxy, and the proxy stopped tracking the thing. The example Natividad uses is HubSpot, whose organic search traffic dropped by around 80 percent after Google's algorithm updates and AI Overviews rolled out. In the same window their Q4 2025 revenue hit an all time high and the stock rose. Her conclusion is blunt: traffic and revenue are not the same thing, and judging your marketing by traffic is measuring the wrong quantity.
That is not a licence to ignore the decline. It is a correction to what you conclude from it. A business can lose most of its visitors and still gain buyers, and a business can lose most of its visitors because it genuinely stopped being recommended anywhere. Those two look identical on the traffic chart, and only one of them is fine.
So what should you measure instead?
Four layers rather than one number. Natividad's suggested stack is audience, reach, interest and sales, with incremental sales lift as the primary indicator rather than attributed sessions. You cannot fully attribute any of it, but you can correlate leading indicators against business outcomes over time, which is a weaker claim and a truer one.
Four concrete substitutions follow from her talk:
- Presence in the answer rather than clicks from it. Whether an assistant names you when someone asks for a recommendation in your category is a first impression whether or not it produces a visit.
- Reach on rented land. For every one visitor to your site, roughly a hundred people encounter you on platforms you do not control. Measure that deliberately instead of treating it as overhead.
- One owned channel, kept strong. Her argument for email is its stubbornness: open rates were around 30 percent in 2005 and around 34 percent in 2024, and click rates have sat between 2 and 4 percent for twenty years.
- Ask the buyer. A post purchase question about how someone first heard of you recovers exactly the information analytics deletes.
For a dated record of the first item on that list, SignalCheck runs a live query from a neutral session asking an AI model who it would recommend in your category and city, and shows the raw answer rather than only a score. The value is mostly in reading the sentence that named three competitors instead of you. On timelines for changing that answer, see how long AEO takes to work.
Frequently asked questions
Is zero click search something AI caused? No, AI accelerated it. Rand Fishkin was publishing research showing more than half of Google searches ended without a click back in 2019, years before AI Overviews existed. What AI changed is the ceiling: an answer engine can satisfy a much wider range of questions in place than a snippet could.
If only 3 percent of searches happen in AI tools, why worry about it? Because it is share of searches, not share of revenue or intent, and it is the fastest growing of the four buckets. It is also the wrong reason to relax, since Natividad's central point is that the zero click behaviour applies just as strongly on the other 97 percent.
Should I stop putting links in my social posts? Test it rather than assuming. The measured effect is large, with Meta reporting 97.3 percent of US post views going to link free posts, but your audience may behave differently. The safer rule is to make the post worth reading on its own whether or not a link is attached.
How do I tell a healthy traffic decline from an unhealthy one? Watch enquiries, close rates and revenue per customer alongside the traffic line, and separately check whether you are still being named where people ask for recommendations. Falling traffic with steady revenue and continued mentions is the reshuffle. Falling traffic with falling mentions is a visibility problem.