I Started a Blog in the Zero-Click Era. Here’s What I’m Measuring Instead of Traffic.

Reading Time: 7 minutes

This morning I spent the better part of an hour trying to find out how many people had visited my own website.

I did not find out.

I started in the obvious place, which turned out to be the wrong place. My newsletter platform has a website dashboard, and it reported five metrics in a tidy row: unique visitors, page views, sessions, bounce rate, session duration. Zero. Zero. Zero. Zero percent. Zero minutes, zero seconds. Above each one sat a small badge reading 100.0% with a downward arrow, which is what a percentage change looks like when both sides of the comparison are nothing.

Then I noticed what I was actually looking at. That dashboard covers five pages — the subscribe page and its immediate neighbours, published twenty-eight days ago. Nobody visits a subscribe page directly. The zeros weren’t a verdict on my writing. They were the sound of an instrument pointed at the wrong object.

So I went to Google Search Console, where the real numbers live. It greeted me with a welcome screen and asked me to prove I owned the domain. I opened my site’s SEO plugin to paste in the verification code, and found one already sitting in the field.

Forty-three characters, beginning with ZoXtkzcx. Not the code Google had just handed me. Somebody had verified this site before — a previous version of me, logged into an account I can no longer identify. The data exists. It has been quietly accumulating. It belongs to a stranger who is also me.

I tried three accounts, then stopped.

I had planned a different piece. It’s the one everybody is writing right now: the traffic collapse memoir. Search referrals fall off a cliff, the author describes the vertigo, and it ends on hard-won wisdom about owning your audience.

There’s a lot of that genre at the moment, and some of it is very good. The chief executive of People Inc. has said Google made up roughly 65% of the company’s traffic three years ago and now accounts for something in the high twenties. Large publishers have built forecasting models for a future in which search sends them nothing at all.

I can’t write that essay. Not because the trend isn’t real, but because I have no cliff to describe. I have a beginning.

After an hour of failing to find my own visitor count, I started to suspect that’s the more useful place to be standing — because the collapse memoir is fundamentally a story about loss, and loss stories carry a buried assumption: that the thing lost was rightfully yours. Traffic from a search engine was never rightfully anyone’s. It was a subsidy, extended for two decades by a company that owed no one an explanation, and now being withdrawn on the same terms.

The people writing that essay have to unlearn a set of habits. I have the odd privilege of never having formed them.

The headline numbers are genuinely alarming and deserve to be stated plainly. Similarweb puts zero-click searches at 69% of all queries. Semrush measured Google’s experimental AI Mode at a 93% zero-click rate, with three-quarters of sessions never leaving Google’s interface at all.

But those aggregates conceal the distinction that actually determines whether any of it applies to you.

Break queries down by intent and the picture separates cleanly. Informational queries — the what is, the how do I, the why does — run at roughly a 74% zero-click rate. Transactional queries run closer to 31%. Analysis of 2026 search results pages found that 99% of AI Overviews are triggered by informational intent, with why questions setting one off nearly 60% of the time. And when an AI summary appears above the top organic result, that result loses about a third of its click-through rate — unless the site in question happens to be the source the summary is citing.

This is why e-commerce operators shrugged through the last two years while writers panicked. They were never in the blast radius.

It also means the honest diagnosis isn’t my writing got worse or the algorithm turned on me. It’s narrower and much less personal: the category of question you answer determines whether anyone still needs to arrive at your page to get the answer.

I write essays about philosophy, AI, and modern life. Every one of them is informational intent. I am, definitionally, in the crosshairs.

Here is the fact that reorganised my thinking. People Inc.’s search share collapsed from 65% to the high twenties — and across the same period, its total audience and its revenue both grew.

Traffic, it turns out, wasn’t the business. Traffic was a proxy for the business, and the proxy broke while the underlying thing kept working.

That reframe changes what an article is for. A blog post used to be a conversion machine: search intent in, session out. For informational writing, that machine is now largely obsolete. What replaces it is slower and less satisfying, but it isn’t nothing.

An essay is now a citation asset. It establishes that you said a particular thing, at a particular time, with a particular degree of rigour. Machines read it. Sometimes they name you. And when someone later searches your name directly, something counterintuitive happens: analysis by Amsive in 2025 found branded queries carrying an AI Overview actually gained click-through rate, by around 18–19%. The summary works as a credential check. It confirms you exist and are roughly who you claim to be, and then the person clicks through to do something.

So the machine doesn’t destroy you if it knows your name. It destroys you if you were only ever a generic answer to a generic question — which is precisely what most search-optimised content was built to be.

The uncomfortable corollary: an entire discipline of writing-to-rank produced exactly the sort of prose that automated summarisation makes redundant. It optimised itself into the shape of its own replacement.

None of these require the data I couldn’t find this morning. That’s deliberate.

Whether I’m cited, not whether I’m ranked. Once a month I search my target phrases by hand and note whether any AI summary names this site. That’s the whole method — a list, kept manually. It’s crude and it doesn’t scale. It’s also the only measurement that touches the thing that now matters, and I’d rather track the right number badly than the wrong number precisely.

Subscribers, counted honestly. Three.

I’ll leave that there without softening it, because the smallness is the point. Three is a real count of real people who made a real decision. Ten thousand sessions is a count of moments when a browser loaded a file. The first number is small and true; the second is large and mostly weather. Given the choice of which to grow, I’ll take the true one. And I’d rather publish the embarrassing figure now, and be held to it, than quietly start reporting once it flatters me.

Whether machines can read me at all. While hunting for analytics I found a warning I’d been dismissing for weeks: seven pages missing custom search metadata, seven missing social sharing images. If the game is being legible enough to get cited, then structured, machine-readable markup isn’t housekeeping — it’s the entry fee. Google has published cases where properly marked-up pages substantially outperformed unmarked equivalents on click-through, including a 25% improvement measured by Rotten Tomatoes across more than a hundred thousand pages. I was failing at the cheapest available layer while worrying about the expensive one.

Things only I can say. Original measurement. First-person experience. Positions I can be wrong about in public, under a name. A summarising machine can compress any explanation I write, because explanations are compressible by design. It cannot compress an account of something that happened to a particular person, because that account doesn’t exist anywhere else to be summarised from.

This essay is the example. Nobody else lost an hour this morning to an orphaned verification code in a WordPress plugin.

I’m not declaring search dead. Google’s chief executive has pushed back on the idea that AI removes the web’s link structure, describing the shift as a continuum in which sources and links remain part of the picture. He has an obvious commercial interest in saying so. That doesn’t make it wrong, and I notice that the incumbent would say that is a way of avoiding an argument rather than answering it. Google’s head of search has separately stated that clicks to websites have held relatively stable year-over-year — a claim publishers dispute, and one worth holding alongside the alarming figures rather than instead of them.

I’m not assuming I’d have won the old game. Semafor’s editor-in-chief has said his company deliberately built around a direct relationship with readers rather than search traffic, and doesn’t expect to be much affected. The honest reading of that isn’t search collapsed. It’s that depending on a single referrer was always structurally fragile, and a great many people mistook a decade of favourable conditions for a durable strategy. With more time, I’d probably have made the same mistake.

I’m not optimising for a number I still can’t see. Somewhere in a Google account I can’t identify, there’s a chart showing exactly how few people have read this website. I’d like to see it. But I’ve noticed that not seeing it has cost me remarkably little, which tells me something about how badly I needed it.

The subscriber count. Not because three is impressive, but because it’s the only figure on any of my dashboards that corresponds to a decision a human being actually made.

The hour I lost this morning was, in the end, the most useful hour I’ve spent on this site. Not because I found anything — I found nothing — but because failing to find it exposed the assumption underneath the search. I went looking for my visitor count on the unexamined belief that it was the number that mattered. It isn’t, and it hasn’t been for some time, and I was fully prepared to spend a year chasing it anyway.

I’ll run these same four numbers again in three months and publish them, whatever they say. If you’d like to see whether any of this worked, that’s what the newsletter is for.