Sixty-seven impressions. One click. A$20.79.
We sell stylus pens. Most of our range sits between $20 and $100. So one click, from one person we know nothing whatsoever about, cost us roughly what the product it advertised sells for.
Before anyone writes in: yes, that is one click on one day. It is not our average. Small samples throw up ugly numbers for ordinary reasons.
But it made us ask something we had never properly asked. When you buy a click, what are you actually buying?
We went and found out. We did not enjoy the answer.
We counted our own traffic. Most of it was machines.
We put our storefront behind Cloudflare so every request could be identified before it reached us. Not sampled. Not modelled. Counted, one by one.
First 24 hours:
Our shop software and our analytics disagreed by a factor of six. Shopify said 316 sessions. Google Analytics said 48. Same day, same store, two systems we pay for. One of them was very wrong and neither of them mentioned it.
92 per cent of our UK traffic was running Linux. Of 29,810 requests from the United Kingdom in a single day, 27,300 came from Linux machines. We sell stylus pens to people who own iPads. Desktop Linux is a rounding error in consumer retail. This was almost all of it.
Our number one UK city was Slough. Not London. Slough. A town of 160,000 people, somehow out-shopping the entire capital for stylus pens. The explanation took about four minutes to find. Slough is where DigitalOcean's London data centre is. Our most enthusiastic British customers were racks of rented servers.
Our conversion rate read 0.31 per cent. We had been staring at that number for weeks wondering what was wrong with our product pages. Nothing was wrong with our product pages. We were dividing real orders by a session count that was mostly automated traffic.
Block the data centre networks, and our daily sessions drop from about 280 to 48.
We did not lose 232 customers. We stopped counting 232 machines.
So what was the $20.79?
We do not know.
That is not us being coy. We genuinely cannot find out, and neither can you, and that is the entire point of this post.
Here is what we do know. On the one part of the internet we can actually measure, our own front door, most of the traffic was automated and our analytics cheerfully reported it as people. For months. If that is what the open web looks like when you count it properly, it is a fair question what proportion of paid traffic is the same.
We are not entitled to an answer. That is the bit that should bother you.
You are not allowed to check
Performance Max, the campaign type that served us that click, was for years effectively sealed. You could not see which channels took your money. You could not see which search terms triggered your ads. You could not see where you appeared.
Google has given ground. In April 2025 it announced channel-level reporting, search terms reporting and asset reporting for Performance Max, phased in over the following year. Credit where it is due. That is better.
But read it again and notice the shape of it. Advertisers had to campaign, for years, to be told where their own money went. Basic reporting arrived as a concession rather than a starting condition. Google's own ranking inputs are still only partly disclosed.
Now put that next to what we learned counting our own logs.
The traffic on the internet is substantially automated. The company selling you clicks is also the company counting them, grading their quality, and deciding which ones were invalid enough to refund. Every number in that chain comes from the same place. There is no second meter. There is no independent auditor. You get a figure, and you get to believe it.
Any other industry would call that a problem. In advertising it is Tuesday.
What we are not saying
We are not saying Google defrauded us. We have no evidence of that, and we are not going to invent some to make this a better read.
We are saying three things, and we can show all three:
- A large share of the traffic hitting our store was automated, and standard analytics counted it as human beings.
- That wrecked our session numbers, buried our real conversion rate, and nearly had us rebuild pages that were working fine.
- When we buy a click we have no independent way to verify what we bought, because the seller is also the scorekeeper.
The third one is the one that matters. Not fraud. Something duller and more corrosive: nobody outside the platform is in a position to check, and the platform has no particular reason to invite them to.
Go and check your own shop. It takes five minutes.
Do not take our word for any of this.
Compare your shop platform's session count against your analytics session count for the same day. Ours were 316 and 48. If yours disagree by more than a little, something is being counted that should not be.
Look at the operating system split of your visitors. If Linux is anywhere near the top, those are not shoppers. Shoppers are on iPhones.
We spent months optimising against numbers that were not real. The single most useful thing we have done this year was stop trusting the count and go and measure it ourselves.
The $20.79 still annoys us. What annoys us more is that we cannot find out what it bought, we are not supposed to ask, and the only people who could tell us are the people we paid.