Nobody Reads Your Blog. The Machines Never Stopped.

01The dashboard is not lying. It is answering a different question.

Open the analytics on a small site you have kept up for a year and the picture is quiet in a way that is hard to argue with. Four visitors yesterday. Two of them were you, checking that the deploy went out. The honest conclusion looks obvious from there, which is that nobody wants the thing you wrote and you should stop writing it. I came close to stopping.

Your last visitor left an hour ago. The machines never left.

What analytics is physically able to see

Analytics is a piece of script that runs inside a browser and reports back. A machine reading your page usually does not run it, and no amount of configuration changes that, so the entire machine readership is absent from the chart by construction rather than by accident. The chart is not wrong. It is a census of one species conducted with a method that can only detect that species, which is the same reason the bot share never shows up in the number you quote at yourself. Analytics counts people. It was never built to count influence.

The number that changed my mind

I stopped reading the dashboard for a fortnight and read the server log instead. On one site of mine a single assistant opened a page thirteen times a day, every day, without missing one. Human visits over the same fortnight sat in single figures a day. Click throughs from that assistant back to the site were zero, and that last part is the one that took me a while to swallow, because zero looks like failure until you work out what the fetches were for.

It is worth being honest about what a fetch is not. A machine opening a page is not an endorsement, and some of those visits are scrapes that lead nowhere at all. A fetch tells you that you are reachable and being read, which is a floor rather than a result. A floor was exactly what I had been missing, because the dashboard had spent a year telling me the floor was zero.

02One day, one site, two audiences

Here is a single ordinary day on that site drawn as a strip of twenty four hours. The top row lights up when a person opened a page. The bottom row lights up when one assistant did. Both rows are the same day and the same site, and only one of them has ever appeared in a report I looked at.

How to read the strip

The human row has three lit hours and they cluster where you would expect, in the middle of somebody’s waking day. The machine row is spread flat across the whole clock, including the hours when the country the site is written for was asleep. There were three hours with a person in them and thirteen with a machine, and only one of those audiences was in the report. The machine row is also one assistant on its own. Every other crawler is off the chart entirely.

03Why publishing into an empty room still pays

The reflex is to treat a machine fetch as a wasted page view, a robot burning bandwidth on something meant for a human. That reading is exactly backwards, and it costs people years of work they then throw away.

You are not writing for a reader. You are writing for the answer.

The machine is not an audience. It is a courier. It arrives, takes a sentence, and carries it to somebody who never sees your site and never will, and that person makes a decision with your sentence sitting inside it. The zero in the referral column is not evidence that nothing happened. It is what a successful delivery looks like from your end. A page nobody visits can still be the reason a stranger buys something. You will not get a click for it and you were never going to.

None of this makes the page any less something a person should want to read. The sentence a machine lifts is the same sentence a person reads, so writing for machines and writing badly are not the same activity and the first has never required the second. What changes is narrower than it sounds. It is only which number you allow to tell you whether the work was worth doing.

What to measure once you accept that

Three things, and none of them are in your analytics. First, whether you are being read at all, which your server log answers tonight and no dashboard ever will. Second, whether you get named when somebody asks about your category without saying your name. Third, whether the sentences you wrote come back with the numbers still attached, which is the whole argument for putting a figure in the claim in the first place. Stop asking how many people came. Start asking how many answers you are inside.

Tomorrow: twenty one small sites that vouch for each other, and why that beats one big site vouching for itself.