Your traffic is not traffic

For a year I believed the site had readers. There was evidence. The dashboard showed thousands of people, and the old articles took the largest share. Most came from Asia, which was odd for English posts about construction work in Skåne, but the internet is odd.

That site was never planned. It started as WordPress years ago, and when I moved it to Next.js I did it in a hurry, because other work needed the hours more. The plumbing is right and the site loads. The rest, the part that decides whether anyone arrives, I left alone.

WordPress does a lot out of the box and hides the rest behind plugins. A stack you control can be tuned everywhere. I had tuned none of it.

Then I opened the second tool.

What the numbers showed

Two numbers, one site, the same weeks.

The dashboard reported 13,480 visitors and 182,000 requests.

The other tool counted 287 people.

Thirteen thousand against two hundred and eighty-seven, for the same small site, and both tools were certain.

The 287 had a shape. Most came from Singapore and Vietnam. Seven from Sweden.

They had a favourite page, and it was not one I would have picked: an old article about bathroom mirrors and hygiene robots, which brought 71 of them. The home page brought 15.

Then I looked at how they got there. In a month, six visits arrived from a search result or a link. The other 281 arrived from nowhere. That is not an audience arriving, that is a handful of people wandering in.

So I asked Google what people searched for. In three months the site was shown a few hundred times and clicked once, by somebody typing its own name. Everything else was an old article topic or a misspelling of one.

One click.

Why the numbers do not match

The two tools count different things, and neither of them counts a reader.

The dashboard counts requests. Every time anything asks the server for a file, that is one. A search engine walking 40 pages is 40 requests. A service that checks the site is up every minute is 1,440 a day, all night included. A script looking for a way in counts too. None of it is a person, and all of it is traffic.

The other tool counts browsers. It throws the rest away, which is why its number is smaller. It is not more honest, it just answers a different question.

Google answers a third one: how often a link was shown, and how often somebody clicked it. It is the only one of the three that says anything about what a person wanted, and the only one that is bad news.

Where they were, how they arrived, and what they read said the same thing. The 287 were not an audience.

The site was not popular, and never had been. The dashboard had been describing machines.

What I did with it

I stopped reading totals. A total is the one number that cannot be wrong and cannot be useful.

I started asking each tool for the thing I actually wanted. People, not requests. Clicks, not how often a link was shown. Every one of them has that setting. None of them turns it on for you.

Then I wrote down what each tool is for:

• The server log tells me what is hitting the site, including the noise, and it finds broken links.
• The browser count tells me roughly how many people arrived. It is good for a trend.
• Google tells me what people typed. It is the only one that says why.

Google knows the site well and people do not. It listed 137 addresses returning not found, and two pages I had removed were still being shown to searchers. The counters were wrong about who was reading, and right about a site that had grown a tail nobody was watching.

The last step was the uncomfortable one. If the whole site earns one click in three months, then the archive is not an asset and not a traffic source. It is a shelf. I kept it, because it is still mine and some of it is still true, but I stopped treating it as proof that anyone was reading. Nothing was deleted. It is linked from one place, not in the main navigation, and not on the home page. Its role just changed, from shopfront to shelf.

The rule

A number whose source you cannot name is not a measurement.

Ask what the tool counted before you ask what the number means. Requests, people, and clicks are three different things, and the one that looks best is usually the one counting machines.

And measure before you rewrite. I nearly rebuilt the whole site around a problem that turned out to be a crawler.

And a site that works is not a site that gets visitors. Putting one up is the easy half. The other half is finding out what it does once it is up, and that is the half I had been skipping.

If you want the full set of numbers, with the file each one came from, mail me and I will send it.