The 119

Last week sloppish.com was visited by 5,066 unique IP addresses. One hundred and nineteen of them read more than one article.

That is a ratio of roughly 43 to 1, and it has been stable for a month. We are not publishing it because it is embarrassing, though it is. We are publishing it because we have the raw logs, we checked them ourselves, and almost nobody else who quotes a visitor count can say the same.

The numbers

Four consecutive weeks, same query, same table:

June 22–28: 1,604 unique IPs · 46 sessions read more than one article.
June 29–July 5: 2,432 unique IPs · 57 multi-article sessions.
July 6–12: 4,429 unique IPs · 84 multi-article sessions.
July 13–19: 5,066 unique IPs · 119 multi-article sessions.

We went looking for evidence that our numbers were fake. The data refused to cooperate. Unique IPs tripled, and engaged readers went up too, 46 to 119, roughly in step. The trend is real. Whatever we are doing is working.

The direction is honest. The magnitude is fiction.

How you can tell

Divide hits by unique IPs for any article and you get a number that tells you what kind of visitor you had. A person who reads an article and clicks another produces two or three or six hits. A crawler produces one, then leaves.

Our six most-visited pages last week, by that ratio: 1.01, 1.02, 1.03, 1.04, 1.03, 1.05. The most popular thing we published was fetched 1,364 times by 1,348 distinct addresses, almost all of them once.

Our busiest day had the least reading on it.

The second tell is an inversion. On July 16 we recorded 1,729 unique IPs, our high for the week, and an average of 2.1 hits per session. On July 19 we recorded 250 unique IPs and an average of 6.9. The days with the most visitors had the least engagement. Sort our traffic by volume and you have sorted it by how machine-like it is.

What we cannot tell you

We run a one-pixel beacon that only fires from a browser executing JavaScript, which is a far better human signal than a log line. We cannot give you a week of it, for two reasons, and both are ours.

The beacon tag is not on every page. It is on some articles and missing from others, including our own front page, so any count it produces undercounts by a factor we cannot state. And our access log retains a single day, so there is no week of beacon history to go back to.

We also spent part of this week believing the beacon did not exist at all. A query for it against our analytics database returned nothing, and one of us concluded the mechanism was gone. It was not. The ingest script rewrites the beacon's path to the article's path before storing it, so the beacon is invisible in the database and perfectly visible in the raw logs. The tooling was right. The person checking was looking in the wrong place and reported an absence.

That is worth stating plainly, because it is the same error the industry makes with visitor counts. A number that is easy to query becomes the number you report.

Why publish it

Every publisher running on unique visitors is off by something like this. Most of them cannot check, because they have a dashboard rather than logs, and a dashboard will not tell you that 1,360 of your readers arrived once and never scrolled.

We are a small site. If we said we had five thousand weekly readers, nobody would investigate. The honest number is one hundred and nineteen, and it is growing, which is the part we would have missed if we had only ever looked at the big one.

One hundred and nineteen people read us twice last week. That is a real audience, and it is a different business than five thousand.