Mostly Machines
This blog’s counter says 336,000 views. Most of them were not people. What three tools show, why they disagree, how to count better, and my own estimate of how many humans actually read here.
Blogger tells me this blog has been read 336,000 times. The line on its chart sat near the floor for a decade, began to climb in 2023 and, this October, went almost vertical: 20,900 views in thirty days, with a single day of about 5,000 around the ninth.[1] For a blog that began in 2008 as a notebook and spent most of the 2010s as a runner’s diary, that looks like success.
It is mostly not. When I opened the other two tools that Google offers for a site like this one, Google Analytics and Search Console, the same four weeks looked very different: 880 views by 361 visitors, and about twenty clicks a month from Google searches. Somewhere between the two numbers are the people who actually read what is written here. This post is an attempt to find them, with my own numbers, their limits, and what can be done to count better.
The cover is the Turk, the chess-playing automaton that toured Europe from 1770: a machine that seemed to think, with a human chess player hidden in its cabinet. A blog counter is the reverse. It shows what look like readers, and inside most of them there is a machine.
Three counters, three different questions
Blogger’s own statistics count every time a page is served. It does not matter who asked for it: a person on a phone, a search engine indexing the page, a company collecting text to train an AI model, or a program probing the site for weaknesses. All of them count as views. The page shows where visits came from, which browser and system they used and which country they were in, but it has no way of telling a person from a program that says it is one.
Google Analytics works differently. It counts a visit only when a small script runs inside the visitor’s browser, which most simple programs never do, and it removes traffic from bots and spiders that Google already knows about. That filter cannot be switched off, and Google does not say how much it removes.[2] Analytics also measures how long a page stays open and in front of the reader, which is the closest thing these tools have to “did anyone read it?”
Search Console answers a narrower question: how often Google showed a page in its results, for which searches, and how often someone clicked. It does not count visits from anywhere else, but what it counts is real.
Who the machines are
Bots are not peculiar to this blog. Imperva, a security company that watches traffic across many large sites, reported that in 2024 automated traffic overtook human traffic for the first time in a decade, at 51 per cent of all web traffic, and that bad bots alone made up 37 per cent.[3] A small blog, with few human visitors to dilute them, can expect a larger share.
Several things in Blogger’s figures for the last thirty days point the same way:
Part of the recent climb is probably the companies that collect text to train language models, whose crawlers have multiplied since 2023; how that text is then used is explained in How a Language Model Learns. A blog that published dozens of long essays in two months is a rich and fresh source. I cannot see from these tools which crawlers came, only that most of the views behave like machines.
Who the people are
In Google Analytics, the four weeks from 30 September to 27 October show 880 views by 361 visitors, who spent an average of 1 minute 28 seconds with the blog. Most entered through the front page (214 views, 32 seconds each) and left. A few pages were read at length:
Two pages show one visitor each, browsing a whole series: the Memórias label (19 views, 5 minutes) and the AI label (13 views, almost 5 minutes). That is what a real reader looks like. It is also what I look like, checking my own posts.
And a large part of that long reading, it turned out, was mine.
Search Console adds the view from Google: 59 clicks in the three months to 27 October, about twenty a month, and 369 pages in Google’s index. Its one recommendation was that a 2022 race report, on the Backyard Ultra World Team Championship, had suddenly been shown four times as often as usual. The running posts are still what strangers find.
And then there are the other running blogs. Eight of them, from ultkm to trilhos míticos, sent 770 views this month according to Blogger, through the lists of “blogs I follow” that show each new post. Over all time, Google sent about 20,800 visits and Facebook about 13,800. The 283 comments, none of them this month, are the surest sign of all that someone was there.
The most faithful reader
Google Analytics can list its visitors one by one, each with a number that it also keeps in the visitor’s browser, in the small file called a cookie. Sorted by activity, two visitors stood far above the rest. Everyone else made at most 25 events in the four weeks; these two made 753 and 396.
The first had been placed in six towns in Portugal, some of them places I had not been to, so at first it did not look like me. The cookie in my own browser settled it: the same number. The second, which Analytics placed in Portugal without naming a town, began its visits on 27 October a quarter of an hour after the first stopped, while I was finishing a post. It was my phone.
The reading time is the striking number. Other visitors made most of the visits, but almost all the time spent reading this blog last month was spent by its author: checking a post after pasting it, rereading, looking up an older one through the search box. That is what a writer does in a month with dozens of new essays. It is also what made the long reading times in the table above look like an audience.
How many people, then
Starting from Google Analytics, because it is closest to counting people, I made three corrections, and one assumption for the whole history, each with a range rather than a single number:
So, in the busiest month in the blog’s history, somewhere between 160 and 210 people read it, opening between 160 and 320 pages: most opened one page, perhaps two, and left. That is about 5 to 10 page views a day. For a new essay that means, in its first weeks, between a handful and a few dozen readers. Since Blogger began counting, the blog has probably been read 53,000 to 89,000 times by people, most of them in the running years, when the share of machines on the web was smaller. These are rough numbers. The point is the order of magnitude: hundreds a month, not tens of thousands.
What the estimate cannot see
Each correction can be wrong in either direction, and some of the numbers disagree in ways I cannot yet explain.
How to count better from now on
None of this can be corrected backwards. Google Analytics does not rewrite its history: its filters, by its own documentation, “do not affect historical data”.[6] What can be done is to make the next months cleaner.
ipconfig show the address inside the local network, which Google never sees. And home addresses change. To find out whether a frequent visitor is me, I open the blog, press F12 and type document.cookie in the Console: the number after GA1.1. in the _ga cookie is the one that Analytics shows for that browser in Explore, User explorer.What the totals say over time
Blogger’s overview page adds a few numbers that help to see past this month.[1]
Seen over the whole history, the two counters tell different stories. Blogger’s line rises steadily from a few hundred views a month in 2011 to about a thousand in the running years, climbs again from 2023, and leaves the chart in 2026. Analytics, which only began counting in January 2024, moves between about 150 and 1,500 views a month: higher in the months when I published most and read most myself, and always far below Blogger. In 2024 it saw one view for every six that Blogger counted; in 2025, one for every eleven; in October 2026, one for every twenty-two. The readers who are not me sit lower still.
Read backwards, the widening gap suggests that the early numbers were more honest than today’s. They were not the truth, though. In 2013 Incapsula, a security company later bought by Imperva, counted every visit to its 20,000 client sites over 90 days and found that bots made 61 per cent of them.[8] A year later it put bots at 56 per cent of visits overall, and at 60 to 80 per cent on smaller sites,[9] the group this blog, at twenty or forty views a day, belonged to. Not every one of those bots would have reached Blogger’s counter, but many did. Split by period, the estimate above says that between a sixth and three-fifths of Blogger’s views from 2011 to 2022 were people, against about one in twenty now. Closer, then, but not close.
Part of that rise follows the writing. I published about 128,000 words in 2025 and close to a million in the first ten months of 2026, eight times as much. The views in Blogger rose with the words: more pages, more for the machines to read. The readers and the comments did not.
The clearest sign of people, though, is the comments. Of the 283 on the blog, 217 were written by 69 other people and 66 are my replies. They are not spread evenly over time. Between 2011 and 2016 other people left 160 comments on 95 posts. Between 2020 and 2026 they left 17, on 390.
My own reading of this is that people stopped following blogs, and now prefer shorter things. The years fit, though not neatly. The comments peaked in 2012 and 2015, when the blog was one of a circle of running blogs whose authors read each other and answered each other, and when a reader like Google Reader brought every new post to the people who followed it. Google closed Reader in July 2013, saying its use had declined.[10] The comments held up for some years after that, with good years in 2015 and 2019, and then faded. Since then, links to the blog have mostly travelled through Facebook, which has sent about 13,800 visits over the years; whatever conversation followed those links happened there, not here. And the move towards short video is measured: across the markets studied by the Reuters Institute, the share of people who watch social video for news each week rose from 52 per cent in 2020 to 65 per cent in 2025.[11]
There are fair objections. Long text has not lost its readers: the same survey finds that most people still prefer to read their news rather than watch it, 55 per cent against 31,[11] and Substack, which delivers long writing by email, passed five million paid subscriptions in March 2025.[12] The readers moved to the inbox and the feed more than they left reading. And the blog changed too. The race reports were written for a community that ran the same races and knew the people in them; the essays of 2026 are long, mostly in English, about economics, biology and AI, and arrive at a pace of several a week that few readers could keep up with, let alone comment on. Both things are probably true: the medium moved, and so did the blog.
Fewer, and real
The honest count is smaller than the one on the dashboard by a factor of about a hundred: fewer than two hundred and fifty people a month, most of whom open one page and leave, and only a handful who stay for minutes. The most faithful reader of all, this month, was me, on two devices. Some of them have been here since the training cycles and race reports, and still arrive through the lists on other runners’ blogs. Others find a note on the ITRA index through Google and never come back.
The machines are not without consequence. Some of them index the blog so that people can find it; others take its text to train the models that will answer questions in its place. But they do not read. A couple of hundred people who do are worth more than a counter that says 336,000, and the counter will be easier to believe next month, when I am no longer among them.
This piece was written collaboratively with Claude Opus 5.5 (Anthropic): human specification and editorial direction, and the screenshots of the three tools; machine analysis and drafting. The numbers were read from screenshots of Blogger statistics (all time and last 30 days), Google Analytics (home page and Pages and screens, 30 September to 27 October 2026) and Search Console (overview, 27 July to 27 October 2026), all taken on 28 October 2026. Blogger’s overview page (totals, followers and the most-viewed posts) was read on the same day, and comments were counted by year from the blog’s public comment feed, without naming the people who wrote them. The monthly series in Figure 5 were read off the charts of Blogger and Google Analytics, which offer no export of them, and checked against their reported totals. The estimates come from a script, scripts/readers.py, which holds those numbers and the assumptions stated in the text. The most frequent visitor was identified by comparing its number in Analytics’ User explorer with the _ga cookie in the author’s browser. The ad-blocker figure is the GWI survey as reported by Backlinko; the original report was not seen.
Cover: the Turk, from Karl Gottlieb von Windisch, Briefe über den Schachspieler des Hrn. von Kempelen, 1783, via Wikimedia Commons, public domain, cropped.
Authored by: Luis Matos Ferreira — Physicist, Developer, Writer
- The Paths Through This Blog — eighteen years of posts, and the themes that run through them.
- Os Caminhos Deste Blogue — the same guide, in Portuguese.
- How a Language Model Learns — what the text collected from the web is used for.
- The Manufactured Public — fake audiences on a larger scale: influence operations and their bots.
- ITRA Performance Index — the post strangers find most often through Google.
- dorsal1967, Blogger statistics (all time; last 30 days), Google Analytics 4 (30 September–27 October 2026; calendar year 2025) and Google Search Console (27 July–27 October 2026), screenshots taken on 28 October 2026; own calculations in
scripts/readers.py. - Google, Analytics Help, “[GA4] Known bot-traffic exclusion”, retrieved 28 October 2026.
- Imperva, “2025 Imperva Bad Bot Report: How AI is Supercharging the Bot Threat”, 15 April 2025.
- StatCounter Global Stats, Desktop vs mobile vs tablet market share, Portugal, September 2026 (mobile 49.5 %, desktop 49.4 %), retrieved 28 October 2026.
- GWI, ad-blocker use among internet users, Q2 2025 (29.5 % use one at least sometimes), as reported by Backlinko, “Ad Blocker Usage and Demographic Statistics”.
- Google, Analytics Help, “[GA4] Data filters”, retrieved 28 October 2026.
- Google, Analytics Help, “Google Analytics opt-out browser add-on”; add-on at tools.google.com/dlpage/gaoptout.
- Incapsula, 2013 Bot Traffic Report, as reported by phys.org, “Incapsula reports that web bots now account for 61% of web traffic”, 13 December 2013.
- Imperva (Incapsula), “2014 Bot Traffic Report: Just the Droids You were Looking for”, 18 December 2014.
- Google, “Powering Down Google Reader”, Google Reader blog, 13 March 2013, as reported by Search Engine Land/MarTech, “Google To Close Google Reader On July 1”, 13 March 2013.
- N. Newman et al., Reuters Institute Digital News Report 2025, “Executive summary and key findings”, 17 June 2025.
- Tubefilter, “Substack surpasses five million paid subscribers”, 12 March 2025.
Comentários
Enviar um comentário