Our Own Record #6: 33K monthly users, and 44% of the growth came from two articles about an October price increase

Our Own Record #5 went out on August 19, when the site’s active users over the last 30 days stood at 11,136. Five weeks later, as of September 25, the same metric reads 33,244. Three times. The number moved, so as promised, we record it. The premises of the series have not changed either. The figures come only from Google Analytics 4, Search Console and the operating repository, and neither the site’s name nor any individual’s is disclosed. Readers therefore cannot verify them from outside. That limitation was stated in part 1, and ad revenue figures remain unpublished.

The biggest fact first. Forty-four percent of the growth over these five weeks came from two articles. Both are current-events pieces about a rule change that raises the price of an everyday product from October, and the first of them alone accounts for 20% of all users in the last 30 days. The concentration we described in #5 as “the top 6% carry 63%” did not ease. It tightened.

The numbers as of September 25, 2026

Against the previous 30 days, users grew 1.98x and search clicks 1.76x. Against the period reported in #5, users grew 3.0x and search clicks 2.9x. GA4 restates figures after the fact, so the period we reported as 11,136 in #5 now reads 11,170 when pulled again. The table keeps the value as reported.

By week, active users went from 1,257 in the week of July 20 to 9,066 in the week of September 14. The week of August 17, just after the last report, had 4,929, so the weekly count nearly doubled over these five weeks. The week of September 21 contains a five-day holiday and is still being counted, so it is left out of the chart.

One article is 20%, two are 44% of the growth

We broke the growth down by page. Subtracting each page’s users in the previous 30 days from its users in the last 30 days gives an increment, and each page’s share of the total increment (17,966 users) is its contribution.

The top two carry 43.9% of the growth. They cover the same rule change, and the people reading them searched for a product whose price goes up in October. The first had 7,363 users in the last 30 days, which is 20.2% of the page-level total of 36,372. In the previous 30 days the leading page held only 9.5%, so dependence on a single article doubled in five weeks.

Measured with the same yardstick as #5, concentration looks like this. In the July analysis for #5, the top 8 of 142 published articles (6%) carried 63% of traffic. This time, 965 pages had at least one user in the last 30 days. The top 6% of those, 58 pages, carried 77.8%, and the top 8 pages alone carried 49.2%. We added 95 articles, and the concentration got thicker, not thinner.

We know why. Rule 3 of the seven, “publish current events and disasters within 48 hours,” was applied to a rule change that touches daily life, an October price increase. In #5 we wrote that “current events build the peaks, and the standards hold up the floor.” This is that peak. It will come down once October passes. The first article still leads over the last 7 days, but searches for it should fall off after the day the price changes, and how many 30-day users disappear at that point is what the next record will show.

The Korean edition overtook English

Multilingual publishing was the measure we listed under “what did not work” in #5. In early August we rolled 63 columns out in English, Chinese and Korean, and the English column index was shown 2,131 times in search with zero clicks. Five weeks later, search clicks by language look like this (Search Console page-level totals, 28 days).

Korean grew 13.5x and passed English. Look inside and it is almost entirely one article. The Korean edition of an explainer on a regional tourism support program drew 614 clicks in 28 days at a 7.4% CTR. The Japanese edition of the same article drew 206 clicks at 1.3%. The Korean edition is read three times as much as the Japanese one. GA4 shows the same reversal, 1,097 users for the Korean edition against 899 for the Japanese one over the last 30 days.

Translation did not do this. The program covers travelers from Korea, so people searching in Korean were there from the start. The English index got zero clicks because nobody was searching in English, and we now read that as a demand problem rather than a translation-quality one. What we wrote in #5, that topic selection has to match the search intent of each language market, was correct, but correct in a way we did not picture. We did not aim and hit. One of 63 columns translated in a batch happened to meet the condition. Reproducing it would require a step, before translating, that asks whether anyone searches for this program in that language.

The Chinese edition, at 37 clicks, is unchanged from five weeks ago.

Read on weekday afternoons, at an office desk

Here is a cut we looked at for the first time: day of week, hour and device.

Weekends run at less than half of weekdays. Traffic concentrates between 10:00 and 17:00, eight hours that hold six tenths of a day, and from 2:00 to 4:00 in the morning it drops to around 220 users an hour. By device, desktop is 26,032 users or 79%, and mobile 6,805 or 21%. Daily figures tell the same story. The five-day holiday from September 19 to 23 ran 619 to 741 users a day, less than half of the 1,300 to 1,800 on the weekdays around it.

Those three cuts pin down the reader. Someone looking up programs and subsidies during working hours, on an office PC. This is not something an individual reads on a phone at the weekend. It has become something people look up as part of their job. The plain reading is that the subjects chosen under rule 1, “put money you can receive first,” landed on the research of business owners and accounting staff. That reader profile feeds the next decision. Whether to shift scheduled publication times to weekday mornings is something we will test over the next 30 days and answer with numbers.

Inside search, and the AI channel follow-up

Search Console for 28 days: 4,088 clicks, 149,410 impressions, average position 8.5, CTR 2.7%. Position improved by 1.5 places from the 10.0 in #5. Pages with at least one click number 313, up 108 from the 205 in #5. Pages with at least one impression number 1,476. More pages are being shown, but the pages that get clicked remain two tenths of the whole. The concentration at the top has the same shape on the search side.

Users arriving through the AI Assistant channel, which #5 described as “beginning to appear as a column that cannot be ignored,” number 249. That is 28% more than the 194 of the previous 30 days, but search grew 1.9x, so the share fell from 1.2% to 0.7%. The absolute number grew and the relative one sank. We keep tracking this column.

One figure on the quality of reading. Average session length is 206 seconds, up from 177 in the previous 30 days. The engagement rate is 65.5% (60.2% before). Pageviews per session are 1.10. People read one article and leave. There is no browsing around.

How many articles went in

Published articles went from 242 to 337, 95 more in five weeks. August saw 118 published and September 78 through the 26th. There are 164 scheduled, filling the calendar to December 1. The repository has 63 commits since August 18, 282 in total.

The 401 pages that had zero users in the previous 30 days (new articles, plus published ones that nobody had read) gathered 5,120 users combined in the last 30 days. That is 14.1% of the growth. The 95 new articles contributed less to growth than one pair of current-events pieces. Adding volume and getting the right single article out fast are both “writing articles,” and they act on the numbers differently.

What did not work, and what we do not know yet

The concentration above is also the weakness. A structure where the top article holds 20% is the same structure that loses 20% the month that article’s searches end. The subsidy guide that led the 90-day ranking in #5 is now 10th with 662 users in the last 30 days, slightly down from 684 in the previous 30. Current events always fade. Until we publish the numbers after the fade, we will not call these five weeks of growth real strength.

Multilingual publishing still does not work, apart from the one Korean article. Chinese at 37 clicks is unchanged. Placing translations stops until we have built the screening step that comes before placing them.

And once more, these numbers cannot be checked from outside. The evidence exists only in the GA4 and Search Console dashboards and in a repository that is not public. When the day comes that the name can be disclosed, we will replace this with a version that carries the URL and can be verified. That pageviews and revenue are separate stories is something we also wrote in the tally of 225 AdSense blogs sold on a site marketplace, published the same day. This site’s revenue record will be written when it reaches a stage where it can be.

We recorded a number that tripled in five weeks, midway up the peak. The number after the peak comes down will appear in the same table.

This series

• Part 1: Started four, three broke
• Part 2: The SaaS two of us started broke over capital allocation
• Part 3: 277 commits and 17 languages, and still it stopped
• Part 4: 9,103 monthly users in three months, the content site in real numbers
• Part 5: 11K monthly users. The summer the top 6% carried 63%
• Part 6 (this article): 33K monthly users, and 44% of the growth came from two articles about an October price increase