Our Own Record #5 | To 11K Monthly Users — the Winning Line Was the "Top 6%": the Summer We Turned Analytics into Topic-Selection Rules

This is a follow-up on the tax-and-money content site covered in Our Own Record #4. The last report (as of August 12) closed at “9,103 users in the last 30 days.” This time, beyond how the numbers moved in the week since, I want to record how a July round of analytics was converted into rules that apply before writing. As the series promises, only measured values from Google Analytics 4, Search Console, and the repository’s article metadata are used, and no service or personal names appear (which means outside verification is impossible, a limit stated in part 1 of the series).

The numbers as of August 18, 2026

Weekly, active users rose from 113 in mid-June to 3,331 in the first week of August. Even in the single week since the last report (226 articles, 9,103 users/30 days), the site moved to 242 articles and 11,136 users/30 days. 85% of traffic is search (9,452 users), plus 1,161 direct, and 173 users from an “AI Assistant” channel. AI-referred traffic appearing as a column too large to ignore is the new observation this time. It is still only 1.6% of the whole, but unlike search, where AI increasingly summarizes before the click, people arriving through this channel come with a clear purpose. Whether the number grows is something this series will keep tracking.

Look one level deeper into search: clicks rose 11.7x against 10.1x for impressions, and average position improved from 15.7 to 10.0. The growth therefore goes beyond “more exposure” to “ranking higher for the same exposure.” The period when a new domain’s evaluation ramps up overlaps with the topic-selection change described below, and the two contributions cannot be separated, I want to be honest about that.

The finding: the top 6% carried 63%

In July I ran an analysis joining 90 days of GA with Search Console. Laying out the breakdown of the then-current 142 published articles and 5,102 users, the top 8 articles, a mere 6% of the total, accounted for about 63% of traffic. Put the other way, 134 articles were “essentially unread.” Seeing that skew, I shifted the focus of improvement from “how to write” to “what to write.”

Out of the analysis came seven rules for topic selection. In outline: 1) always put “money you can receive” (subsidies, benefits, rebates) first in line, 2) when a theme hits, expand it horizontally by region and fiscal year, 3) publish current-events and disaster topics within 48 hours, 4) fill unwritten gaps among the “standard systems people look up when in trouble”, 5) time practical procedure-and-paperwork pieces to the season, 6) don’t write generic big-keyword explainers, 7) cap the number of articles in low-PV differentiation clusters. These were codified in the repository’s editorial-policy file, and every topic since is judged in that order.

How the rules showed up in measurements

Rule 2 (horizontal expansion) shows the clearest result. A single article on electric-vehicle subsidies reached 1,851 users in 90 days, the site’s top article, but instead of stopping there, it was expanded into a by-prefecture series. As a result, the prefecture articles now line up as a band near the top of search clicks (Kyoto 38, Saitama 32, Shizuoka 31, Chiba 29…). One hit can be luck. A hit surface built by expansion is reproduction.

Rule 3 (48-hour current events) also worked. An explainer on the tax deduction for disaster relief donations led the last 30 days with 1,067 users, and a related piece on “how tax money is used” posted an outlier search CTR of 13.6%. An explainer on the liquor-tax revision drew 922. These trend articles, however, also decay fast. Just as the subsidy guide that led the 90-day ranking has slipped to 5th over the last 30 days, the numbers show a two-layer structure: current events build the peaks, and the standards (high-cost medical care 917, dependents explainer 981) hold up the floor.

What did not work, recorded too

In early August, 63 columns were bulk-translated into English, Chinese, and Korean (985 URLs added). So far no clear result. The English column index has been shown 2,131 times in search with 0 clicks. It gets displayed but not chosen. Multilingualization is not a “place translations and traffic comes” play. It needs topic selection matched to each language market’s search intent, the very lesson of rule 4 in Japanese, apparently to be relearned per language. Too early for a verdict, but at minimum there was no quick effect.

The July finding that “generic big-keyword pieces stall at 20–30 users” has also been held since as rule 6. The act of not writing leaves no trace in the numbers, but given the prehistory in which 134 of 142 articles went unread, deciding not to write is the largest time-saver of all.

Per this series’ policy, ad-revenue figures are again not published here. The record of traffic and the record of revenue are separate stories. When the latter can be published, it will be written then.

What transfers, and what doesn’t

Three things are portable. 1) For a small site, analytics, even on a small base, yields the answer to “what to write” early: in our case, the moment the concentration into 6% became visible, the material for rule-making was already there. 2) Horizontal expansion of a hit is the smallest editorial strategy a solo operation can run. 3) Record the measures that did not work as well, the 2,131 impressions and 0 clicks of the multilingual bet are base data for the next decision.

The limits, stated plainly: much of this growth comes from having picked “money you can receive,” a domain with strong search demand, in the first place. The same multiples will not appear in every genre. The weekly figures include a week still being tallied, and August’s peak includes a current-events tailwind. When a down month comes, that will be recorded too. On how genre choice shapes revenue structure, see the blog PV-unit-economics case. For the rest of this media-operations record, see Our Own Record #4.