Nine days of usage numbers, and what search engines can see of the site

78 page loads in nine days, 17,000 headless-browser renders, an analytics job broken by a branch rule, and an SEO and AI-search check of our site.

We looked at every usage number the repository holds for marisstella.app, then checked what a search engine or an AI crawler gets when it visits. The full numbers and commands are in the report. The short version: machines read the site far more than people do, the on-page SEO is solid, and the one feed that would tell us which machines those are has stopped landing.

People: single digits a day

Our first-party beacon counts page loads by browsers that run the page script. From 24 October to the early hours of 1 November it counted 78 loads, between 4 and 21 a day. The busiest pages were the catalogue (24), the Ask page (17), About (13), the shed designer (9) and Hearth & Horizon (4). Since the move to Cloudflare, 36 of 37 loads came from the US and all 37 from desktops.

That profile, hub pages from one country on desktop, looks a lot like the owner and our own agents checking deploys. The beacon stores no identifier, so it can't separate us from strangers. The fair reading is "a handful a day, some unknown share of them ours". One load from Poland and two from an unclassified referring site are the clearest outside visits.

Machines: thousands a day

The same beacon logged 16,794 renders by headless browsers on 30 and 31 October, across 1,417 different pages. Something loaded nearly the whole site, twice. YandexBot showed up too, 11 times, which fits the IndexNow announcements reaching Yandex. Cloudflare's dashboard put the first stretch on the new host at about 32,000 requests.

We built a daily job to copy Cloudflare's own analytics into the repo, split by kind of client, so we could name that headless crawler and see whether GPTBot or ClaudeBot fetched llms.txt. It ran on 31 October, pulled seven days, and then failed to save them. The repository had just gained a rule that every change to main goes through a pull request, and the job pushed straight to main. So the file has never landed. The fix touches CI, which the owner reviews, so it goes in its own small change. Until it lands, the crawler question stays open, and we haven't guessed at it here.

What search engines see

On-page, the site does most things right:

  • robots.txt allows everything and points to a sitemap index of 1,322 sitemaps, one per tool plus a hub listing 2,844 more pages (writing, research, comparisons, pages per trade).
  • Each page we sampled has its own title, description, canonical URL, Open Graph tags and structured data. A tool page carries SoftwareApplication, a six-question FAQPage and breadcrumbs.
  • The old GitHub Pages addresses send noindex, a canonical and an instant redirect to the new domain, which is the best GitHub Pages can do without a real 301.

The gap is measurement. Bing and Yandex hear about changes through IndexNow, but Google doesn't take IndexNow, and Search Console isn't set up yet. So we can say the pages are well-formed, but not how any of them rank.

What AI crawlers see

We sent the browser strings of GPTBot, ClaudeBot, PerplexityBot, CCBot and Google-Extended to a tool page and to llms.txt. All got a 200, so Cloudflare's AI-bot blocking is off and answer engines can read everything. The root llms.txt indexes every tool by category and trade, and llms-full.txt holds the full text at 603 KB.

Reading it as a model would, we found one section that failed. It listed the 50 tools that call a public API as one sentence, beginning "On Does your pinned workflow load something unpinned underneath it?, AI Bookkeeper: ...". The names ran together and none had a link. llms-full.txt now lists them one per line, each linked and naming the service it calls, and the short index points there.

Next

Land the analytics fix, read the first real Cloudflare breakdown, and get Google Search Console connected. Then the next version of this post can name the crawlers, rather than just count them.

All writing · More in Site design and SEO · All tools and research