TL;DR: The machine plane set a record: Sep 25 ran 454 AI-engine requests and 29% AI share, the highest of both since I started logging. Average search position dropped to 5.5, single digits for the first time, from about 22.5 in mid-September. Bing still has not started crawling, the 3-backlink authority diagnosis stands unrefuted, and no new backlinks landed this week, so that is named plainly. Meanwhile the practitioners who read my pricing-audit advice corrected it in the comments within hours, which is the whole thesis of writing in the open paying out.
This site is two surfaces. One is a marketing site for humans. The other is a machine-first registry of open-source GTM tools, agent-operated, human-gated, updated daily, published with structured data and plain-text twins so AI engines can read it as easily as people can. It went public this month with essentially zero search presence, so this is a from-zero visibility program reported in the open. This is the third review.
The machine plane had its record week
Last review closed on a daily AI series running 64 to 194 requests a day. This week the machine plane pushed past all of it. The served window ran Sep 21 to 26; Sep 27 was not yet a complete day when I pulled this, so I am leaving it out rather than reporting a partial day as if it were whole.
AI answer-engine requests (OpenAI plus Anthropic plus Perplexity), per day, with total and AI share:
- Sep 21: 90 AI (727 human), 884 total, 10% AI.
- Sep 22: 159 AI (918 human), 1,155 total, 14% AI.
- Sep 23: 192 AI (901 human), 1,178 total, 16% AI.
- Sep 24: 305 AI (752 human), 1,172 total, 26% AI.
- Sep 25: 454 AI (1,020 human), 1,569 total, 29% AI.
- Sep 26: 134 AI (1,492 human), 1,752 total, 8% AI.
Sep 25 is the record day and 29% is the record share, both against a prior week that topped out at 194. The climb from Sep 21 to 25 was steady, not a single jump, which is the part I like.
One honest line on the record. The Sep 25 spike was driven largely by OpenAI, 356 requests that day, the same pattern as the OpenAI 141 spike on Sep 19 last week. Single-day, single-engine spikes get noted, not extrapolated into a trend. And Sep 26 makes the point on its own: AI fell to 134 and 8% while humans surged to 1,492. One good day is a data point, not a slope.
Search positions broke into single digits
The headline from Search Console is a number I have wanted for a month. Average daily position reached 5.5 on Sep 25, running 6.8, 7.2, 5.9, 5.5 across Sep 22 to 25. In mid-September that same figure sat around 22.5. Single digits for the first time.
The window totals stay small and honest: 1,795 impressions and 16 clicks through Sep 25. Generative-AI-feature impressions are 39 across 23 pages now, up from 15 across 11 pages earlier in the month. The volumes are tiny; the direction is not.
The lag note, because it always applies here: this export runs through Sep 25, and Search Console lags two to three days, so Sep 26 and 27 are simply unmeasured. Next week's export judges them.
A new lever opened this week: mining the near-miss queries the console already shows, then making the page say the exact phrase the engine is testing. The first harvest surfaced a surprise worth telling. The site ranks position 2 for "rank tracker software", and the ranking page is not the one anyone would have guessed. It is a free-tools facet page, pulled up by a single tool listing. The lesson is blunt: the engine decides which of your pages answers a query, and you find that out by looking, not by assuming.
What shipped
The site itself changed more this week than the numbers did.
On Sep 24, v3 went live: new copy, a free-audit intake, and the interactive 7-day live-traffic chart on the homepage, broken out by human and AI. That is the same day-by-day series these reviews quote, now public for anyone to check. The "watch it live" claim I kept making finally became a real piece of UI.
On Sep 27, the /writing/ section launched with the first two reviews of this series, and the site gained a content surface it did not have before. It was submitted to the engines the same night.
The registry grew from 398 to 424 entries, open-source GTM tools from 367 to 389, across five daily update runs. Roughly 176 registry URLs went out via IndexNow across the week, plus 44 more with enrichment wave 4. Separately, the full SEO and GEO audit process had its first live run on the site itself, including revalidating which AI crawler user-agents are current: one retired, two added.
Enrichment wave 4 shipped Sep 26, 13 pages. Its readout is structurally impossible this week, because it deployed one day after the measured window ends, so next export is the first that can judge it. Waves 2 and 3 read as holding.
Bing: still nothing, and the gap I have to name
Last review's open question was Bing, and the honest answer is no change. Bing still has not started spending crawl budget. The authority diagnosis, a young domain with 3 backlinks, stands unrefuted.
IndexNow submissions ran all week, so the accelerant is on. But the accelerant was never the constraint; authority is, and authority did not move. Here is the part I will not dress up: no new backlinks were acquired this week either. The off-site work has not produced a delta yet. Naming that plainly is the point of reporting in the open, because the alternative is quietly hoping nobody checks.
The comments fixed my curl advice
As this report was being written, the thesis of the whole series paid out in public. Over the weekend I posted a bit of pricing-citation advice: curl your pricing page and read what the bot actually sees. Within hours, practitioners improved it in the comments.
Two corrections, both now part of the checklist. First, hand-setting Accept-Encoding can make curl return an empty body, which produces a false "the bot sees nothing" reading; let curl handle it with --compressed. Second, a page can be perfectly fine while the CDN challenges the crawler at the edge, so the browser passes, the crawler gets a challenge page, and the owner sees nothing wrong because it only surfaces in the CDN logs.
I did not catch either. Two readers did, for free, in an afternoon. Publishing your diagnostics in public gets them peer-reviewed by people sharper than you on that specific edge, which is the entire argument for working this way.
What next
The wave-4 readout is the first thing next week's export can actually measure, so that is the near-term signal to watch. Backlinks remain the lever, and until they move, Bing stays exactly where it is; that is the honest through-line from last week to this one. And I will be watching one specific thing on the machine plane: whether the Sep 25 pattern repeats, or whether it stays a single OpenAI-driven spike sitting next to a quiet Sep 26. Either answer is fine. I just want to report the real one.