A New Site Doesn't Show Up in AI's Recommendations——A Fixed-Point Observation Starting on Day 2 After Launch

On July 13, 2026, on the second day after this site's launch, we asked ChatGPT for "recommended sites where you can research AI search and LLMO." The response listed 10 sites, including global SEO tool vendors, official documentation from search engines, and domestic explanatory media. This site was not included.

That a site on its second day after launch is not recommended is a natural result in itself. What was interesting was the breakdown. Among the recommended sites were some that broadly restrict automated access via robots.txt. On the other hand, a site right after launch that had put in place the machine-readable foundations——allowing AI crawlers, structured data, llms.txt——did not appear in the list.

What can be read from this is the distance between "being in a state readable by AI" and "being recommended by AI." Readability is no more than a necessary condition; what determines recommendation is considered to be a different factor.

This observation is consistent with findings from academic research. A study that compared AI search and conventional search at large scale (Koudas, 2025) reports that AI search systematically prioritizes earned media——mentions by third parties and authoritative sources——over a brand's own communications, and that a bias toward major brands exists. In other words, AI's recommendation lists are likely to reflect how much the web as a whole talks about a site, rather than what that site itself writes.

For a new site, this is a harsh reality. No matter how thoroughly the machine-readable foundations are put in place, without an accumulation of mentions, a site does not make it onto the recommendation stage. And accumulating mentions takes time.

So what should a new site wait for, and what should it measure? There are broadly two routes by which a site is reflected in AI search. One is the route where a page is cited when the AI performs a search; the other is the route where it is incorporated into the training data of a future model. The former allows the crawl status and the emergence of citations to be observed to some degree.

Therefore, on this site, we begin a fixed-point observation using this site itself as the subject. What we observe are three lines: (1) named questions that include the site name, (2) data questions that this site's database can directly answer (for example, "which facilities in Osaka offer 24-hour painless (epidural) delivery"), and (3) general recommendation questions like this one. We will submit them regularly across multiple AI search environments and record the timing and context in which citations emerge.

The record of "not being recommended" right after launch is the starting point of this observation. When change occurs, or when it does not, we will report the progress with numbers.

Note that this article does not assert the recommendation criteria of any particular AI service; it is an observational record at a single point in time.