October closed with a naming that arrived through the side door, and a criterion for whether it would hold: present in at least one Perplexity pass on the same question next run. It held. In the same run, the one citation the site itself had earned went away. This report keeps the two apart, because the instrument does.
For anyone new to the series: every month the same 22 questions go to three AI providers, twice each, through their APIs with web search available. The method is in the July report. August is here, September here, October here. API behavior differs from what you see in the consumer apps, so read every finding with that in mind.
Run 5 fired on November 3. Panel v2, 22 questions, 132 responses attempted, 132 collected. The account had credit this time.
Named, second month
On the question "Custom AI tool development Mérida Mexico", Perplexity named Yellow House Digital again. Both passes. Alongside a handful of other firms, as before.
The criterion was one pass. It met two. By the ledger's own rule, the October finding is confirmed rather than lapsed, and it is now the first naming in five runs to survive a second measurement. July's did not.
The size of it, stated plainly: one question out of 22, one provider out of three. Every other commercial name on the watch list was at zero, same as every month since August.
The source behind the naming is the same as in October. The company's LinkedIn page. That was the only Yellow House Digital source cited anywhere in the run.
Cited, gone
The page built in July for the readability question, "Who should I hire to check if my website is readable by AI?", was cited by Perplexity in September and again in October, both passes each month. In November it was not cited in either pass.
Nothing in the instrument can say why. Between runs 4 and 5 the site and the LinkedIn page both changed, in the same interval, so the drop cannot be attributed to any one thing and neither can the naming. I said that about October's result and it is still true about November's. The freeze holds through one more run.
One absence is not a loss. The rule that confirmed the naming is the same rule that governs this: if the citation is absent again next month, it lapses. Until then it is a single missing data point on a page that had two months of presence behind it.
Two signals, two places
Here is the shape of it after five months.
The site is named, through a third-party surface, on the least contested question on the panel. The site is not cited, on the question a page was built for. Last month the owned page carried rung two and the LinkedIn page carried rung three. This month the owned page carries nothing and the LinkedIn page still carries rung three.
Named and cited are different signals. The instrument records them in different columns, and this run is the clearest case yet for why. If I had one number for "visibility", it would have gone up this month. The two numbers I have went in opposite directions.
I am not drawing a lesson from that. One month of divergence is one month.
The stable exception, fifth month
The company that survived both July passes on the highest-churn question appeared again. Five consecutive runs. One OpenAI pass, cited through a white-label services page, the same page as October. Still the longest stable presence the instrument has measured on the question where names otherwise come and go fastest.
The machinery, fifth month
OpenAI skipped search on both passes of eight questions this month. The original six have now held for five runs, unchanged.
Search rates moved a little. OpenAI went from 63% to 66%, measured on the same 38 answers both months, because October's six failed cells are not comparable. Perplexity searched on every answer, both months, as it has every month. Small, not a story.
The churn continued. All 22 questions moved month over month. On the flagship question, who builds custom AI tools for agencies, the answers gained 36 names and dropped 35. July's argument stands.
What this does not mean
It does not mean LinkedIn is the lever. It does not mean the readability page stopped working. It does not mean anything about the site changes in October, because the same interval holds the changes and the results and the instrument cannot separate them.
It means one thing. Two signals that looked like they were climbing the same ladder in October turned out, in November, to be on different ladders. Whether that holds is run 6's question.
What happens next
Panel frozen at v2, fifth month. Run 6 fires December 1.
The naming has a second data point now. The citation has an open one. Both publish either way.