AI Bot Visibility Across Five Matched Websites

The first AISymantix study found a narrow page-level discovery signal but no reliable sitewide prediction. The obvious next question was whether AISymantix was simply too small a sample. So I picked four other websites with similar launch timing, current page counts, GSC coverage, and validated crawler data. Then the comparison did something useful. It made the crawler story harder to defend.

Five Sites, One Common Window

The cohort was selected before calculating the result. First GSC observations fall within twenty-one days of one another. Current sitemaps remain in the same general size band. Every site supplies a complete GSC series from July 9 through September 9 and the same sixty-one validated Good Bot days after the two unavailable cache dates are left as gaps.

The Obvious Explanation Is Work

AISymantix received most of my attention during this period. That is not a subtle difference hiding in the data. Sixteen of the cohort's nineteen logged SEO Events belong to AISymantix. It also generated nearly two-thirds of the measured impressions. If I worked on one site constantly and barely touched the others, of course the worked site should look different.

Then AITOAIConnect refused to make the explanation that tidy. It had two logged events and finished with a larger impression increase than AISymantix in the fourteen-day comparison. That does not mean two events beat sixteen. Events vary wildly in scope, event logging is not work-hour tracking, and two sites are not a controlled test. It means event count cannot be used as a dosage meter.

The fair reading is narrower. Sustained work probably contributed to AISymantix owning 63 percent of cohort visibility. The current data cannot separate that contribution from topic demand, content quality, internal linking, query competition, or the unusually strong search-visibility cluster AISymantix developed during the same period.

Bots Rose Everywhere. Search Did Not.

This is the contradiction I was looking for. Good Bot activity increased on all five sites. Search impressions fell 50 percent on SymatixAI and slipped 2.5 percent on CustomCakesOrlando. SymatixAI recorded more Good Bots than AISymantix, yet AISymantix produced almost eight times as many impressions. CustomCakesOrlando also recorded more bots than AISymantix and produced about one-seventh of the impressions.

If crawler volume were a practical leading indicator, those are difficult results to explain. If crawler volume measures machine access while search visibility depends on a larger system, they make perfect sense.

The Longer Daily Test Still Finds No Forecast

The cohort test normalizes daily bots and impressions by current sitemap size, removes each site's average, and compares the remaining within-site movement. That keeps AISymantix's larger impression total from swallowing the other four sites. The test then asks whether today's bot activity aligns with the change in impressions up to seven days later.

The strongest bot-leading growth relationship is 0.119 at six days. That is not nothing. It is also nowhere near strong enough to treat as a forecast, especially after testing seven possible leading lags and keeping the best one. The longer, broader test reaches the same conclusion as the original AISymantix analysis.

Page-Level Discovery Still Survives

The sitewide theory failed twice. The page-level sequence did not. Across the retained raw-request window, twenty-seven current pages received a Good Bot request before their first GSC impression. That is real discovery order. It is surrounded by eighty-one bot-only pages and six same-day, search-first, or search-only counterexamples. That is why discovery order cannot be promoted into ranking prediction.

What the Five-Site Study Supports

• Websites with similar age and page counts can produce radically different search visibility while receiving much closer Good Bot totals.
• Good Bot growth is common across the cohort and does not distinguish sites with rising impressions from sites with falling impressions.
• AISymantix's work intensity is a credible part of its result, but event count is not a clean measurement of effort and does not isolate causation.
• AITOAIConnect is an important counterexample because its impression growth exceeded AISymantix in the period comparison with far fewer logged events.
• Crawler requests remain useful evidence of machine access and page discovery order.
• Neither crawler totals nor the current pooled lag test support using bot visibility as a ranking forecast.

Method and Limits

• The sites were selected using current sitemap size, first GSC observation date, and overlapping validated Good Bot coverage before the cohort result was calculated.
• Study totals use only dates available for every site in both GSC device totals and the algorithm-version-2 Good Bot cache.
• Good Bots count successful content GET requests from active search, AI search, training, and user-retrieval crawler fingerprints.
• The pooled lag test normalizes by current sitemap size and removes each site's mean before calculating correlation.
• SEO Events are evidence that recorded work occurred. Their absence does not prove that nothing changed, and one event is not equal to another event.
• The five sites have similar age and size, but their subjects, competition, content, internal linking, and search demand remain different.
• Raw request retention is thirty-five days, so page timing and human-request comparisons use a shorter window than the sixty-one-day site series.
• Crawler activity proves machine access. It does not prove indexing, citation, training use, a human prompt, ranking causation, or future visibility.

Public living dataset /data/research/ai-bot-visibility-cohort.json. It includes the cohort definition, daily site rows, crawler-purpose totals, lag tests, event windows, page timing, bot-first receipts, bot-only counterexamples, method, and limits.