A 404 is supposed to mean nothing is there.
Wrong URL, broken link, mistyped path — the server says no, the log captures it, and you move on. Philip K. Dick asked whether androids dream. It turns out AI does — of pages that don’t exist. In standard web analytics, 404s are noise. You filter them out, fix the obvious ones, and forget the rest.
But when an AI crawler hits a 404, something different is happening.
AI can’t say “I don’t know.” It fills gaps. It constructs answers. And sometimes, in constructing an answer, it reaches for a page that doesn’t exist — a page it has decided should exist. The 404 log, in that case, isn’t an error report. It’s a record of intent.
We pulled the 404s from mare-interno.com’s EdgeShaping logs and read them as exactly that: evidence of what AI thinks this site is, and what it expected to find here.
The AI Already Knows What This Site Should Look Like
Some of the 404s were almost flattering.
/blog/case-study//blog/category/looker-studio//blog/author/k-uchiumi/
None of these exist. But all of them make sense — for a site like this one. An analytics consultancy focused on GA4 and measurement infrastructure would reasonably have case studies, a Looker Studio category, and an author page. The AI correctly identified what kind of site mare-interno.com is, then extrapolated what its URL structure ought to contain.
This isn’t hallucination in the pejorative sense. It’s structural inference — the AI reading the shape of a site and filling in what logic suggests should be there. The irony is that the 404s generated by this process are, in a way, a sign of comprehension. The AI understood the site well enough to predict it.
Someone Asked a Question. The AI Went Looking for the Answer Here.
One path stood out from the rest:
/blog/ga4でデータを理解する：レポートの見方と解析方/
That’s Japanese, left raw in the URL. Decoded, it translates roughly to: “Understanding data with GA4: how to read and analyze reports.”
This wasn’t a crawler probing site structure. This was an AI mid-conversation, trying to answer someone’s question, and reaching for a source. It decided mare-interno.com would have an article with that title — and went to retrieve it.
The article doesn’t exist. But someone’s question does. The AI generated a citation before the content existed to support it. The query came first; the page was assumed to follow.
That’s a different kind of 404. It’s not a broken link. It’s an unfulfilled expectation — one that originated outside this site entirely.
Some of the Dream Belongs to Someone Else’s Site
Not all the 404s fit the pattern above. Some were stranger.
/blog/item/air-max-1-running-shoe-31-12-2025-07-40-22//blog/item/red-black-stripped-sneakers-31-12-2025-07-39-38//blog/メニュー//blog/事業内容/
Sneaker product pages. A navigation menu path. A “business overview” slug. None of these have anything to do with analytics or measurement infrastructure.
What appears to have happened: somewhere in the AI’s internal model, mare-interno.com got blended with other sites. An e-commerce store. A generic corporate page. The boundaries between sites aren’t as clean in an AI’s representation as they are in reality. The result is 404s that aren’t predictions about this site — they’re projections from somewhere else, misdirected here.
Less a dream, more a case of misdelivered mail.
The 404 Log as a Mirror
Taken together, these three patterns reveal something that standard analytics can’t show you.
Normal traffic analysis tells you who came, what they read, how long they stayed. The 404 log — specifically the AI-generated portion of it — tells you something else: what the AI thought was here. Its model of your site. The gaps it tried to fill. The questions it was already answering on your behalf.
Some of that model is accurate. Some of it is logical extrapolation. Some of it is noise from other sites bleeding through.
Reading the 404 log this way inverts the usual direction of analysis. Instead of measuring what users did, you’re inferring what AI expects — which may be closer to what users will ask than anything currently on the page.
This site has articles that don’t exist yet. AI knows that. Or more precisely — AI is dreaming them.