A UK employee wellbeing platform's site carries roughly 92 live articles. Sitting behind them, in the CMS, are 121 more - a whole earlier generation of blog posts, unpublished during a site rebuild and never brought back. Their old URLs return 404. To every crawler, engine and dataset on the internet, those pages simply stopped existing.
Buried in that dormant pile are the slugs that made me stop scrolling: stress statistics. Mental health statistics. Financial wellbeing statistics. A UK employee wellbeing statistics roundup, whose old address I fetched directly during the audit and confirmed dead.
Of all the content this company has ever produced, those are precisely the pages AI engines are most likely to cite - and they are precisely the ones the site no longer serves. This case study is about why statistics pages punch so far above their weight in AI answers, how content culls keep destroying exactly this asset class, and what a rational rescue looks like.
§ Why statistics pages are citation magnets
Watch how an AI engine actually uses a source and the pattern is hard to unsee. Engines cite pages that let them extract a specific, attributable fact: a number, a percentage, a dated figure with a source. A statistics roundup is that, industrialized - dozens of extractable claims, each one a potential citation, each one answering the factual questions ("what percentage of UK employees report workplace stress?") that people put to machines constantly.
Compare that with the thought-leadership essay, the product update, the trend piece. An engine can paraphrase those; it rarely needs to cite them, because they contain few facts that require attribution. In the same engagement's own sweep data, the pattern showed up from the other side: on the informational how-to query I tested across three runs, the engines cited nobody at all - they answered from model memory, because no source on the open web had made itself the extractable authority worth pointing at. Informational queries do not go citation-less because engines refuse to cite. They go citation-less until someone publishes the page that earns it. Statistics pages are the shortest path to being that page.
§ How the cull happens - and why nobody flags it
No one decided to delete the citation surface. What happened is what always happens: a site rebuild, a migration to a new template, a reasonable instinct to launch clean with current content. The older collection - dated posts, inconsistent formatting, uncertain traffic - stayed behind in the CMS, unpublished, pending a review that never got scheduled. The URLs went to 404, and no redirect map was built because the pages "weren't coming back anyway."
Every criterion in that decision was visible in analytics: traffic, freshness, brand fit. The criterion that was not on the spreadsheet is the one this whole series is about: citability. A three-year-old statistics roundup can look like the deadest wood in the archive by pageviews and still be the single most machine-quotable asset the company owns - because its value was never mostly the visitors it drew, but the facts it staked out. Culls run on visitor metrics amputate citation assets by accident, and the loss registers on no dashboard, ever.
There is a compounding cost, too: even the 404s themselves are dead weight. Every crawler that still remembers those URLs - and crawlers remember for years - spends its visit on a dead end instead of a redirect to something alive. The site is paying crawl attention on content it is not even serving.
§ The rescue, sequenced honestly
Not all 121 posts deserve resurrection, and pretending otherwise would repeat the original mistake in reverse. The sequence I put on the queue:
First, triage by citability, not nostalgia: the statistics pages and anything else dense with extractable, verifiable facts goes in the rescue tier; the rest gets judged normally. Second, refresh before republish - a statistics page earns citations with current, sourced, dated numbers; republishing 2023 figures unedited would manufacture the stale-fact problem engines punish. Third, republish at the old URLs where possible, or 301 the old addresses to the new ones, so the years of memory pointing at those slugs resolves to content again instead of a 404. Fourth, only after that, extend: the statistics surface becomes a maintained annual asset with a named owner, because a statistics page with a maintenance owner compounds, and one without decays back into cull bait.
§ The audit for your own archive
- List every URL your CMS knows about that is not currently published, and check what the old addresses return. A pile of 404s with content sitting unpublished behind them is recoverable value at close to zero production cost.
- Score the pile on one question: does this page contain facts a machine would need to attribute? Numbers, percentages, dated claims, original data. That score, not last year's pageviews, decides the rescue tier.
- Before any future cull, add citability to the kill criteria alongside traffic, and require a redirect decision per URL - even if the decision is "let it die," make it a decision.
- If you own no statistics page in your category, notice that the slot may simply be open: test the factual queries in your space and see whether engines cite anyone at all. An uncited query is an invitation.
- Give every statistics asset an annual refresh owner at birth. The format's citation power is entirely a function of being current, and currency is a process, not a property.
Findings dated as of the May and August 2026 audits: CMS collection counts from the site's own platform, dead URLs verified by direct fetch, citation behavior from 3-run sweep data in the same engagement. The site is anonymized as a matter of client confidentiality.
The summary with the sting in it: this company spent years producing the exact content format that AI engines most want to cite, then unpublished it in an afternoon of tidying, and the whole loss was invisible because deleted pages do not appear in any report. Your archive may be holding your best AI-visibility assets right now. Check what you have buried before you commission what you already own.