The Record Decays Unevenly

The public record is usually imagined as an accumulation. Things are published, they join the pile, and the pile grows. The measured behaviour is different: the record loses material continuously, at a rate large enough to change what can be established about an organisation, and it does not lose material evenly.

The Pew Research Center’s study “When Online Content Disappears”, published in May 2024, found that 38% of webpages that existed in 2013 were no longer accessible by October 2023, along with a quarter of all pages sampled from across the 2013–2023 period. The same study found that 54% of English Wikipedia pages contain at least one dead reference link, that 23% of news pages and 21% of government pages carry a broken link, and that nearly one in five tweets vanished within months of being posted.

The headline figure is arresting but the composition matters more, and it is the part that gets dropped when the study is summarised.

Loss is selective

Material does not disappear at random. It disappears according to who is paying for it and why.

Pages vanish when a company is acquired and its site is folded into the buyer’s. They vanish when a publication changes platform, when a marketing campaign ends, when a subdomain is retired, when a contract with a hosting provider lapses, when a founder loses interest. What survives is what somebody continues to have a reason to maintain.

That selection has a direction. Well-resourced organisations preserve their archives; defunct and small ones do not. Material that supports a current commercial position is migrated; material that documents a superseded one is not, and its removal is rarely deliberate enough to be called suppression. The surviving record is therefore not a smaller copy of the original record. It is the subset that remained useful to someone with the means to keep it, which is a different thing and a systematically flattering one.

Anyone reconstructing an organisation’s history from what is currently reachable is reading that subset, and the reading tends to be more coherent, more favourable and more continuous than the events were.

Citations outlive what they cited

The Wikipedia figure is the one with the most immediate consequences, because a dead reference does not behave like a missing reference. The claim in the article stays. The footnote stays. What has gone is the ability of any reader to check whether the second ever supported the first.

An assertion in that condition is functionally an assertion without a source, but it does not look like one. It looks better sourced than a bare claim, since it carries the apparatus of verification. Systems that weight material by the presence of citations, and readers who do the same, will treat it accordingly.

A related failure occurs while the link is still working. A widely repeated claim holds that around 60% of Wikipedia articles about companies contain factual errors, credited to academic research from 2012. The underlying study surveyed 1,284 public relations professionals about the accuracy of articles concerning their own organisations and clients. It measured the opinions of interested parties about their own coverage. It did not audit anything, and it does not support the claim made in its name.

Here the source is available, live, and freely readable. It simply says something other than what it is cited for. Link rot removes the ability to check; citation drift removes the incentive, because a live URL beside a claim signals that the checking has already been done. The second is harder to detect than the first and there is no automated test for it.

The governance reading

Treated as a technical matter, decay is a broken-links problem and there are crawlers for it. Treated as a records problem it is more awkward, because the material at risk is mostly not on the organisation’s own servers.

An organisation’s evidentiary position consists of registry entries, coverage, directory listings, documentation, references and analysis distributed across parties who have their own reasons for keeping or discarding things. None of those parties is under an obligation to preserve what it published, and most will not notice when they stop. The material disappears without a notification to anyone, least of all to the subject.

Three consequences are worth separating.

The first is that a claim once well supported may quietly become unsupported. The support was never owned; it was borrowed from third parties whose decisions about their own archives were made for unrelated reasons.

The second is that the balance of the record shifts as it thins. If four accounts of an organisation existed and two are gone, whichever two remain now constitute the record, and there is nothing in the surviving material to indicate what was lost. A description assembled from it will be confidently constructed from an unrepresentative sample.

The third is that the loss is invisible from the inside. An organisation’s own site can be complete and current while the external material that corroborated it has thinned to nothing. On any internal measure the position looks unchanged.

What is actually available to do

Very little of this can be prevented, and it is better to say so than to propose a procedure that does not work. Nobody can compel a third party to keep a page, and archives are partial.

What can be done is closer to bookkeeping than to strategy. An organisation can know which external sources currently carry an accurate account of it, and check periodically whether they still exist and still say what they said. It can keep its own copies of material it relies on, with dates. Where a claim depends on external corroboration, it can note which source corroborates it, so that the disappearance of the source is a visible event rather than a silent one. Where the supporting material has gone and nothing has replaced it, the honest position is that the claim is no longer supported, which is a normal thing for a record to say.

None of this is difficult. It is merely unrewarded, which is why the measured decay rates are as high as they are.