Features
Part of The ethics and mechanics of archiving internet culture properly
Internet culture archives move constantly: writing a citation that survives
Internet culture archives move under your citation in five ways, only one of which looks broken, and this is how to write a reference that survives it.
A citation to an archived page implies that the page will still be there and still say the same thing. Neither is guaranteed. Archives are working systems, run by people making decisions, and the decisions change what a reader finds when they follow your reference.
This page covers the five ways an archived source moves under a citation, and how to write a reference that keeps working, or at least fails visibly instead of silently. The services you are most likely to use — the Internet Archive's Wayback Machine, which has captured the web since 2001; archive.today; and Perma.cc — are all subject to the same five, and none of them is a permanent record.
What to take away
- An archive can remove material, re-crawl over it, reorganize its addresses, change its access rules, or quietly change its rendering. Only one of those looks like a broken link.
- The dangerous failure is the one that returns a pagesilently different content under an address that still resolves.
- A citation is repairable if it records enough to find the item another way. Most citations record only an address.
- Quote the sentence you are relying on. A quotation is the only part of a citation that survives the source changing.
The five ways it moves
What the reader sees
- Removal on request
- A missing page, or a notice
- Re-crawl
- A newer capture at the same address
- Reorganization
- A redirect to something adjacent, or a search page
- Access change
- A login wall, a paywall, or a regional block
- Rendering change
- The same content, differently assembled, sometimes losing pieces
Detectable how
- Removal on request
- Immediately, because it fails loudly
- Re-crawl
- Only if you recorded the capture date and check it
- Reorganization
- Only by reading the page and comparing it to your note
- Access change
- Immediately for you, invisibly for readers elsewhere
- Rendering change
- Almost never, unless you kept a copy
The middle three are the problem. All of them return a working page, so nothing announces a failure, and the reader concludes that your citation supports whatever they are now looking at. A dead link is honest by comparison, which is the useful thing to notice about link rot: the visible kind is the mild kind.
Five Ways a Citation Breaks
Change
- Removal on request
- Missing page or notice
- Re-crawl
- Newer capture, same address
- Reorganization
- Redirect or search page
- Access change
- Login wall, paywall, block
- Rendering change
- Same content, reassembled
What You See
- Removal on request
- Immediately, fails loudly
- Re-crawl
- Only if you logged capture date
- Reorganization
- Only by reading and comparing
- Access change
- Immediately for you, not others
- Rendering change
- Almost never, unless you kept a copy
Detectable How
- Removal on request
- Re-crawl
- Reorganization
- Access change
- Rendering change
Why removal happens, and why it is legitimate
Archives remove things. They do it because a person asked, because a rights holder objected, because material was captured that should not have been, or because a policy changed. Each of those is a defensible reason, and a reader who treats removal as vandalism has misunderstood what an archive is.
It also means an archive is not a permanent record and never claimed to be. Its holdings are what it holds at that moment. Citing it cites a state of a system, so a citation that omits which state is incomplete.
Removal is not only a takedown request. When Yahoo closed GeoCities in 2009, one of the largest collections of early web pages went dark, and volunteers from Archive Team, working with the Internet Archive, copied what they could before the shutdown. What they missed did not survive.
The ethical side of what archives choose to keep and let go is set out under internet culture archives.
Writing a citation that survives
Record these, in this order of importance:
- The quotationthe exact words you are relying on, copied character for character.
- The creatorthe author, account, or community that posted the item.
- The title and publication dateof the item itself.
- The archive and the capture datewhich service holds the copy, and which snapshot you used, because the same page is captured many times.
- The persistent identifier, when the item has one, recorded alongside the address rather than instead of it.
- A description and the date you checkedone plain sentence that would let another person find the item even if every address in the citation fails.
Citation Elements in Order
- The quotation, exact words, kept short
- The date you checked it yourself
- The capture date, if archived copy
- The original address, separate from archive
- The archive and its kind
- A description in your own words
That is more than a typical reference carries, and every element earns its place by covering a different failure. The general argument for recording what a claim rests on, inside the claim, is made under platform timelines.
Addresses, and why they move
Addresses move because organizations reorganize, and reorganizing usually means a new address scheme applied to everything at once. The case against doing that was made clearly in Cool URIs don't change, written by Tim Berners-Lee for the W3C in 1998, and it is worth reading because it explains why the reasons people give for changing addresses are almost all bad ones.
Where an institution has thought about this, it issues a persistent identifier: a name for the item that survives the item moving, with a service that resolves it to wherever the item currently lives. The systems you are most likely to meet are the DOI, used by journals and publishers; the ARK, issued by libraries and archives; and the Handle, the older scheme that several of the others are built on. When one exists, cite it in addition to the address rather than instead of it, since the resolver is itself a service that can stop.
Most of the material this site concerns has no such identifier and never will. For that material the quotation is your persistent identifier, which is why it is first on the list above.
Maintaining references you already published
Set an interval and check. Not every reference, and not for correctness of the underlying claim, but for whether the address still returns the thing you cited.
What to Do When a Reference Changes
What does the address now return?
Gone -> say so in place, with the date
Changed -> keep old note, add new state
Never edit a citation to match a source that changed. That converts your record of what a source said into a record of what it says, and destroys the only evidence that it ever said anything else. The same reasoning applies to a term whose meaning drifted, which is why social media slang entries keep senses rather than replacing them.
Common questions
Should I archive a page myself before citing it?
Where you can, yes, and it does not remove the need to quote. The Wayback Machine's Save Page Now, archive.today, and Perma.cc — the last run by the Harvard Law School Library for legal and scholarly citation — all capture a page on demand and give you a durable address for the copy. A capture you made is a capture that can also disappear, and it carries no more authority than any other.
What if the archived copy and the live page disagree?
That is a finding, not a problem. Record both, with dates. The disagreement is often more informative than either version.
Is it acceptable to cite a page I could not load?
Only if you say so, in the citation, with the date you tried. A reference to something you never read is a fabrication with a footnote.
How often should references be checked?
Often enough that a reader is unlikely to be the first to find a break. A yearly sweep, on the anniversary of publication, is a workable default, and any interval you actually keep to beats a better one you do not. The general problem of how much of this material disappears is covered under early social networks.
