October 27 marked World Day for Audiovisual Heritage. The UNESCO observance highlights a fragility that extends to the present: publishing something online does not guarantee its survival. Deleted pages, shuttered websites, and obsolete formats contribute to digital decay: the gradual loss, corruption, or inaccessibility of digital information. Understanding why this material disappears means asking how to preserve it and who decides whether we will still be able to access it. Today, even efforts to prevent artificial intelligence companies from using content can hinder those who collect copies to preserve it.

Established in 2005, the observance commemorates UNESCO’s 1980 Recommendation for the Safeguarding and Preservation of Moving Images. The 2026 theme, “Your Window to the World,” highlights the role of moving images in connecting people, cultures, and generations. But digitizing content and publishing it online is not enough: keeping it accessible requires care over time.

The idea of the internet as an archive with no expiration date runs up against stories like that of journalist Kevin Vaughan. In 2007, the Rocky Mountain News published “The Crossing,” his series on the aftermath of a 1961 accident that killed 20 children in Colorado. Text, photographs, and video told the stories of survivors and families. After the newspaper closed in 2009, access to the project was lost. The series was later brought back online with help from the Denver Public Library.

Digital decay happens this way, too: a newspaper closes, leaving published work inaccessible, along with the personal accounts it gathered. This is one of the forms of loss explored in Vanishing Culture: A Report on Our Fragile Cultural Record, released by the Internet Archive in April 2026. The volume brings together essays and case studies on articles, films, music, and websites that become unavailable as platforms, licenses, or terms of access change.

Fifty-seven years have passed since the launch of ARPANET, the U.S. research network established in 1969 that was a precursor to the internet; 37 have passed since the web was conceived in 1989. The internet is old enough to have a history, though that does not give it the wisdom to preserve it. The gaps in its memory stem from maintenance and decisions, not from a network that has somehow become “full.” There is no deus ex machina tasked with saving every fragment: behind its apparent unity are authors, publishers, platforms, search engines, and archives, each with different powers and interests.

A document’s survival, after all, requires practical care. A paper document needs space, protection, and someone who knows where to look for it; a digital one also needs functioning services, software that can read it, and devices that can open it. Even an intact file can become unreadable if the programs or storage media needed to access it fall out of use.

A bookmarked link can feel like a small assurance: this content interests us; we can return to it. But a URL only tells us where to look. It does not keep a copy. All it takes is for a website to shut down or a page to move without the links being updated, and we find ourselves staring at an error message: that is link rot. Other times, the page opens, but its content has changed. The link works; what we were looking for is gone. This is known as content drift.

A broken link, however, does not prove that all is lost. In Gone but Not Forgotten: Recovering the Dead Web, an analysis published in April 2026, the Internet Archive revisited data from a 2024 Pew study, checking 5.4 million URLs against the Wayback Machine. About 26% were unreachable on the live web. Archived copies were available for URLs accounting for 16 percentage points of that total, leaving roughly 10% unavailable in either place. The existence of a copy, the authors note, does not guarantee that every element of a page has been preserved.

To retrieve a vanished page, an archive must have copied it before it disappeared. This is where concerns about artificial intelligence come in: some publishers, worried that their articles will be used to train AI models, also restrict the Internet Archive’s automated collection of content. 

In May 2026, Nieman Lab identified 342 local news websites with restrictions of this kind. Among the reasons cited was concern that AI companies might obtain content from archives, although no publisher had confirmed to Nieman Lab that this had happened through the Wayback Machine. We do not know how many pages were left without a copy. But the risk is clear: efforts to control how articles are used today may make them harder to find tomorrow.

The desire to control what remains online extends to our private lives. We do not want to carry every trace with us: photographs and comments can continue to circulate after the original is deleted, and removing a search result does not erase the content from the website hosting it.

For what we do want to keep, meanwhile, the phone has become a personal archive that grows every day. Newly taken photographs sit alongside others retrieved from drawers and digitized, together with messages, videos, and documents. We upload them to services such as Apple’s iCloud, Google Drive, or Microsoft’s OneDrive so they remain within reach even when we switch devices. This is the cloud: files stored on remote servers and accessed over the internet. Behind the lightness of the word “cloud,” however, are infrastructure, contracts, and login credentials. Available storage and whether an account remains active can affect syncing, backups, and how long files are kept, depending on each service’s terms.

The digital world has become a second fabric of memory. In a study published in Memory in February 2026, first published online in December 2025, 40 participants in Germany reconstructed their birthday celebrations from 2019 and 2024. When allowed to consult materials and records to help them remember, they chose digital ones almost exclusively. They relied on them especially to recall the earlier celebration, in 2019, recovering details that memory alone had not supplied.

Preserving these traces takes more than an upload. The Library of Congress offers concrete guidance: keep high-quality originals, describe files, create multiple copies in different locations, and periodically check that they can still be read. As storage media and formats age, the content needs to be transferred. Personal memory needs maintenance, too.

Digital decay undermines the idea that publishing means preserving. Clicking “save” does not guarantee that we will find a document ten years from now; clicking “delete” does not ensure that every copy will disappear. Digital memory depends on the maintenance work and resources of those who preserve it, and on the choices made by publishers, platforms, and archives. Perhaps trust in the internet begins with recognizing that its memory, like our own, needs regular exercise.