The Internet Is About to Lose Its Memory

Show someone a screenshot of MySpace, the first iPhone home screen, or Facebook in 2008, and they can often place it in time. These interfaces function like old shop signs or newspaper front pages. Millions of people saw the same arrangement of buttons, colours, and menus, so the design became part of a shared past.

That common visual record may soon become harder to produce. Researchers in human-computer interaction are developing generative interfaces: screens assembled by artificial intelligence for a particular person and task. Instead of opening the same fixed page as everyone else, a user may receive a temporary dashboard, comparison tool, or menu based on a request, previous behaviour, and inferred preferences.

In a 2025 preprint, participants preferred task-specific interfaces generated by a language model to ordinary chat responses in more than 70 per cent of the tested cases. A 2026 study went further by learning individual design preferences and using them to select different interfaces for different users.

For users, the appeal is clear. A complex service can remove controls that a person rarely needs and foreground the information that matters at that moment. Yet a personalised screen creates an awkward historical question: which version should be preserved?

The web already loses large amounts of material. Pew Research Center examined pages that appeared in the Common Crawl repository between 2013 and 2023. By October 2023, one quarter were no longer accessible. Among pages recorded in 2013, 38 per cent had disappeared. Usually the entire website had not vanished. A single article, file, or page had been removed.

Web archives recover part of that lost record by storing copies at particular times. Their task becomes more difficult when content changes according to the visitor. Archive-It, a service of the Internet Archive, describes behaviour-based and preference-based content as a known archiving problem. Pages that require interaction or depend heavily on JavaScript can be captured incompletely or fail during replay.

Generative interfaces deepen this problem because the page may never exist as a stable object. It is calculated during the visit. Two people can ask the same service the same question and receive different layouts. The output may also change after a model update. Saving the address and source files would resemble preserving a theatre building while discarding the performances.

This matters because memory is partly social. Research published in Nature Human Behaviour found that the way French participants organised memories of the Second World War reflected the structure of their culture's shared public narrative. Media provide common cues through which private memories become connected. A familiar website layout can serve a smaller version of that function. It gives a generation a visual reference that others recognise.

If interfaces become fully individual, people may remember using the same platform without remembering the same place. Future historians could know what a service was designed to do while remaining unable to reconstruct what any particular user saw. Recreating the experience would require screenshots, the model version, the prompts or actions that shaped the page, relevant user settings, and the generated output itself.

The internet's next memory problem may therefore begin before anything is deleted. A page that appears once, for one person, and is never recorded has already entered history in a form that history cannot easily retrieve.