Two people can open the same address at the same minute and read different pages. One gets the national edition, the other local headlines; one the full site, the other a stripped-down page built for a phone. The Web Archive holds copies of that address, but no single copy stands for what either person saw, because each page was constructed separately for its visitor.
One address, many versions
The variation was documented precisely in a 2013 web-archiving paper. Researchers at Old Dominion University crawled the CNN front page twice with the same URL, changing only the user-agent string so the site saw a desktop browser in one crawl and an iPhone in the other. The site returned a different page each time, and both captures sat in the archive under the same address, distinguishable only by content. The difference was rarely subtle: one ESPN page held 201 links on desktop and 58 on mobile, and NBC News served different local stories and weather depending on where a request appeared to come from.
The same variation arrives through signals no user-agent can carry: geolocation, cookies, login state, A/B tests, and reading history. None of these make the URL lie; they mean a URL can name a family of representations rather than one fixed document. The New York Times experimented with personalized homepage modules based on signals including location and reading behavior, exactly the kind of publishing where one archived homepage cannot represent every visitor’s experience.
What one capture keeps
A crawler arrives with a particular browser profile, network location, session state, and crawl configuration. The Library of Congress notes in its web-archiving FAQ that archived sites may differ from the live web because interactive and dynamically generated material cannot always be captured completely. A personalized page adds another problem: even a technically perfect capture may represent only one visitor state. Everything the crawler did not look like goes unrecorded: the local section a New York IP would have seen, the recommendations built from a logged-in reading history, the variant served in an A/B test.
The archived copy is faithful to its own visit. Nothing preserves the relationship between that visit and the ones it missed, and nothing but the content distinguishes the two CNN pages under one URL. A history of the web that treats each address as a single evolving document quietly misreads a web where documents are a population.
Context needed to read a capture
Interpreting a preserved version later requires knowing what the page was built for. When was it captured, with which software, from which region, as which device, logged in or not, after which scripts ran? The Old Dominion study argues that those environmental dimensions belong beside the timestamp if future readers are going to understand why two captures of one URL differ.
Recovery has a hard limit. A version nobody captured is gone regardless of how much context survives, and the only full record of what a particular person saw was that person’s own browser. As pages are assembled per viewer, the archive captures one witness’s testimony to a page that existed in many forms. Preserving the personalized web means asking who saw it — because the thing nobody saw was never on the web at all.