Simon Willison discovered that Firefox automatically prefetches “next page” links for Datasette database tables, even though only a small share of users actually click through. This roughly 24-year-old browser behavior has now been addressed by eliminating unnecessary server-side computation.
What this is
Datasette is an open-source tool for publishing and querying SQLite databases through the web. Pages declare the “next page” URL in the HTTP Link response header. When Firefox sees this, it fetches the page in advance before the user clicks. Prefetching—a browser loading content the user may access in advance—is designed to reduce waiting time. But on database-query websites, it can force servers to render pages that nobody will view.
Industry view
We note that this change does not reject prefetching as a practice. Datasette lead Simon Willison still considers it reasonable for websites to provide next-page information through links. The problem is that browsers treat a navigation cue meant for human judgment as a high-probability access instruction. After the fix, browsers wait until users actually click before requesting the next page, sacrificing some of the convenience of having content prepared before it is opened.
Such problems are not uncommon. Automated systems often take additional actions based on probabilistic judgments about user intent. When server and user objectives differ, prefetching can turn efficiency into cost. Another risk is that Datasette’s fix primarily targets Firefox; resource-consumption problems may remain if other browsers adopt similar strategies.
Impact on regular people
For enterprise IT: Operations teams managing databases and content platforms need to check whether browsers are prematurely requesting resources such as “next page” links, images, and scripts. In particular, they should monitor whether these requests drive wasted computation and higher bandwidth consumption.
For professionals: When using a browser to query public information, prefetching may make pages feel more responsive, but it may also make servers slow down faster. The key is not to reject automation, but to distinguish between “what users explicitly need” and “what systems infer users might need.”
For the consumer market: Website performance may vary slightly, but the main beneficiaries of the fix are website operators. Most users will feel no difference, while avoiding the unintentional use of computational resources for large volumes of content that nobody clicks.