02 · AI · 5 min read
Why AI Needs Records, Not Just Websites
A model can read a page. It cannot easily trust one.
A website is an excellent way to persuade a human and a poor way to inform a machine. It mixes fact with narrative, omits what it assumes, and rarely says when anything was last true.
Three failures of page-derived knowledge
- —Ambiguity: which of four similarly named businesses is this page about?
- —Silence: hours, service areas and credentials are implied rather than stated.
- —Decay: nothing on the page distinguishes a fact from 2019 from a fact from this morning.
A record addresses each of these directly. Identity resolves ambiguity. Attributes replace inference. Timestamps and provenance replace guesswork about freshness.
Being online may no longer be enough. Information may need to be understandable, verifiable and citable.
We are not arguing that models cannot read the web. They plainly can. We are arguing that the cost of reading it — the ambiguity, the staleness, the unattributable claims — is a cost someone eventually pays, and that structured records are one plausible way to lower it.