LANDMARK · 8 MIN · JUDGE
Trace a claim
After this landmark, you can take one concrete claim from an AI answer and follow it to the original source that either confirms it or doesn’t exist.
Tracing is the concrete act behind “verify it.” You don’t re-check a whole answer at once — you pick the pieces that can be true or false and follow each to where it came from: a named person, a figure, a direct quotation, a date, a specific study. A claim you can’t trace to a real, openable source should be treated as unconfirmed, not as probably-fine. Two failure modes hide here. The model can invent a plausible source that does not exist, and it can cite a real source that does not actually say what it claims. So tracing has two steps: reach the source, then check that the source supports the specific claim — not just that it’s on the same topic. The habit is small and repeatable, which is what lets you use it under time pressure.
This is the canonical concept. It stays the same across learner lenses so personalization never changes the underlying facts.
◌ EVERYDAY LENS
What this looks like for you
When an answer names a study, a statistic, or a quote you might repeat, search for that exact thing — not the topic. If you can’t find the original in a minute or two, treat it as “unconfirmed” rather than passing it on as fact.
MAKE A DECISION
An AI answer states: “A 2021 WHO report found remote workers were 34% more productive.” You need to use this. What do you do first?
CARRY THISPick one claim from a recent AI answer and trace it to a source you can open — or note that you couldn’t.