Does That AI Citation Actually Exist?

Watch the briefing

Watch the full video on YouTube

Briefing

M. Rowan, 2025. Reliable Reasoning in Everyday AI. Journal of Applied Machine Judgment. That looks like a citation. Does its appearance establish that the work exists? No. This reference is invented for our demonstration. Before relying on a convincing citation, find the original record and match its details. Formatting cannot do that work for us.

Start by separating the reference into fields: the title, the author or organization, the publication date, and an identifier if one is provided. An identifier might be a report number or a digital object identifier. These details give us something specific to check. Copying the whole citation into a search engine can provide a lead, but the lead is not yet the original record.

Now compare a real example. The official NIST document is titled Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile. The authoring organization is the National Institute of Standards and Technology. Its date is July 2024, and its report identifier is NIST AI 600-1. We opened the official PDF for this briefing. Those are record details we can match, not details we are accepting because an answer sounded confident.

Imagine someone supplies the same title and organization, but lists 2025 instead of July 2024. The document exists, yet that supplied reference does not fully match this record. Correcting the year matters because we want the right edition, not merely something with a similar name. A revised publication might have different content. In this comparison, the mismatch is the date. We are not claiming that a later edition could never exist.

Use three clear outcomes. Found means you located a record and matched the relevant details. Mismatched means you located something, but a field disagrees. Not located means your checking has not located the reference. That last result is a reason to keep checking or ask for a working source, not proof that no such work exists. Our invented reference illustrates the unresolved status; we have not performed or reproduced a live search for it.

NIST discusses confabulation, including citations that can mislead people into trusting an answer. Our checklist is an editorial method for reducing that risk, not a NIST-endorsed verification protocol. It also stops short of an important second question: does the document support the claim? A real source can be misread, quoted out of context, or attached to a conclusion it never makes. Finding it is progress, not the finish.

For the next citation you receive, save the original link, record the fields you matched, and name anything still unresolved. Do not turn uncertainty into a verdict just to make the notes look complete. Find the original record. Match its details. Then read what it actually supports. A later briefing can examine that final step; this check already helps us avoid treating a convincing reference as verified evidence.