A citation is the moment an answer engine points at a source: a small numbered link sitting beside the reply, or a name woven into the sentence itself. It borrows the look of scholarly habit and works differently. A scholar cites the work that shaped the argument. An answer engine cites a passage it retrieved while composing, which may or may not be where the claim in that sentence actually came from — a model can reach for a plausible source after the fact as readily as it reaches for a plausible fact.
That gap produces a failure worth learning to recognise. A citation can be entirely real, load correctly, and still not support the sentence it is attached to. It can also be fabricated outright, with the precise shape of a genuine reference — plausible journal, plausible year, plausible author — which is hallucination applied to attribution rather than to fact. The presence of a link reads as diligence, and most readers treat it as the end of the checking rather than the beginning of it.
For anyone who publishes, the citation has become the unit that matters, because the click it replaces may never arrive. Ranking highly used to mean being read. Being cited means being named to somebody who will most likely never visit. That is worth something — it is the machine saying this is where I would look — but it is a thinner thing than a reader, and it cannot be bought, demanded or reliably engineered. The engines lean toward sources that are specific, consistent and corroborated elsewhere, which is less a technique than a description of being good.
The human translation is being quoted rather than visited. Somebody repeats your sentence at a dinner you were not invited to, with your name attached, to people who will never meet you. It is real recognition and it is poorer than a reader. There is no lever to pull here, which is oddly restful: the only strategy available is to be worth quoting.
