Sources & citations: 20 ways AI gets it wrong
Fabricated citations, broken links and sources that don't support the claim. Each failure mode below is phrased as the question people actually ask, with what it looks like in real work — and the layer of the Trust Stack that catches it.
Does AI cite sources that don't back up its claims?
A footnote points to a document that says nothing about the point it supposedly proves.
Why does AI cite sources that aren't relevant?
A reference is attached that has no real bearing on the argument.
Can AI rely on sources that are out of date?
An argument leans on a report that has since been replaced.
Does AI treat low-quality sources as reliable?
A weak forum post is cited as if it were authoritative evidence.
Why does AI cite commentary instead of the actual law?
A summary article is cited where the underlying statute should be.
Can AI misquote a source it's citing?
A quotation appears in the text that the original source never contains.
Does AI paraphrase sources inaccurately?
A restated finding shifts the source's meaning enough to mislead.
Why does AI fabricate citations that don't exist?
A reference list includes a paper, author, and date that cannot be found anywhere.
Does AI give broken links as references?
Every supporting link in the answer leads to a dead or wrong page.
Can AI guess a source's content from its title alone?
A document is summarized confidently from its headline without the text being read.
Why does AI cite marketing copy as neutral fact?
A vendor's own brochure is used as impartial proof of its claims.
Does AI ignore sources that contradict it?
A confident answer omits the well-known sources that say the opposite.
Can AI fail to check which source is more recent?
An older and newer source are cited side by side with no note of which supersedes.
Does AI mix up sources from different jurisdictions?
Authorities from several legal systems are blended without flagging which applies.
Why does AI cite rules from the wrong country?
Guidance for one country is built on the laws of another.
Can AI rely on sources that have been replaced?
A repealed regulation is cited as if it were still in force.
Does AI blur official sources and opinion?
A government notice and a pundit's take are presented as equally official.
Can AI cite a source for the wrong point?
A reference that supports point A is attached to unrelated point B.
Does AI produce a long reference list that's actually weak?
A bibliography looks thorough but few of its entries actually support the work.
More from the library
One model can’t reliably catch its own mistakes. A council of independent minds can.
Run your work through the councilAll 250 failure modes · See also: the Trust Stack