Source and Citation Verification

How do I verify a citation properly?

Five questions in order, stopping at the first failure. Does the item exist, located as an item rather than as a search result or a mention elsewhere. Is it what the citation says it is, particularly its type. Does the relevant passage, not the abstract, say what the citing text claims. Does the quoted figure match on value, base, unit, period and definition. And is this source the origin of the claim or is it citing someone else. Existence is the cheapest rung and the least informative.

How can I tell if an AI made up a citation?

By trying to find it, which is the only reliable method. The signature is a real author who works in the field, a real journal that publishes in the field, a believable year, and a title that reads exactly like a paper that would exist. Frequently there is a malformed DOI, a page range outside the volume, or a volume and year that do not correspond. Sometimes two real papers are merged into one reference, and sometimes a real paper carries an invented finding. If the item cannot be located, treat it as fabricated.

What do I do with a citation I cannot find?

Remove it. It does not become "personal communication", it is not softened to "research suggests", and it is not kept because the underlying claim is probably true. Find a real source for the claim, or restate the claim as the author's own assertion, or delete it along with anything that rested on it. Softening the language while keeping the unfindable source is the same fabrication with the evidence hidden. And when you find one fabricated citation, verify the whole document, including sources that already passed.

How do I trace a statistic back to its original source?

Look at what your source cites for the figure, go to that, and repeat until you reach a source reporting its own measurement or a dead end. Record the value at each hop. You are looking for three things: a chain where many sources collapse to one origin, a chain that terminates at a figure with no attribution at all, and an origin whose actual finding is narrower or older than the claim. Also watch the number itself, because bases widen, scopes creep, periods detach and projections become forecasts as figures are repeated.

Does it count as corroboration if several sources say the same thing?

Only if they are independent. Collapse each source to its origin and count the origins, not the sources. Three independent measurements agreeing is corroboration. Three documents repeating one measurement is one measurement, and calling it widely reported is true and irrelevant. Frequency is routinely mistaken for strength, and the most repeated statistics in most fields are the least verified.

How should I handle a paywalled source I cannot read?

Record it as partially verified if you have confirmed its identity and type, or unverified if you have not, with the reason. Then either obtain access, find an accessible source for the claim, or label the claim as resting on an unverified source in the document itself. Reputation is not verification and a source is not checked because it is probably right. An unverifiable source labelled unverified is a respectable component of a document; the same source presented as verified is a defect.

What is a secondary citation and how do I handle it?

It is citing a source for something that source in turn cites from elsewhere. There are two honest treatments: go to the origin, read it and cite it, which is always preferred; or where the origin is genuinely unobtainable, cite the origin and state that you read it in the secondary source. What is not permitted is citing the origin as though you had read it, which is the default in most documents and is how errors propagate for decades unchecked.

How do I tell whether a source is real research or marketing?

Ask whether a method is available. A survey of 2,000 professionals with no sampling description, no fieldwork dates, no instrument and no base breakdown is an assertion with a number attached. Press releases, vendor reports, sponsored surveys and trade-press summaries are not disqualified automatically, and a vendor's data about its own market can be the best available, but the type has to be named in the citation and the commercial interest disclosed.

How do I check a source that changes, like a web page or a statistical table?

Record the version or edition, the access date, and a captured extract of the specific passage or figure you relied on. Then check whether it has changed since it was cited. Revisions are common and quiet: series get rebased, guidelines are superseded, pages update with no visible history. A link records where you looked; an extract records what it said, and it is the only thing that survives the source being revised.

When should citations be verified?

While sources are being collected, not at the end. A source checked at the moment of reading costs a minute. The same source checked after the document is written costs the paragraph, the argument that paragraph was carrying, and the credibility of everything near it. Verification is mandatory rather than discretionary wherever any part of the document was produced with AI assistance.