Professional fact-checking comes down to one habit: never let a claim rest on a single source, especially not the source who benefits from it. Before publication, a checker traces every factual statement back to an original record, an original document, or a person qualified to know. If the claim cannot be traced, it does not run, or it runs with the uncertainty stated plainly.
The work is slower than it sounds and mostly unglamorous. It is phone calls, document pulls, and re-reading your own copy line by line. What follows is the sequence professionals use, in the order they use it, and what each step actually settles.
What counts as a fact in the first place?
Start with the definition, because half of fact-checking is knowing what you are checking. Wikipedia describes a fact as a true datum about an aspect of the world or an occurrence in it, and notes that standard reference works are often the first stop for checking one. Dictionaries say much the same in plainer terms: the Cambridge Dictionary treats a fact as something known to be true, while Dictionary.com separately flags a second, older usage in which "fact" can simply mean something said to be true, whether or not it is.
That second usage is the checker's whole problem. A claim is not a fact because someone stated it. The job is to move a statement from the second category to the first, or to say honestly that you could not.
It also helps to know what a fact-check will never settle. Values, predictions, and matters of opinion are not checkable the way a date or a dollar figure is. A good checker separates those out early, so the verification effort goes where it can actually do something.
Where does a checker start with a claim?
The first pass is a triage, and it happens fast:
- Name the claim precisely. "The mayor raised taxes" is not a checkable claim. "The mayor signed an ordinance raising the property tax rate in March" is. Vague claims cannot be verified or refuted, so sharpen them first.
- Sort by risk. A misspelled street name embarrasses you. A wrong budget figure or a false accusation damages someone. High-risk claims get the most time, not the ones that are easiest to check.
- Ask what the ideal source would be. For a spending claim, the budget document. For a quote, the recording or the transcript. For a legal status, the court file. Then go get that source, not a retelling of it.
That last step is where most shortcuts happen, and where most corrections are born. A wire story, a blog post, or another outlet's report is a retelling. It can point you to the original, but it is rarely the original.
Why does the original source matter so much?
Because every retelling adds a chance of error and strips away context. When a checker finds the primary source — the study itself, the filing itself, the full video rather than the clip — three things happen at once. The claim gets confirmed or corrected. The context around it becomes visible, including the parts that complicate the story. And the wording becomes exact, which matters when a number or a quote is going into print under your outlet's name.
Documents deserve their own caution. A record that looks official may not be authentic, and a real document may not say what a source says it says. The craft of reading records — matching names, dates, and stamps against independent anchors — is its own discipline, covered in depth in Document Verification in the Field: How to Authenticate Records Before Publication.
How do you handle a source you cannot independently confirm?
Real newsrooms face this every week, and the honest answer is a hierarchy of fallbacks rather than a bright line. In rough order of preference:
- A second, independent source with no stake in the claim and no shared origin with the first. Two outlets running the same press release is not independence.
- Physical or documentary evidence — a receipt, a timestamped photo metadata trail, a public filing.
- An expert qualified in the specific subject, who can tell you whether a claim is plausible and what would confirm it. Experts confirm plausibility, not facts, and a quote from one should say so.
- On-the-record attribution of the uncertainty itself. If nothing else works, the story says who claims it, what their stake is, and that it could not be independently confirmed. That is a weaker story, but an honest one.
Anonymous sourcing sits at the bottom of this ladder and needs its own rules, not improvisation. If your desk has not written them down, How to Write an Anonymous-Sources Policy Readers Can Actually Understand walks through why the policy has to be public.
What does the final check before publishing look like?
The last pass is not about new reporting. It is about your own copy. Read every sentence and ask three questions of each: Is this a fact or an inference? If a fact, where did it come from? Does the story say what the source actually supports, no more?
Watch for the quiet failures. A number that was true of one year presented as current. A quote trimmed until it means something the speaker did not say. A hedge in your notes — "source believes," "reportedly" — that hardened into flat assertion by the third draft. These are the errors that survive every draft because each one looked fine in isolation.
A written checklist beats memory under deadline, and a one-person operation needs one most of all. The Pre-Publish Verification Checklist a One-Person Desk Can Run offers a version sized for small staffs. Readers following this should also see The Pre-Publish Verification Checklist a One-Person Desk Can Run.
What this means for your newsroom
Verification scales down better than people assume. A solo reporter cannot run a dedicated fact-check desk, but can run the sequence: sharpen the claim, find the original, seek independence, and state what remains unconfirmed. The expensive part was never the checking. It is the correction after publication, the source call that goes badly, and the reader who stops trusting the byline.
The limits are real, too. This sequence works well for discrete factual claims — figures, dates, quotes, documented events. It works poorly for contested judgments, ongoing events where records lag, and claims about motive, which no document can settle. In those cases the professional move is to show the evidence and let readers see the reasoning, rather than to declare a verdict.
And when a check fails after publication anyway, the response is the same discipline in reverse: move fast, be visible, and fix the record. Issuing Corrections — The Correction Conversation: How to Call a Source Back When You Got It Wrong covers the hardest part of that, the call to the person you wronged. We covered a connected angle in Issuing Corrections — The Correction Conversation: How to Call a Source Back When You Got It Wrong.
