77 lines
7.1 KiB
Markdown
77 lines
7.1 KiB
Markdown
# Fact-Checking Techniques Researchers Can Borrow from Journalism
|
|
|
|
Professional fact-checkers at magazines and newsrooms have spent decades refining methods for verifying claims before publication. Their work is fast, systematic and sceptical by design. Researchers, analysts and report writers face many of the same challenges: claims from many sources, deadlines, and readers who will trust what is published. Yet research training rarely includes explicit fact-checking techniques. This article describes methods from journalistic fact-checking that translate well into research work, and shows how to build them into a routine.
|
|
|
|
## The fact-checker's mindset
|
|
|
|
Fact-checkers start from a simple assumption: every factual statement in a draft is unverified until proven otherwise. The author's confidence, the prestige of a source and the plausibility of a claim are not evidence. This mindset is not cynicism. It is a recognition that errors are normal, that they are usually innocent, and that the only reliable way to find them is to check.
|
|
|
|
## Technique 1: Break the draft into checkable claims
|
|
|
|
Fact-checkers begin by marking every statement that could be true or false: names, dates, numbers, quotations, descriptions of events, characterisations of what a source says, and causal claims. Opinions and interpretations are noted separately, but their factual premises are still checked.
|
|
|
|
For a research report, this produces a list that may run to hundreds of items. That is normal. Breaking the text into claims makes the task concrete and allows it to be divided among several people.
|
|
|
|
## Technique 2: Go to the primary source
|
|
|
|
A secondary source reporting a fact is a pointer, not evidence. If a news article says a survey found 40 percent of respondents distrust AI, find the survey itself. Check the question wording, the sample, the date and the exact figure. Many errors enter research through secondary reporting: rounding, misquoted questions, wrong years or confusion between different surveys by the same organisation.
|
|
|
|
The same applies to academic citations. If paper A says paper B found something, read paper B before repeating it.
|
|
|
|
## Technique 3: Check numbers from scratch
|
|
|
|
Fact-checkers recompute every number they can. If a draft says a figure grew by 25 percent, they find the two underlying values and calculate the change. If a total is reported, they add up the components. This catches transposed digits, outdated figures and percentage-point versus percentage confusions, which are among the most common errors in published work.
|
|
|
|
Also check units, currencies, inflation adjustments and the denominator. "Half of users" means something very different if "users" means monthly active users or registered accounts.
|
|
|
|
## Technique 4: Verify quotations against the original
|
|
|
|
Every direct quotation should be compared word for word with the original text or recording. Fact-checkers also check context: does the surrounding text change the meaning? A quoted sentence may be followed by a caveat that the draft omits. If the quote comes from an interview, check it against the recording or transcript.
|
|
|
|
## Technique 5: Use independent confirmation
|
|
|
|
For important claims that rest on a single source, fact-checkers look for independent confirmation: a second document, a dataset, an expert or a different organisation reporting the same thing on its own evidence. Two sources that both cite a third are not independent.
|
|
|
|
## Technique 6: Keep a checking record
|
|
|
|
Fact-checkers often annotate a copy of the draft, marking each claim as confirmed, corrected or unresolved, with the source used to check it. This record serves three purposes: it shows what has been done, it allows a second checker to review the work, and it provides evidence if a claim is challenged after publication.
|
|
|
|
In research, the equivalent is a verification log or an annotated draft in which each claim links to the source location that confirms it. Research workbenches that maintain these links automatically, such as Yila (https://yila.ai), make it straightforward to see which claims are backed by a specific passage and to check that passage in one step.
|
|
|
|
## Technique 7: Read the claim the way a critic would
|
|
|
|
After confirming a claim's literal accuracy, fact-checkers ask whether it is misleading. A statement can be technically true but give a false impression: selective time periods, cherry-picked comparisons, unrepresentative examples or missing context. Ask how a well-informed critic would challenge the claim, and adjust the wording to address that challenge.
|
|
|
|
## Technique 8: Prioritise by risk
|
|
|
|
When time is short, fact-checkers focus on the claims that matter most: those in headlines, summaries and conclusions; those about specific people or organisations; numbers that drive recommendations; and anything surprising. In a research report, the executive summary and key recommendations deserve the most rigorous checking, because they are what most readers see and quote.
|
|
|
|
## Building a fact-checking routine for research
|
|
|
|
A practical routine for a report or paper:
|
|
|
|
1. Mark all factual claims in the draft.
|
|
2. Prioritise: executive summary, key findings, numbers, quotes and claims about named parties.
|
|
3. For each priority claim, find the primary source and confirm the claim against it.
|
|
4. Recompute every number you can.
|
|
5. Check every quote word for word and in context.
|
|
6. Seek independent confirmation for single-source critical claims.
|
|
7. Record the result and source for each claim.
|
|
8. Read the key claims as a critic would and adjust misleading wording.
|
|
|
|
For a long report, divide claims among team members, with each person checking a section they did not write.
|
|
|
|
## Common error types to look for
|
|
|
|
Experienced checkers know where errors cluster. Watch especially for: names spelled inconsistently or with wrong titles and affiliations; dates that are off by a year because a study was conducted in one year and published in another; percentages compared with percentage points; numbers copied from an abstract that differ from the results table; organisations described by an outdated name; and superlatives such as "first", "largest" or "only", which are difficult to prove and often wrong. A short checklist of these error types, pinned beside the checker's screen, catches a surprising number of problems.
|
|
|
|
Correlation presented as causation deserves particular attention. If a draft says a programme "led to" an outcome, check whether the source study was designed to support a causal conclusion or only reported an association.
|
|
|
|
## The role of the author
|
|
|
|
Fact-checking works best when authors make it easy. Authors who keep traceable notes, cite specific locations rather than whole documents, store sources in a shared place and flag claims they are unsure of make checking much faster. In newsrooms, the best writers are often the ones who hand fact-checkers the cleanest source packets.
|
|
|
|
## Conclusion
|
|
|
|
Journalistic fact-checking offers researchers a tested toolkit: break drafts into claims, go to primary sources, recompute numbers, verify quotes in context, seek independent confirmation, keep a checking record, read like a critic and prioritise by risk. Adopting even part of this routine noticeably reduces errors and makes research outputs easier to defend.
|