FILE 015 / SOURCES & VERIFICATION
Featured artwork: an original AI-generated conceptual illustration of repeated reports, not a real collection of source documents.
Ten websites carry the same striking story. Before treating that spread as confirmation, follow the claim backwards. Those pages may lead to several investigations, or to a single announcement that has travelled widely.
The useful question is what each publication adds. A new observation, a separately obtained record and a rewritten headline do different jobs. You can keep those differences visible without dismissing accurate reporting or pretending that a large stack of links settles the issue.
Follow the trail: trace the source · use six checks · try the fictional example · weigh the support · make your record
Follow the evidence behind each page
Begin with the exact proposition you want to investigate. “An organiser announced a sighting” and “the reported object was present” require different support. A genuine announcement can establish what its issuer said while leaving the event itself unresolved. Our primary-source and authenticity definitions explain that distinction.
Then follow attribution: which interview, document, dataset or observation supports that particular sentence? Different domains, bylines or wording do not by themselves establish independent sourcing. One article can contribute its own observation to one part of a story while borrowing another part. Mark a relationship as unknown when the reporting does not reveal it.
Credited reuse is a normal part of publishing. The Associated Press’s standards describe permitted use of others’ material and clearly identified retransmission. For news releases, AP requires credibility assessment, additional sourcing and attribution in its own reporting. These are AP’s standards; the broader lesson is to inspect what was added. Shared wording is a reason to trace a connection, not grounds to accuse a publisher of fraud.
Read beyond the page
A polished website can supply a great deal of information about itself. Check outside it as well. Search the publication, author or institution, examine other accounts of their work, and locate the documents behind the specific claim. This habit is often called lateral reading.
In a 2019 study by Sam Wineburg and Sarah McGrew, 45 participants—ten historians, ten professional fact checkers and 25 Stanford undergraduates—evaluated online material while explaining their thinking. The publisher’s abstract reports that fact checkers commonly left the original site to investigate it elsewhere and reached more warranted judgments in less time. This was an expert/novice comparison, not a guarantee that the method or any profession is infallible.
Checking a publisher’s reputation and tracing a claim’s evidence are related tasks. A well-regarded outlet may accurately attribute an assertion it has not independently established. Read the attribution carefully before converting “the organisation says” into a finding of your own.
Trace a repeated claim in six steps
AFP’s published verification methods describe seeking original material, checking data at its source and preserving a trail readers can inspect. Use the following sequence to apply that approach to a cluster of pages.
- Make the claim precise. Separate the event, its date, the reported observation and any proposed explanation. Write down which part you are testing.
- Record each item. Save its URL, title, credited author, stated publication or update date, access date and relevant passage. Keep an archived copy when available; distinguish what the page states from what you have verified.
- Follow attribution upstream. Open cited reports, releases and records. Search a distinctive phrase to find other versions. Call a result the earliest version you located unless you have stronger grounds for identifying the actual origin.
- Group shared material. Note which pages use the same release, interview, image or dataset. Trace paraphrases too. If a page provides a fresh interview, ask whether it adds new knowledge or repeats the same originating assertion.
- Assess what each record supports. Check its origin, integrity, access to the event, methods and limitations. Compare agreements and discrepancies. Separate records can still be mistaken or irrelevant to the precise question.
- Write a limited finding. State the traceable accounts, the support for each part of the claim and the gaps. Identify the next record or observation that could change the assessment.
A fictional example: five pages, one announcement
This example is entirely invented. The harbour, event, documents and websites do not refer to a real incident.
Five websites carry a claim that a blue ring hovered above a harbour at dusk, after boat operations had closed. Tracing the pages produces the following record:
- Site A — the announcement. An evening-watch organiser publishes a release describing the ring and saying several visitors saw it. Their separate accounts are not included.
- Site B — the roundup. It attributes the story to the organiser’s release, repeats its quotation and adds no separately obtained observation.
- Site C — the newsletter. It links to Site B. The wording changes; the account still comes from the announcement.
- Sites D and E — two reposts. D links to Site A; E credits Site C. Neither adds reporting. They are two webpages, grouped here because they play the same role.
- A separate harbour log. In this invented file, an attendant independently recorded an evening-watch group on the promenade after boat operations ended. The entry predates the announcement and says nothing about a ring of light.
The five webpages trace to one published account of the ring. This does not establish how many witnesses existed, whether the release is authentic, or whether its account is true. It establishes the information path specified in this fictional example.
The log adds separate support for the gathering and timing. It does not confirm the ring, its behaviour or its cause. In a real investigation, the log’s origin, date and integrity would also need checking; an earlier timestamp alone would not establish independence.
Five webpages repeat one announcement. A separately recorded log supports the gathering and timing. This file contains no independently obtained observation of the claimed ring.
That finding leaves the sighting unresolved. Another page repeating the announcement would extend its reach while preserving the same evidential gap.
Ask what an additional record changes
The Library of Congress’s source-analysis guidance treats corroboration as comparison: examine agreement, disagreement, context and the questions that remain. Apply that comparison to individual claims. A record confirming a gathering need not confirm what participants reportedly saw.
A fresh interview with the person who issued the original statement may clarify their account without supplying another witness. A different document may independently establish a date while offering no evidence about a proposed explanation. Record the contribution at that level, rather than labelling an entire article independent or dependent.
If two pages cite each other and neither exposes an underlying record, write down the loop. You have found a gap in the available support. That gap does not, by itself, prove the claim false or establish that no other evidence exists. The next useful step is to locate the record the claim depends on.
No fixed number turns links into proof
AFP states that it requires at least two independent information sources for the central claim of its fact-checks. That is AFP’s editorial policy, not a universal formula for truth. The adequacy of evidence depends on the proposition, the records and the methods used.
One authentic catalogue entry can establish that the catalogue lists a particular date. Establishing the object’s actual date may require more work. Several genuinely separate witnesses can still have limited viewpoints or make similar mistakes. Independence matters alongside relevance and quality.
Likewise, different ownership does not establish different sourcing, and shared ownership does not prove every observation was shared. Investigate how the information was obtained. Avoid turning a source count into a confidence percentage without a justified model.
Make your own source record
Use the Evidence Field Notes worksheet to keep the exact claim, source trail, observations and provisional conclusion together. In the source field, group pages by the material they actually use. State uncertain relationships explicitly.
For a focused comparison, use the Source Trail Record to list page identifiers and trace shared origins. Distinguish independently obtained support, mark uncertain relationships and name the next evidence to seek.
- Documented: what the available records establish, and for which part of the claim.
- Interpreted: the reasoning connecting those records to your conclusion.
- Unresolved: the missing or conflicting evidence and what you would look for next.
Take the wider investigation through the Evidence Engine’s thirteen checks. Use the Evidence Glossary for terms, and The Phantom Footnote when the trail includes AI-generated citations. For an image-based claim, begin with the image-verification guide.
A useful record lets the next reader retrace the support and see where your conclusion stops.
Sources and research note
- Wineburg & McGrew (2019), Lateral Reading and the Nature of Expertise — original research; the study description above uses the publisher’s public abstract.
- AFP Fact Check, How we work — original material, independent sourcing and transparent verification.
- The Associated Press, Telling the Story — attribution, credited reuse and news-release standards.
- Library of Congress, Evaluating Sources and Using Evidence — corroboration, context and comparison.
Research checked November 18, 2026. This is a methods guide with an invented teaching example, not a report of an actual sighting, a systematic survey of websites or a firsthand experiment. The linked publisher abstract was reviewed; full-text access to the Wineburg–McGrew paper was not obtained. The practical workflow is DEVIL13’s explanatory synthesis of the linked methods.
