How to tell a primary source from a repeated claim

A primary source is the closest available evidence for the particular question you are asking. It is not simply the first result, the official-looking page, or the oldest document. And “primary” describes the source's relationship to a claim—it does not certify that the source is accurate, complete, or honest.

To distinguish a primary source from repetition, define the claim, identify who could directly know or record it, and trace every citation back to the item that actually contains the evidence.

Make the claim specific first

“The company changed its policy” is too broad. Break it into testable parts:

• what exact language changed?
• which document or product did it govern?
• who had authority to make the change?
• when did the previous and new versions apply?
• is the claim about what was announced, implemented, or experienced?

The best primary source depends on that question. A company announcement is primary evidence that the company announced something. It may not prove the system actually behaved that way. A user's dated receipt can be primary evidence of that transaction, but not of the policy for every user.

Match the source type to the claim

Use a source map:

| Claim | Likely direct evidence | Common repeated version | | --- | --- | --- | | A person said specific words | Full recording, transcript, speech, letter, or attributable interview | Quote graphic or article quoting another article | | A rule has legal effect | Enacted text, regulation, order, docket, or official register with status | Blog summary or search snippet | | An organization published a policy | Versioned policy, official notice, or archived page | News rewrite or customer forum post | | A number was reported | Original dataset, table, filing, or methodology from the responsible body | Chart screenshot or article repeating the figure | | A study found an effect | The study, protocol, data, and correction record | Press release or headline | | An event occurred at a time and place | Contemporaneous recording, records, direct observations, and physical evidence | Later retelling or viral caption |

“Official source” and “primary source” overlap but are not identical. An agency FAQ summarizing a statute is official and secondary to the enacted text. A company statement about a competitor is primary evidence of the company's statement, not of the competitor's conduct.

Follow the citation chain backward

Start from the page carrying the claim and record exactly what it cites. Open the cited item. Confirm that it contains the same claim, then repeat until you reach the underlying document, data, or recording.

Do not accept phrases such as “reports say,” “according to research,” or “it is widely known” as citations. Search a distinctive sentence or figure in quotes and look for an earlier attributable version. Check footnotes, image credits, dataset identifiers, docket numbers, report appendices, and transcript timestamps.

Watch for citation laundering:

1. article A makes an unsupported claim
2. article B repeats A without checking it
3. article C cites B
4. article A is later updated to cite C

The loop can look like several sources in search results while containing no direct evidence. Draw the links. Sources are independent only when their evidence or reporting is independent.

Inspect the item, not the label

Once you reach a candidate primary source, verify:

• Identity: Who created, signed, recorded, or issued it?
• Date and version: Is this the edition or state relevant to the claim?
• Completeness: Is it the whole document or a crop, excerpt, draft, or selected table?
• Context: What was its purpose, audience, and surrounding event?
• Provenance: Where did this copy come from, and what happened to it since creation?
• Status: Was it enacted, final, corrected, retracted, superseded, or merely proposed?

The Library of Congress describes primary sources as original documents and objects created at the time under study, while noting that they are often incomplete and require comparison with other sources. The National Archives' analysis method similarly moves from identifying and observing a document to making sense of it and using it as evidence.

That process matters because a primary source can be mistaken, biased, staged, fraudulent, or limited to one viewpoint. Directness answers “how close is this to the event or record?” Credibility answers a different question.

Record provenance as a chain

For material claims, keep a compact record:

| Field | What to write | | --- | --- | | Claim ID | One sentence that can be tested | | Direct item | URL, document ID, edition/page, dataset version, or timestamp | | Creator and role | Who made it and why they could know | | Creation or issue date | Date supported by the item or catalog | | Copy provenance | Archive, repository, publisher, or transfer path | | Transformations | OCR, transcription, crop, translation, reformat, or excerpt | | Supports | The narrow fact this item establishes | | Does not support | The tempting broader conclusion it cannot establish |

The W3C provenance model distinguishes entities, activities, and agents. You do not need its full ontology for ordinary research, but the separation is useful: what is the item, what process produced or changed it, and who was responsible?

Compare versions and corrections

Primary sources change. Policies are updated, datasets are revised, transcripts are corrected, and studies are amended or retracted. Record the exact version you used and look for a correction log, replacement file, later docket entry, or archived state.

A later corrected version may be the best evidence of the current record while the earlier version remains primary evidence of what readers saw at the time. Do not silently substitute one for the other.

Translations and transcripts add another layer. When possible, retain the source-language item or recording alongside the translation or text. Label who translated or transcribed it and whether the version is official, automated, or independent.

Use secondary sources for the work they do well

Secondary sources can add context, compare multiple records, identify expert consensus, explain technical material, and expose problems in a primary source. They are not inferior by definition. A rigorous synthesis may answer a broad question better than one raw document.

The mistake is to present a secondary source as direct evidence it never inspected. Prefer a transparent combination:

• cite the primary item for what it directly records
• cite high-quality synthesis for interpretation and wider context
• disclose why no suitable primary source is available when that is the case
• show evidence that supports and evidence that challenges the claim

The International Fact-Checking Network's standards embody this approach by requiring significant evidence to be identifiable, suitable primary evidence to be used when available, and key claim elements to be checked against more than one named source except where only one relevant source exists.

Label the result precisely

Use source-role labels instead of a binary good/bad score:

• Direct primary source: the item directly records the event, statement, rule, measurement, or transaction being checked.
• Official summary: authoritative about the issuer's position but secondary to the underlying text or data.
• Contemporaneous account: created near the event, with a stated viewpoint and access that still require evaluation.
• Secondary analysis: interprets or synthesizes identifiable evidence.
• Repeated claim: restates the proposition without adding inspectable evidence.
• Source role unresolved: the chain ends before the underlying evidence can be checked.

If the chain depends on a circulating quotation, use how to trace a quote to its original source. To retain the item and its version for others to audit, follow how to preserve and document a web source.