# Check the source behind a claim before sharing it

Complete original constructed teaching cases. No real people, studies, source datasets, customers, accounts or external publications.

Separate the exact claim from the publisher, trace its original evidence and check context and independent support. Leave the page to investigate who is behind it instead of relying on its own presentation. Record supported, contradicted and unresolved parts separately. A familiar logo, many repeated articles, a citation or a current date does not verify the conclusion.

## Write the checkable statement first

Copy the exact proposed sentence into a private review note and identify the source version. Separate a factual assertion, a forecast, an opinion and an advertising promise. “Eighteen downloads completed” is a different assertion from “eighteen people found the guide useful.” Write the claimed population, unit, period and comparison before searching for a number that appears to fit.

Decide what evidence would answer that particular question and what would contradict it. A source about one account permission cannot establish every private metric. A current help page can describe a documented control without proving that an actual account has it or that a publication succeeded. Keep a narrow answer when the available source only supports a narrow claim.

## Look outside the publisher’s own account of itself

Check who operates the source, relevant expertise, funding or commercial interests, intended audience and correction practices. Search outside the page for reliable independent context. The Digital Inquiry Group calls this lateral reading; the public introduction contrasts it with remaining on one page. Its linked teaching materials require access, and we did not inspect those lessons or reproduce an underlying study.

The Commons Library’s source-evaluation briefing treats authority, relevance, transparency, purpose and worldview as useful questions. An interest or perspective does not by itself make a statement false, and a prestigious affiliation does not make every statement true. Expertise must match the claim. Use these clues to choose the next evidence check rather than manufacture a numerical trust score.

## Trace the evidence and its actual scope

Open the referenced original where possible: the actual data definition, full method, relevant document or first-hand record. Read the passage and surrounding limits rather than a snippet. Follow who cites whom. Three articles repeating one announcement provide three publication locations, not three independent measurements. A primary source can also be mistaken, partial or self-interested; inspect what it actually establishes.

FactCheck.org’s published process emphasizes primary evidence, contextual editing, factual review and explained corrections. That is a description of its newsroom, not a claim that our automated guide has its independent editors or has contacted a source. The Commons briefing also notes that peer review is no guarantee. Retain conflicting evidence and explain why one source is more relevant to the exact question.

| Constructed reference | Origin | Bounded finding |
| --- | --- | --- |
| P cites R | R | No new measurement |
| Q cites R | R | Same declared origin |
| R reports18/24 | Request log | Requests, not preference |

## Resolve three complete source checks

V1 — Constructed repetition: pages P and Q cite announcement R; R reports 18 completed file requests out of 24 attempts, without people or usefulness outcomes. Proposed headline: “Three independent sources prove 75% of readers prefer the guide.” Decision: reject the independence, people and preference claims. Bounded draft: “In this fictional request log, 18 of 24 file requests completed; reader usefulness and preference were not measured.” Outcome: a corrected interpretation, not validation of R’s collection method.

V2 — Constructed scope change: help source H describes permission to create a member post, while a promotional article claims that the permission also retrieves every historical private impression. Decision: inspect the exact documented scope and leave the additional reading/history claim unresolved unless relevant evidence supports it. Bounded draft: “The described permission concerns publishing; this source does not establish the requested private history.” Outcome: no invented vendor certification or conclusion that every connected tool has the same access.

V3 — Constructed unavailable origin: a widely repeated statistic points to a report that cannot be inspected, and no preserved method or relevant independent result is available. Decision: hold the statistic or attribute its unverified status explicitly; do not relabel absence as proof of falsity. Bounded draft: “The cited source was unavailable at this check, so the stated result could not be verified.” Outcome: a documented limit, not a recovered study, a true/false verdict or an accusation about a person.

## Keep a claim-level decision and a correction route

For each important clause, record support, contradiction, missing evidence and the proposed public wording. Separate “the source states” from “the underlying result was independently checked.” If a changeable limit matters to the action, retain the date and relevant surface. A readable quotation with a working link can still omit a decisive qualification.

Before a real reshare, review the actual caption, image, translation and destination for the same unsupported assertion. Correct affected outputs rather than merely appending a vague source list. Keep a clear hold decision when material evidence is missing. This source review does not grant permission to reuse a photograph, reveal private correspondence or issue a legal or medical conclusion.

## Save the trace, not an automatic truth badge

The private record separates exact claim/version, publisher context, original reference, scope, citation lineage, independent evidence, contradictory or unavailable material, final wording and remaining checks. Minimize personal details. The original V-cases and fixture demonstrate how repeated origins and changed units affect a conclusion; a deterministic checker cannot decide whether an arbitrary outside claim is true.

Relevant primary bodies were read on 1 November 2026. The Commons page returned 403 to the direct fetch but its actual official web body was read. A former Full Fact process URL returned 410 and is excluded despite a cached search result. No subject was contacted, no underlying DIG study or gated lesson inspected, and no independent newsroom review or reader outcome is claimed.

1. Record exact claim, unit and source version.
2. Read original evidence with context and method.
3. Investigate the publisher beyond its own page.
4. Retain lineage, independent evidence and contradictions.
5. Save bounded wording or hold decision with open evidence.

## Primary sources

- [Digital Inquiry Group: Public lateral-reading introduction](https://cor.inquirygroup.org/curriculum/collections/teaching-lateral-reading/)
- [Commons Library: May 2026 source-evaluation briefing](https://commonslibrary.parliament.uk/research-briefings/cbp-10822/)
- [FactCheck.org: Primary evidence, review and correction process](https://www.factcheck.org/our-process/)