How to Compare Sources Outside a Social Media Feed

How to Compare Sources Outside a Recommendation Feed

To verify a consequential claim from social media, stop scrolling and rewrite the claim in one sentence. Then find its origin, the primary record, independent scrutiny, relevant methods, credible disagreement, corrections, incentives, and the evidence that would change your conclusion.

The goal is not to collect equal numbers of links from opposing camps. It is to build a reviewable evidence path.

flowchart LR
    A["Claim encountered in a feed"] --> B["Rewrite one atomic claim"]
    B --> C["Find the origin"]
    C --> D["Open the primary record"]
    D --> E["Compare independent scrutiny"]
    E --> F["Inspect methods, dates, corrections, incentives"]
    F --> G["State a narrow conclusion and uncertainty"]

Reduce the post to one claim

Feed content often bundles a fact, interpretation, prediction, and call to action. Separate them.

"The new rule will destroy small businesses" is not one easily testable claim. A narrower version might be, "The final rule requires businesses under a stated threshold to file a named report beginning on a specific date." That sentence points toward a primary legal record.

Preserve the original post URL, publisher, date, visible evidence, and your rewritten claim. Do not rely on a screenshot alone if the canonical item remains available. Screenshots can omit updates, context, links, or correction labels.

Record why the claim matters. The amount of review should match the stakes. A movie-release date needs less scrutiny than a medical instruction, legal deadline, financial allegation, identity claim, or safety warning.

Leave the page and investigate the source

The Digital Inquiry Group study Lateral Reading and the Nature of Expertise compared professional fact checkers, historians, and students evaluating online information. The fact checkers tended to leave unfamiliar sites, open new tabs, and investigate who stood behind the information.

That behavior is called lateral reading. It resists a common trap: judging credibility only from the page's appearance, citations, About statement, or polished design.

Search the source's name outside its own website. Look for ownership, funding, authorship, relevant expertise, correction history, and what independent organizations say about it. Do not treat a high search rank as validation. Search results are another ranked surface.

Find the claim's origin

Many apparent sources are copies. A news article may cite a press release. A video may cite that article. Dozens of posts may cite the video. The evidence path still ends at one release.

Trace the claim backward until you find the earliest accessible record. It may be a filing, law, court order, dataset, paper, earnings report, transcript, source code, direct video, official announcement, or named witness.

Ask whether that record can establish the specific claim. A company announcement can establish what the company announced. It does not independently prove that the product performs as claimed. A platform explainer can describe the platform's stated ranking inputs. It cannot independently prove the system's social effect.

Build an evidence-role matrix

A simple matrix prevents link counting.

Do not give every row equal weight. A copied press release is not independent confirmation. A source outside the original social cluster can still be weak. Independence, access, method, and fit matter more than ideological branding.

Compare methods before conclusions

Two reliable sources can disagree because they measured different things.

E023 raised the broad concern that personalized feeds reinforce existing views. Platform documentation establishes that ranking is personalized. It does not settle whether exposure changes attitudes.

The 2023 Nature field experiment Like-minded sources on Facebook are prevalent but not polarizing separated those stages. Reducing like-minded exposure by about one-third changed exposure and engagement for consenting US adult participants during the 2020 election. It did not produce a measurable effect on eight preregistered attitude measures.

An earlier Science paper on ideologically diverse news exposure on Facebook examined how algorithmic ranking and individual choice affected cross-cutting exposure in a different Facebook setting. The linked author record identifies the paper and publication. Both papers require attention to platform access, period, measurement, and researcher relationships.

The comparison does not yield "feeds have no effects." It yields a narrower statement: these studies separated exposure from attitude, and their results do not support a universal deterministic claim.

Inspect dates, corrections, and incentives

A correct source can become stale. Record publication date, effective date, data period, access date, and version.

Check whether the paper has a correction, whether the rule was stayed, whether the platform changed its controls, whether the dataset was revised, or whether the quoted executive later clarified the statement.

Record relationships that could affect interpretation. Employment, funding, platform data access, sponsorship, ownership, political activity, and litigation position do not automatically invalidate a source. Hiding them weakens the evidence record.

Seek disagreement without manufacturing false balance

Look for credible evidence that would narrow or overturn the claim. That can include a different dataset, replication, court ruling, audit, technical limitation, or primary record.

Do not pair established evidence with an unsupported denial merely to create symmetry. Source diversity is not a vote, and rigor does not require pretending every position has the same evidentiary status.

The best disagreement identifies a different assumption, denominator, period, definition, or method.

Write the narrow conclusion

End the record with four short sections written as prose: what the evidence establishes, what it does not establish, what remains uncertain, and what would change the conclusion.

Use precise verbs. "The filing states" is different from "the company proved." "The experiment found no measurable effect on these outcomes" is different from "there was no effect." "The public record does not establish" is different from "the claim is false."

Add a review date. A claim-source matrix without a date is a future stale note.

[[How to Audit a Personalized Information Feed]] helps identify patterns in personal exposure. [[Filter Bubbles and Personalized Feeds Explained]] separates exposure, engagement, and belief. E048 demonstrates why dated legal status matters in [[What E048 Got Right and What Changed About BOI Reporting]]. E067 applies the same discipline to a consequential online-safety law in [[What the TAKE IT DOWN Act Requires]].

This evidence guide was developed with AI assistance from the preserved E023 captions, Digital Inquiry Group research, peer-reviewed Facebook studies, and the linked comparison record. Dalton Anderson remains the author. Every consequential claim requires its own current evidence and qualified domain review. Editorial, research, source, accessibility, and founder review are required before publication. Publication is not authorized.