A Repeatable Workflow for Checking Online Claims

Why a repeatable check beats a quick impression

A polished post can make a weak claim feel settled. A chart may have a convincing title, a familiar logo, and thousands of shares; none of those features establishes where the numbers came from or whether they support the caption. The reverse is also true: an unfamiliar website or an awkwardly designed page may link to sound primary evidence. Appearance is a poor substitute for checking.

A repeatable workflow helps separate several questions that online posts often blur together: Who is making the claim? What evidence is offered? Does that evidence actually support the wording? Have independent sources reached the same conclusion? And what remains unknown? You do not have to become an expert in every subject to ask those questions in a useful order.

This guide uses SIFT, a method developed by digital literacy educator Mike Caulfield, alongside lateral reading. Instead of staying on one page and trying to infer its trustworthiness from its design, lateral reading means opening other sources to learn about the publisher, the claim, and the evidence. The goal is not to label every source “good” or “bad.” It is to decide how much confidence a particular claim deserves, given the evidence available.

The four SIFT moves

SIFT offers a memorable sequence, not a rigid investigation with only one correct path. You may move back and forth—for example, a primary study may point you to a better search term, which leads to an independent report, which then raises a question about the study’s population.

1. Stop

Before reacting, sharing, or letting a striking claim set the terms of the discussion, stop. Notice what the post is asking you to feel or do. Is it urgent, alarming, flattering, or framed as a secret that “they” do not want you to know? Emotional force does not prove a claim false, but it can push you to skip verification.

Write the claim in one plain sentence. Preserve its scope. “A small trial found a possible benefit in a particular group” is not the same claim as “This treatment works.” If a post combines several claims, split them up. A statistic, a proposed explanation, and a recommendation may each require different evidence.

2. Investigate the source

Find out who is behind the page or account. Search the publisher’s name in a new tab rather than relying only on its own “About” page. Look for independent descriptions of its ownership, expertise, track record, funding, and corrections practices. A biography supplied by the source can be useful, but it is still a claim made by the source itself.

Check the author and the publication, not just the logo. Is the author identified? Do they have relevant expertise for this specific topic? Does the organization have a clear way to correct errors? Are conflicts of interest disclosed? A scientist may be knowledgeable about one field without being qualified to evaluate every health or policy claim. Expertise is relevant, but it is not a blank cheque.

Also check the date and the original format. An old article may describe evidence that has since changed. A clipped video may omit the question that prompted a speaker’s answer. A screenshot may strip away a headline’s qualification or make a joke look like a report. Whenever possible, locate the live page, full document, or complete recording.

3. Find better coverage

Search outside the original post. Try a distinctive phrase from the claim, the named institution, the study title, or a key figure. Compare reporting from sources that gathered their information independently. If every page repeats the same sentence and links to the same press release, that is not the same as multiple independent confirmations.

Lateral reading is useful here: leave the page and consult other sources about both the publisher and the claim. Researchers at the Stanford History Education Group have described how professional fact-checkers use this approach, moving across the open web rather than examining a single site in isolation. The practical lesson is straightforward: do not ask one source to vouch for itself.

For a scientific claim, better coverage might include a review article, a professional scientific organization’s explanation, or reporting that links directly to the paper and explains its limitations. For a legal or government claim, look for the relevant law, court document, agency publication, or official data release. For a public figure’s statement, search for the full speech, transcript, or recording rather than relying only on a short excerpt.

Independent sources need not agree in every detail. Look at what they share, where they differ, and whether their reporting gives you a way to inspect the underlying evidence. “Several outlets reported it” is less useful than knowing whether those outlets checked the same primary document or simply repeated one another.

4. Trace the claim to its original context

Follow links and citations until you reach the source closest to the evidence: a study, dataset, official record, transcript, or full report. Then compare the original material with the wording of the post. A link is not proof that the linked item supports the claim. Headlines and captions often make broader statements than the documents beneath them.

Ask what was measured, who was included, when the evidence was collected, and what comparison was made. In a study, inspect the abstract and conclusion, but do not stop there if the claim depends on a particular number. Check the methods, the results, and any limitations the authors report. In a dataset, look for definitions, date ranges, exclusions, and notes about revisions. In a quotation, check the surrounding passage.

Keep the distinction between an observation and an explanation. A study may find that two things occurred together without demonstrating that one caused the other. A rise in an outcome after a policy change does not, by itself, show that the policy caused the rise. The comparison group, alternative explanations, and study design matter.

Worked example: follow a viral statistic back to its source

Consider a fictional post that says: “A new study proves that 80% of people who use a certain app sleep better.” The post includes a chart and a link to a research paper. This example is invented to demonstrate the checking process; it does not describe a real study or establish anything about an actual app.

Stop: make the statement testable

The word “proves” is stronger than the evidence may justify. “People who use” could mean daily users, anyone who tried the app, or people who completed a survey. “Sleep better” needs a definition. Before searching, rewrite the claim as a set of questions: Did a study report an 80% figure? What did the figure measure? Who participated? Was app use compared with anything? Did the researchers measure sleep directly or ask participants how they felt?

Investigate: identify the publisher

Open the account’s profile and the linked page. Suppose the post comes from a wellness company that sells the app. That commercial connection does not automatically invalidate the study, but it is relevant context. Check whether the company funded the research, whether the authors disclosed financial relationships, and whether the post identifies the study accurately. Search the organization’s name separately to see how outside sources describe its research claims.

Find better coverage: seek independent interpretation

Search the paper’s title, author names, and a distinctive phrase from its abstract. Look for reporting or expert explanations that inspect the paper rather than repeat the company’s announcement. If independent sources describe the study as a small, short-term survey, that detail may materially change what the viral statistic means. If no reliable coverage exists, that absence is not proof of deception; it is a reason to avoid presenting the claim as established.

Trace: check what “80%” refers to

In this fictional example, imagine the paper says that 80% of respondents in a voluntary survey reported that they felt their sleep had improved. That is not necessarily the same as 80% of all app users sleeping better. The survey may have included only people who completed the questionnaire, relied on self-reports, and lacked a comparison group. Those features would limit what the result can establish.

The post’s wording has now shifted several times: “80% of respondents reported perceived improvement” became “80% of users sleep better,” and then “the study proves the app works.” Each step makes the assertion broader or more certain than the hypothetical evidence supports. A careful summary would preserve the limits: “In a voluntary survey, 80% of respondents said their sleep felt better; the result does not, on its own, show that the app caused the change or that the figure applies to all users.”

That is the key move in a source check: report what the evidence can bear, not what a headline would like it to bear.

A checklist to use before sharing

Save this short record in a note, document, or research log. A few written details make it easier to revisit a claim later and show where your conclusion came from.

• Claim: What exactly is being asserted? Have you separated factual statements from interpretation or advice?
• Original post: Who published it, and where? Record the account or publisher, page title, and link.
• Date: When was the claim posted, and when was the evidence produced? Is the information still current?
• Evidence: What is the closest available original source—a study, dataset, legal text, official record, or full transcript?
• Context: Who or what was studied? What was measured? What limitations or qualifications appear in the source?
• Independent coverage: Which sources checked the evidence themselves, rather than repeating the same announcement?
• Remaining uncertainty: What could not be verified, or what information would change your assessment?
• Confidence: Is the claim well supported, partly supported, misleadingly framed, unsupported, or unresolved?

A confidence label is a summary of your evidence, not a badge of certainty. “Unresolved” can be the most accurate result when a source is inaccessible, the original data are missing, or reputable accounts disagree for understandable reasons. Record why you chose the label; that explanation is more useful than a bare verdict.

Common traps that derail a check

Treating a search result as verification

A search snippet can omit a qualification, show old text, or combine words from different parts of a page. Open the result and inspect the source. For a consequential claim, follow its citations rather than stopping at the first page that sounds confident.

Counting copies as independent confirmation

Ten websites repeating one press release still leave you with one underlying account. Check the links and wording. If reports all trace back to the same source, note that dependence instead of treating the number of mentions as a vote.

Assuming a primary source settles everything

Primary sources are essential, but they are not automatically self-explanatory. A paper can be preliminary, narrow, or methodologically limited. An official document can establish what an agency recorded without answering every question about the real-world cause of an outcome. Read it alongside informed interpretation, and keep the distinction between data and conclusion intact.

Confusing a missing link with proof of a false claim

A post with no citation is difficult to verify, but lack of a citation alone does not show that the claim is false. It means you should locate supporting evidence independently or refrain from repeating it as fact. Likewise, a dead link may reflect a moved page, not necessarily an attempt to hide evidence. Search for the title, author, or archived citation before drawing conclusions.

Checking only the details that support your first impression

Once you suspect a claim, it is easy to notice confirming evidence and overlook inconvenient qualifications. Ask what evidence would change your mind. Search for the strongest credible account on the other side, then compare its sources and reasoning. This is not a demand to give every position equal weight; it is a way to test whether your own reading has missed important context.

When time is short

Not every post deserves a full investigation. Match the effort to the stakes. A low-impact claim about a film release may need a quick check of the publisher and date. A claim about health, safety, elections, money, or a person’s reputation merits a closer look before you act on it or pass it along.

For a fast first pass, pause, search the publisher’s name outside its own site, look for independent coverage, and open the original evidence if it is available. If you cannot finish the check, do not upgrade uncertainty into certainty. You can wait, ask for the source, or share a narrower statement about what you verified. Not sharing is a legitimate outcome of source evaluation.

Be especially cautious with screenshots and short videos. Search a distinctive sentence in quotation marks, look for a longer recording, and check whether the image or clip has appeared in another context. If a post quotes a document, search for its title and inspect the surrounding text. The original context can change the meaning of a sentence without changing a single word.

Make the workflow useful to someone else

A good check should be understandable by a reader who did not watch you search. Keep a compact trail: the claim as presented, the source and date, links to the primary evidence and independent coverage, and a short note about what the evidence does and does not show. If a link is likely to change or disappear, record the document title and publication date as well.

Use precise language when you share your findings. Instead of “debunked,” say which part is contradicted and by what evidence. Instead of “confirmed,” explain whether the source supports the full claim or only a narrower version. If evidence is mixed, say what the sources agree on and where the uncertainty remains. These distinctions help readers judge the reasoning rather than simply accepting another confident-sounding verdict.

Be willing to update your assessment. A later correction, a newly released dataset, or a better translation may change the picture. Keeping dates and links in your notes makes revision possible—and makes it clear that confidence should follow evidence, not loyalty to an earlier post.

A final practical test

Before relying on or sharing a claim, try to complete this sentence: “I believe this much because this source, published on this date, supports this specific part of the claim; this is what remains uncertain.” If you cannot fill in those blanks, the next step is not to guess. It is to return to SIFT: stop, investigate, find better coverage, and trace the evidence to its original context.

The method will not eliminate disagreement, guarantee that every source is accurate, or make complicated evidence effortless. It does something more useful: it gives you a consistent way to notice who is speaking, distinguish repetition from independent confirmation, and keep a claim within the limits of its evidence. That is enough to make a more informed decision about what to trust, what to question, and what not to pass along yet.