How to Fact-Check Sources for Academic Research
Fact-checking is claim work, not a vibe about a website. When a number, quotation, image, or “new study” shows up in a paper, leave the page and ask whether the claim is true, whose evidence it rests on, and whether that evidence still stands. This page uses Mike Caulfield’s SIFT moves, Caulfield and Wineburg’s 2023 book Verified (excerpted in the Spring 2025 American Educator), Wineburg and McGrew’s lateral reading, Wardle and Derakhshan’s information-disorder terms, and the IFCN Code of Principles. It is not how to tell if a source is credible and not the CRAAP walkthrough.
By CitationEasy Editorial Team23 min read
On this page
Quick Answer: How to Fact-Check a Claim
Stop before you paste the sentence into your draft. Identify the claim (a number, a quote, a causal finding, a date, a photo). Then do the four things Caulfield named in 2019 and restated with Wineburg in Verified: stop, investigate the source, find better coverage, and trace the claim, quote, or media back to original context. If you cannot reconstruct that context from a primary source, do not cite the claim.
Claim-checking templates (fill from the source in front of you)
Claim sentence:
[Who] claims that [specific fact], based on [named study, dataset, speech, or image], as of [date].
Trace step:
The version I first saw is a [tweet / news lede / syllabus PDF / AI summary]. The original is [DOI, transcript, dataset table, or archive item] at [stable URL].
Coverage step:
Independent coverage of the same claim appears in [second named source] and [third named source]. They [agree / disagree / cite one shared origin].
Use-or-drop decision:
Cite only if I can point a reader to checkable evidence. If the trail dies in a press release, a screenshot, or “everyone knows,” drop the sentence.
What This Page Is For
CitationEasy already has a page for is this publisher trustworthy enough to cite? (credible sources) and a page for run the official CRAAP questions on the item (evaluate sources). This URL answers a narrower question: is this particular claim true enough to put in a paper? A respected newspaper can still mangle a denominator. A peer-reviewed article can still be retracted. A true leak can still be mal-information if someone published private facts to injure a person. An AI chatbot can invent a DOI that looks scholarly and still points at nothing.
Professional fact-checkers at FactCheck.org — a verified IFCN signatory at the Annenberg Public Policy Center, last updated on that process page on 4 December 2025 — put the burden where academic writing already puts it:
FactCheck.org, “Our Process” (page updated 4 December 2025)
The burden is on the person or organization making the claim to provide the evidence to support it.
We rely on primary sources of information.
If you cannot meet that bar, the claim does not belong in the paper, even if you have not proved it false. “I could not verify it” is a reason to delete, not a reason to hedge with “some say.”
Name the Problem: Information Disorder, Not “Fake News”
Claire Wardle and Hossein Derakhshan wrote the 2017 Council of Europe report Information Disorder (DGI(2017)09) because “fake news” collapses satire, honest error, propaganda, and leaked private records into one insult. They split the mess along two axes — falseness and intent to harm:
Wardle and Derakhshan, Council of Europe (2017)
Mis-information
Mis-information is when false information is shared, but no harm is meant.
Dis-information
Dis-information is when false information is knowingly shared to cause harm.
Mal-information
Mal-information is when genuine information is shared to cause harm, often by moving information designed to stay private into the public sphere.
| Type | False? | Intended harm? | What you do in a paper |
|---|---|---|---|
| Mis-information | Yes | No | Correct the fact. Cite the better source. Do not treat the first teller as a liar. |
| Dis-information | Yes | Yes | Do not cite the claim as evidence. If you discuss the campaign, cite a verified reconstruction. |
| Mal-information | No | Yes | The fact may be real. Ask whether publishing it is ethical, lawful, and relevant to your argument. |
For academic work, intent is secondary to accuracy. A classmate who pasted the wrong year and a troll who invented the year produce the same unusable footnote. Wardle’s labels still help you choose a response: correct, refuse, or handle with care. They are not a reason to psychoanalyze a source you have not identified.
SIFT When You Care About the Claim, Not the Page
Caulfield published SIFT (The Four Moves) on 19 June 2019 as a short list of things to do, not a scoring rubric. In 2023 he and Sam Wineburg restated the same habit in Verified: How to Think Straight, Get Duped Less, and Make Better Decisions about What to Believe Online (University of Chicago Press). Their Spring 2025 American Educator excerpt is the public wording you can open without a paywall. The credibility page covers the first two moves as a way to size up a publisher. This page spends its space on the last two, because that is where students lose citations: they polish a URL and never check whether the sentence is true.
Caulfield and Wineburg, “Learning to Verify,” American Educator (Spring 2025)
The first task when confronted with the unfamiliar is not analysis. It is the gathering of context.
When we encounter something online, our first question shouldn’t be “Is this true?” but rather “Do we know what we’re looking at?”
They name three contexts to gather in about thirty seconds: the source (who is talking), the claim (what other coverage says), and you (why this sentence is tempting). That last context is why a statistic that helps your thesis needs the same trail as a statistic that threatens it.
Mike Caulfield, “SIFT (The Four Moves),” 19 June 2019
Sometimes you don’t care about the particular article or video that reaches you. You care about the claim the article is making.
You want to know if it is true or false. You want to know if it represents a consensus viewpoint, or if it is the subject of much disagreement.
Stop
Pause before you highlight, screenshot, or “just drop it in for now.” Ask what you already know about the claim and the channel. Outrage, triumph, and “share before they delete this” are not evidence. If you are already writing a paragraph that needs this number, you are the easiest person in the room to fool. Stop long enough to write the claim as a single sentence you could show a librarian.
Investigate the source — then decide whether that is the job
Open a new tab and search the author or organization minus the site you are on. That is still required. It is not sufficient. Caulfield’s own economics example is the point: knowing you are reading a Nobel economist or a dairy-industry video changes how you interpret the piece. It does not, by itself, tell you whether the statistic in paragraph four is calculated honestly. For the publisher decision, go to credible sources. For the number, keep moving.
Find better coverage
When the thing you need is the claim, Caulfield says you may ignore the article that reached you and look for trusted reporting or analysis of that claim. His worked prompt is a conservation headline, not a fake crime rate:
Caulfield’s coverage example (keep the wording; then go find real reporting)
If you get an article that says koalas have just been declared extinct from the Save the Koalas Foundation, your best bet might not be to investigate the source, but to go out and find the best source you can on this topic, or, just as importantly, to scan multiple sources and see what the expert consensus seems to be.
The 2025 excerpt says the same thing as a move, not a vibe:
Caulfield and Wineburg, “Learning to Verify,” American Educator (Spring 2025)
Engage in “lateral reading” by opening up new tabs in your browser and using the internet to check the internet.
That is not a “three-source rule” invented for this website. It is a search for independent coverage. Three news stories that all quote the same press release are one source wearing three bylines. The IFCN asks professional checkers to test key elements against more than one named source “save where the one source is the only source relevant on the topic.” A unique archival letter can be that one source. A viral chart almost never is.
Trace claims, quotes, and media to original context
Caulfield’s last move is the one students skip when the abstract already “sounds like” the finding they wanted:
Mike Caulfield, “SIFT (The Four Moves),” 19 June 2019
Much of what we find on the internet has been stripped of context.
In these cases we’ll have you trace the claim, quote, or media back to the source, so you can see it in it’s original context and get a sense if the version you saw was accurately presented.
The second sentence is Caulfield’s wording, including his “it’s.” Click through to the paper, the dataset table, the hearing transcript, or the uncropped file. Read the limitation the press release dropped. Check whether the photograph’s caption matches the event. If the trail ends at a screenshot, you do not have a source.
Lateral Reading: Leave the Page the Way Fact-Checkers Do
Sam Wineburg and Sarah McGrew watched 10 professional fact-checkers, 10 Ph.D. historians, and 25 Stanford undergraduates evaluate live sites. Historians and students stayed inside the page — logos, About text, a polished URL — and were easier to trap. Fact-checkers left. The peer-reviewed paper is “Lateral Reading and the Nature of Expertise” in Teachers College Record 121 (2019); a Stanford copy of the accepted manuscript is open.
Wineburg and McGrew, Teachers College Record (2019)
In contrast, fact checkers read laterally, leaving a site after a quick scan and opening up new browser tabs in order to judge the credibility of the original site.
Compared to the other groups, fact checkers arrived at more warranted conclusions in a fraction of the time.
Fact checkers, on the other hand, engaged in three practices—taking bearing, lateral reading and click restraint—that allowed them to read less but learn more about the topics they investigated.
Click restraint stands in contrast to whimsical clicking; before clicking on any one result, users evaluated the list of search results to understand the digital terrain in which they have landed.
The Civic Online Reasoning curriculum — now hosted by the Digital Inquiry Group, formerly the Stanford History Education Group — teaches three first questions: who is behind the information?, what’s the evidence?, and what do other sources say? For a citation, add a fourth the moment you leave: does the object I opened still say what the recap said, and has it been corrected or retracted? Vertical reading — staying on one About page until you feel convinced — is how a lookalike domain survives. It is also how a real journal article’s retracted status never gets noticed, because you never opened Retraction Watch or the publisher’s notice.
Lateral tabs for one claim (not one website)
- Tab 1 — the item that reached you. Write the claim in one sentence. Copy the date and any DOI or report number.
- Tab 2 — who is talking. Search the author or organization in Wikipedia, a library catalog, or news coverage. Do not finish the About page first.
- Tab 3 — better coverage. Search the claim in quote marks plus a field word (
mortality,census,opinion). Prefer statistical agencies and scholarly sources over recaps. - Tab 4 — the original object. Open the paper, table, transcript, or file. Confirm the sentence still means what the recap said.
- Tab 5 — later fate. Search the title plus
retraction,correction,erratum, or the author’s name on Retraction Watch.
When to Use a Professional Fact-Check — and How to Read One
A student paper is not a newsroom. You still borrow the profession’s standards. The International Fact-Checking Network at Poynter vets organizations against a public code. The current commitments page — not a random homepage badge — is the text to use. A June 2025 IFCN PDF notes that the Code itself, last rewritten in early 2020, is under a community review that runs through 2025 and into 2026. Until a replacement is published, these are the live rules:
IFCN Code of Principles, commitment to standards and transparency of sources
Signatories want their readers to be able to verify findings themselves. Signatories provide all sources in enough detail that readers can replicate their work, except in cases where a source's personal security could be compromised. In such cases, signatories provide as much detail as possible.
The applicant uses the best available primary, not secondary, sources of evidence wherever suitable primary sources are available.
The applicant checks all key elements of claims against more than one named source of evidence save where the one source is the only source relevant on the topic.
Use the IFCN signatory directory, not a random “fact-check” badge on a homepage. FactCheck.org, Full Fact, Africa Check, and Chequeado are the kind of outlets the directory is for; confirm current status on the directory itself before you treat any badge as current. A fact-check is a secondary reconstruction. Cite it when you are writing about the rumor. When you are writing about the underlying phenomenon, follow its footnotes to the primary source and cite that.
- Read the rating and the evidence list. A colorful meter is not a source.
- Open the linked dataset, statute, or paper. If the links are missing, the piece fails the IFCN source commitment for your purposes.
- Note the date. A 2018 check does not settle a 2026 statistic.
- If two IFCN signatories disagree, you have a controversy to report, not a coin flip to hide.
Worked Example 1: A Statistic with the Wrong Denominator
Use a documented case, not a made-up town statistic. In April 2021, Shimabukuro and colleagues published “Preliminary Findings of mRNA Covid-19 Vaccine Safety in Pregnant Persons” in the New England Journal of Medicine (DOI 10.1056/NEJMoa2104983). Among 827 completed pregnancies they reported 104 spontaneous abortions. A later NEJM correction told readers not to turn that count into a rate:
NEJM correction to Shimabukuro et al. (Table 4 footnote)
No denominator was available to calculate a risk estimate for spontaneous abortions, because at the time of this report, follow-up through 20 weeks was not yet available for 905 of the 1224 participants vaccinated within 30 days before the first day of the last menstrual period or in the first trimester.
FactCheck.org (Kate Yandell, 14 August 2026) reconstructed the viral 82 percent figure that still recirculated after senators released January 2021 texts from Dr. Anthony Fauci’s government phone:
FactCheck.org, 14 August 2026 (reconstruction of the arithmetic, not a quotation from the NEJM article)
The false claim of an 82% miscarriage rate comes from dividing the 104 miscarriages by the 127 people in the study who got vaccinated in the first or second trimester — a calculation that the paper did not present.
- Most people vaccinated early in pregnancy would not yet have completed a pregnancy unless it ended early, so the 127-person denominator selects for loss.
- A later CDC COVID-19 Vaccine Pregnancy Registry analysis in Vaccine 76 (18 February 2026): 128340, DOI 10.1016/j.vaccine.2026.128340 — followed 12,097 people vaccinated before 20 weeks.
- That paper’s fetuses-at-risk estimate: cumulative risk of spontaneous abortion 10.79% (95% CI 10.01–11.56), below or within pre-pandemic estimates — not an 82 percent catastrophe.
The lesson is mechanical, not partisan. Always recover the numerator, the denominator, the time window, and whether the authors computed the percentage you are about to repeat. If you cannot open the table, you cannot cite the percentage. For more on charts that lie with a real axis, see how to present data.
Worked Example 2: A Journal Article That Did Not Survive
Peer review is a filter, not a guarantee. Wakefield and colleagues published “Ileal-lymphoid-nodular hyperplasia, non-specific colitis, and pervasive developmental disorder in children” in The Lancet in 1998 (351:637–641). On 2 February 2010 the journal retracted it. Quote the notice, then never cite the 1998 paper as evidence for a vaccine–autism link:
The Editors of The Lancet, retraction notice (published online 2 February 2010)
Following the judgment of the UK General Medical Council's Fitness to Practise Panel on Jan 28, 2010, it has become clear that several elements of the 1998 paper by Wakefield et al are incorrect, contrary to the findings of an earlier investigation. In particular, the claims in the original paper that children were “consecutively referred” and that investigations were “approved” by the local ethics committee have been proven to be false. Therefore we fully retract this paper from the published record.
A student who only read a 1999 news recap, or a blog that still cites volume 351, would miss the notice. Before you lean on an extraordinary biomedical claim, search the title in PubMed, open the publisher page, and check Retraction Watch. Cite the retraction if you are writing about the scandal. Do not cite the retracted article as if it still supported the finding. How the peer-review process works explains the filter; this page is the later check.
Worked Example 3: A Video That Lost Its Glass
On 14 October 2022, Just Stop Oil protesters threw tomato soup at Vincent van Gogh’s Sunflowers in London’s National Gallery. A tweet from Guardian journalist Damien Gayle reached tens of millions of views. Caulfield and Wineburg open their Verified excerpt with that clip because the outrage skipped one recoverable fact:
Caulfield and Wineburg, “Learning to Verify,” American Educator (Spring 2025)
All this outrage and concern missed a crucial fact: Sunflowers was behind glass.
Apart from minimal damage to the frame, the painting emerged unscathed. The soup had splashed harmlessly on the painting’s protective case—a fact the protesters knew, many bystanders knew, and the gallery knew.
The pixels were real. The caption “destroyed a van Gogh” was not. That is Caulfield’s trace step in one clip: recover the original context before you treat a viral still as evidence. A genuine 2017 protest photo reused as “yesterday” fails the same test. Reverse-search the still (Google Images or TinEye). For moving images, investigators often use the InVID/WeVerify/vera.ai verification plugin to extract keyframes and check upload dates. Then find the earliest publication you can. Do not cite a social post for a factual claim unless you can also cite the underlying record.
Worked Example 4: A Citation a Chatbot Invented
Students now paste AI summaries the way they once pasted news ledes. William H. Walters and Esther Isabelle Wilder tested that habit in “Fabrication and errors in the bibliographic citations generated by ChatGPT,” Scientific Reports 13, 14045 (2023). They asked GPT-3.5 and GPT-4 for 42 short literature reviews (84 papers, 636 citations) and then searched for each work:
Walters and Wilder, Scientific Reports 13:14045 (2023)
Within this set of documents, 55% of the GPT-3.5 citations but just 18% of the GPT-4 citations are fabricated.
It is important to realize, however, that ChatGPT is fundamentally not an information-processing tool, but a language-processing tool.
A fabricated citation can mix a real surname, a real journal name, and a DOI that belongs to a different paper — or to nothing. The University of Mississippi library’s spring 2023 intake of 46 suspected AI citations, later published by A. P. Watson in The Serials Librarian (DOI 10.1080/0361526X.2024.2433640), found the same pattern: none of the supplied URLs or DOIs matched the cited works. Treat an AI bibliography as a lead list, not as a source list. Resolve every DOI at https://doi.org/. If it does not resolve, or resolves to a different title, drop the sentence. For how to cite an AI chat you actually used, see how to cite AI. This page is the prior question: did the object exist?
How to Fact-Check the Claims Students Actually Paste
Statistical claims
- Find the original table or microdata, not a screenshot of a screenshot.
- Write the denominator in words: percent of what, among whom, during which dates.
- Check whether the comparison is like-with-like (same age group, same definition of “unemployed,” same inflation year).
- Prefer a national statistical office, the OECD, or a methods section you can open over a campaign graphic.
- If two reputable series disagree, report the disagreement and the definitions. Do not average them into a fake third number.
Scientific claims
- Open the paper (or the preprint labelled as a preprint). Do not cite a university press release as if it were the experiment.
- Read the sample size, the design, and the authors’ own limitations — especially “preliminary,” “associational,” or “in mice.”
- Ask whether later work replicated the result, and whether a correction or retraction exists.
- Separate correlation from the causal sentence you are about to write.
- For clinical questions, prefer a systematic review or a methods-reporting guideline over a single dramatic trial.
Quotations and attributions
- Find the speech, book page, hearing transcript, or recording. A quote card is not a source.
- Read the sentences on both sides. Ellipses that hide a negation are a different quote.
- Confirm the speaker, the date, and the venue. Famous names attract invented lines.
- If you cannot find the original after a serious search, drop the quotation. Do not “keep it” with a weaker citation to a blog that also lacks the original.
Images, video, and “I saw it on social media”
Caulfield’s trace step is built for cropped video and mis-captioned photos. Reverse-search the still. Recover the earliest publication you can. A genuine photograph with a false date is still a failed citation. Do not cite a social post for a factual claim unless you can also cite the underlying record. For platform-specific citation formats, use the style guides — this page is about whether the content is usable.
AI summaries and invented references
- Treat every title, author string, and DOI from a chatbot as unverified until it resolves.
- Search the title in PubMed, Crossref, or a library catalog. A near-miss title is often a mash-up, not a typo.
- Open the PDF or HTML. Confirm the sentence you want is on the page you opened.
- If the model cannot show you the object, you do not have a source. Rewrite from a work you actually read.
Tools That Help You Leave the Page
| Tool | Use it to | Do not use it to |
|---|---|---|
| IFCN signatory directory | Find a professional check of a circulating public claim | Replace the primary source in a literature review |
| FactCheck.org / SciCheck | See a documented reconstruction with footnotes | Treat the rating graphic as citable evidence |
| Retraction Watch | See whether a paper was withdrawn or heavily corrected | Assume every criticism equals a retraction |
| PubMed / Crossref / Google Scholar | Find the article, citing literature, and publisher notices | Assume a high citation count means the finding is true |
| Wayback Machine | Recover a page that moved or was quietly edited | Prove the first archived copy is the original publication |
| Reverse image search | Find an earlier upload or a stock-photo original | Prove a photo is “real” in the sense of unstaged |
| WHOIS / domain records | See how new a lookalike news domain is | Identify every legitimate owner (privacy redaction is common) |
Wikipedia is a map, not a destination. Caulfield and Wineburg treat it as a first lateral tab. Use it the way Wikipedia for research describes: read the talk page and the citations, then leave. A NewsGuard score or a media-bias chart is someone else’s opinion of a publisher. It does not verify the cell in the table you are about to copy.
Red Flags That Mean You Must Trace, Not Skim
The claim is trying to rush you
- A headline that cannot survive without an exclamation point
- “Share before they take this down”
- A statistic with no named dataset
- A quote with no speech, book, or timestamp
- A study described only in a press release or an AI chat
The trail is already broken
- Citations that do not support the sentence they sit under
- A DOI that does not resolve, or resolves to a different paper
- Three articles that all cite one unpublished slide deck
- A journal you cannot find in a library catalog or DOAJ
- An author with no institutional footprint beyond the one page
None of those flags proves the claim is false. Each one proves you do not yet have a citable claim. Keep going or cut the sentence. For funding and viewpoint distortion after the fact is established, use bias in sources.
Mistakes That Survive a “I Googled It” Pass
- Stopping at a source you already like. Confirmation bias is why you must fact-check the sentence that helps your thesis as hard as the one that threatens it.
- Counting reprints as independent confirmation. Syndicate the same wire story three times and you still have one reporting chain.
- Treating consensus as a citation. “Everyone knows” is not a scholarly source.
- Discarding a biased outlet’s one checkable fact. A partisan site can still quote a statute correctly. Verify the statute.
- Confusing opinion with a factual claim. “This policy is unwise” is not fact-checkable the way “this policy spent $X” is.
- Citing a fact-check instead of the record when your paper is about the record. The check is a map.
- Trusting a chatbot bibliography. Walters and Wilder showed that even GPT-4 still fabricates roughly one citation in six.
Build Fact-Checking into the Paper, Not the Night Before
- While you read. Mark each sentence you might use as verified, unverified, or contested. Keep the URL or DOI in the same note. How to take research notes is the capture habit; this page is the verification habit.
- While you draft. Do not park a “look up later” statistic in the paragraph. If you cannot verify it in the session, leave a visible hole or delete it.
- Before you submit. Re-open every percentage, quotation, and “recent study.” Confirm the link still resolves and that no retraction appeared after you first read the piece.
- When you cite. Cite the object you actually verified — the table, the paper, the transcript — not the social post or chat that introduced it. Then format that object in the style your assignment named.
Verified the claim? Cite the object you opened
Fact-checking decides whether a sentence may appear. Citation records the version you used. After the claim survives, generate the reference for the paper, dataset, or report you actually opened — not for the recap that first tempted you.
Frequently Asked Questions
- How is fact-checking different from evaluating whether a source is credible?
- Credibility is a judgment about the publisher or author for a job. Fact-checking is a judgment about one sentence. A CDC table can be a credible source for mortality and still be the wrong table for your year. A famous newspaper can be a credible publisher and still invert a denominator. Run both checks. Start with credible sources for the publisher, then come back here for the claim.
- Is SIFT just another CRAAP checklist?
- No. Caulfield designed SIFT as a short list of moves that take you off the page. CRAAP is a list of questions you ask about the item. Verified (2023) restates the same habit: gather context before you argue. Use CRAAP on how to evaluate sources after you already know the publisher is real. Use SIFT first when a claim arrives from an unfamiliar URL, a group chat, or an AI summary.
- Can I cite FactCheck.org or Snopes in a research paper?
- Yes, when the paper is about the rumor, the public debate, or how a claim travelled. When the paper is about the underlying phenomenon, follow the fact-check’s footnotes to the dataset, statute, or journal article and cite that primary source. Prefer an IFCN signatory, and open its sources rather than stopping at the rating.
- What if experts genuinely disagree?
- Report the disagreement instead of laundering it into a fake consensus. Name the camps, the evidence each one treats as decisive, and whether a methods-reporting guideline or a living review exists. “Studies are mixed” is honest only if you have read more than one study and can say how they mix.
- What if I cannot verify a claim after a serious search?
- Do not use it. FactCheck.org’s rule is the academic rule: the person making the claim owes the evidence. A hole in the paragraph is better than a footnote that points at a screenshot. Find a different, checkable supporting fact, or narrow the thesis.
- Do I need to fact-check peer-reviewed journals?
- You do not re-run every experiment. You do confirm that the article exists, that it says what you are about to say it says, that you have not upgraded correlation into causation, and that it has not been corrected or retracted. Extraordinary claims and single small trials get more of your time than a routine methods citation.
- How do I fact-check a historical date or quotation?
- Find a primary source or a scholarly edition. University press books, documentary editions, and archive catalogs beat quote aggregators. If two serious historians disagree about interpretation, that is a debate to cite, not a reason to pick the date that fits your outline. Primary sources in history covers the archival half.
- How should I handle images and viral video in a paper?
- Reverse-search the still, recover the earliest publication you can, and cite that publication — with a date — rather than the most recent share. The 2022 Sunflowers soup video is the worked case: the clip was real and the “destroyed painting” caption was not. If you cannot establish when and where the media was made, it is an illustration of a rumor, not evidence of the event.
- How do I fact-check a citation a chatbot gave me?
- Resolve the DOI or ISBN. Search the exact title. Open the work. Walters and Wilder (2023) found 55% of GPT-3.5 citations and 18% of GPT-4 citations in their set were fabricated; many real-looking leftovers still had wrong pages or years. If the object does not exist, delete the sentence. See how to cite AI only after you have an object you actually used.
- Does Wikipedia count as fact-checking?
- It counts as a start. Use the article to collect names, dates, and citations, then open those citations. Do not stop on the Wikipedia page for a contested statistic. Wikipedia for research is the longer rule.