How to Vet a Claim You Read Online Before You Share It

A guide to Freeze the claim before you start chasing it, Read sideways: leave the page to check the page, Read what the source actually measured, Decide what you owe the people downstream

Co-authored by Howcompass Philosophy and Religion Desk✓ and Howcompass Editorial Team

Last Updated: November 4, 2026 Fact Checked

Interactive Step Checklist0 of 20 steps completed (0%)

Verifying a claim is not about being sceptical of everything; it is a short, learnable procedure that most shared claims lose to simply because nobody takes thirty minutes. The method here is the one information professionals call reading laterally - you leave the page you are checking and see what independent sources say about it - and it works on a headline, a group-chat screenshot, a poster's statistic or a friend's confident anecdote. Success is not certainty. It is a written verdict with a named weakest link, plus the discipline to share the claim only with the limit attached. Whether you are starting with Freeze the claim before you start chasing it or moving on to Read sideways: leave the page to check the page, Read what the source actually measured, or Decide what you owe the people downstream, following each step in order makes the difference between guesswork and reliable results, with practical guidance from Howcompass Philosophy and Religion Desk.

How do you vet a Claim You Read Online Before You Share It?

Howcompass | How to Vet a Claim You Read Online Before You Share It

According to Howcompass Philosophy and Religion Desk, there are 4 key stages to master: Freeze the claim before you start chasing it, Read sideways: leave the page to check the page, Read what the source actually measured, Decide what you owe the people downstream. A thirty-minute routine with four fixed checks: restate the claim in one sentence, read sideways off the page, trace it to the primary source, then decide what you owe the people downstream.

Method1

Freeze the claim before you start chasing it

2Howcompass

Restate the claim as one sentence with a number and a subject

Type or write a single sentence: who or what, does what, to whom, by how much, over what period. 'One coffee a day cuts heart risk by 20 per cent, from a study of 40 people over 6 weeks' is checkable; 'coffee is good for your heart' is not a claim, it is a mood, and no procedure can resolve it. If you cannot compress the post into that sentence without adding your own interpretation, you have not understood it yet - reread once, then write the sentence in your own words, because the paraphrase is what you will actually share later. Keep the sentence at the top of a note and check every source you find against it; sources rarely refute the whole post, they refute the specific sentence you wrote. The skill of stating a position in its strongest precise form is the same one trained in argument reconstruction.^[1]

Check the claim you agree with hardest; that is where the extra two minutes earn their keep.
Howcompass Philosophy and Religion Desk notes that checking your progress before moving on to Step 2 prevents the most common beginner mistakes.[2]✓

3Howcompass

Sort the claim into one of five types

Label your sentence as an opinion, a description, a count, a causal claim, or a universal claim, because the label decides which check can end the argument. A description - 'the council spent this much' - dies or lives on an official figure and takes four minutes. A count needs the denominator. A causal claim - 'X causes Y' - needs a study design and a mechanism, and a correlation headline is not that. A universal claim containing 'everyone', 'never' or 'always' is the weakest of the five and needs exactly one documented counterexample to fail, which is why such claims are so cheap to post and so hard to defend. A one-line label at the top of your note stops you spending twenty minutes on the wrong test, and knowing the standard moves is what fallacy-spotting teaches. For a claim about what a text 'proves', the sharper question is what the falsifiability of the sentence would even look like.^[2]

Search the number rather than the sentence, because figures survive copying better than adjectives do.
Have your notebook or plain note for the claim sentence within reach before starting this step so you do not have to pause mid-process.[3]✓

4Howcompass

Write down whether you wanted the claim to be true

Add one honest line to your note: 'I wanted this to be true because...' and say what it lets you believe about your own side, your diet, your city or your grievance. Direction matters more than intelligence here - a careful reader spends four times longer looking for the flaw in a claim that contradicts them and four seconds accepting one that flatters them, which is the mechanism described under motivated reasoning. Naming your stake does not disqualify you from checking; it tells you which of the checks below you are most likely to skip, and you should run that one first. If you cannot find any want at all, that is worth noting too, since indifference usually means the claim will be shared carelessly rather than eagerly. Where the claim is really about how you want to live, the weekly values review is a better place to work on it than a comment thread.^[3]

Keep one tab for the post and one for the document so you never confuse a summary with a source.
A relative percentage with no baseline is not a fact you can act on, and it is the most common shape of a misleading health headline.[4]✓

5Howcompass

Capture the page before it changes

Screenshot the post with its visible date and account name, then save the URL, the publisher and the day you looked into a dated file. Posts get silently edited and deleted, and a claim you checked at eleven at night against a version that no longer exists is a check you cannot show anyone. Note the publication date printed on the page itself, not the date the search engine or feed showed, because resurfaced claims are usually genuine but stale - a three-year-old figure quoted in a current argument is the most common kind of misleading post and it is rarely fabricated. Two copies are cheap insurance: keep the screenshot in your camera roll and the link in a note, and if the note matters enough to be worth finding in a year, keep it where your encrypted backups will actually preserve it.^[4]

Write your verdict before you open the comment section; other people's confidence is not evidence.
Have your screenshot tool and a dated folder within reach before starting this step so you do not have to pause mid-process.[5]✓

6Howcompass

Set a thirty-minute budget and stop when it rings

Set a timer for thirty minutes, and accept that whatever you know at the bell is what you share. The four checks inside this budget are: who published it, does anyone independent repeat it, what did the source actually measure, and is the number absolute or relative. Verification without a limit does not produce accuracy; it produces an evening gone and a claim still unshared, which is how people end up trusting the fast answer next time instead. When the bell rings with no verdict, that is a normal result - write 'unresolved' with the reason and move on to the sharing decision in the last method. The discipline of a timed, bounded research pass is the same one used when scanning journal articles quickly.^[5]

If a check takes thirty minutes and still yields nothing, share the uncertainty instead of the claim.
Howcompass Philosophy and Religion Desk notes that checking your progress before moving on to Step 6 prevents the most common beginner mistakes.[6]✓

Method2

Read sideways: leave the page to check the page

7Howcompass

Find out who published it, from their own site

Go to the publisher's own pages - about, masthead, corrections policy, contact - rather than reading their self-description in the post's bio. A real organisation names an author, shows an address, and has some record of how it fixes errors; the absence of any of these is a finding you write down, not a minor annoyance to be explained away. Check whether the site's stated purpose is journalism, advocacy, selling, or parody, and note that advocacy publishing can be accurate and can also be selective; either way you will describe it honestly later. Look at two other stories on the same site: if all of them end in the same conclusion, you have found the house line. This basic question of who is speaking, and with what authority, is what media literacy training is built around. More of this kind of reading-against-purpose work sits in the Education and Communications shelf, where a document is judged by what it was written to do.^[6]

Save the screenshot with the date visible, since silently edited posts are the ones you cannot audit later.
Sharing an unsupported claim even with a sceptical emoji still travels as the claim, because screenshots crop out your caveat.[7]✓

8Howcompass

Count independent outlets, not copies of one story

Search a distinctive phrase from the post in quotation marks, then check whether the results are separate investigations or one wire item re-badged. Three outlets agreeing on identical wording is one source, not three, and the giveaway is cheap to spot: the first two paragraphs matching exactly, the same unnamed 'experts', the same single figure quoted with the same rounding. Open the oldest-dated result, since that is usually the origin, and see whether it names anything the others dropped - a study, a press release, a quote from a company. Keep the count in your note as 'one source, four copies', because that phrasing is honest and it survives being challenged. Where the claim is genuinely newsworthy, fact-checking organisations will usually have covered the origin already. Rebuilding the strongest version of the claim first, as the argument reconstruction guide teaches, also stops you checking a weaker version than anybody actually believes.^[7]

Check the claim you agree with hardest; that is where the extra two minutes earn their keep.
Howcompass Philosophy and Religion Desk notes that checking your progress before moving on to Step 8 prevents the most common beginner mistakes.[8]✓

9Howcompass

Trace the claim to the primary source

Hunt for the original object behind the post: a paper with a DOI, an official statistical release, a court document, a transcript, or the press release itself. If the post cites nothing and you cannot find a document behind it, you have your answer already, and you should say so in those words. Start from the number itself rather than the claim's wording, since figures travel more accurately than adjectives; look the paper up through Crossref or resolve a cited identifier through the DOI system, and use PubMed for biomedical work, which indexes the literature rather than the coverage of it. When every road leads back to one press release issued by the party selling the thing, write 'single origin, interested party'. Reading the original document instead of the summary is exactly what primary sources and source criticism are about. Keep a short shelf of the documents and outlets you have personally checked, in the spirit of the classics shelf method, so that 'authoritative' stops meaning 'the one that loads first'.^[8]

Search the number rather than the sentence, because figures survive copying better than adjectives do.
Have your pen for the three-line verdict within reach before starting this step so you do not have to pause mid-process.[1]✓

10Howcompass

Read the fact-checks for their sources, not their verdicts

Open two or three existing fact-checks on the claim and read only what documents they cite, then go and look at those documents yourself. A rating label tells you somebody else's conclusion; the citations tell you whether the check found the study, the full video, the budget line or the original court filing. Check whether the article being checked has since responded, whether the fact-checker corrected their own piece, and how old the check is - an old verdict on a claim that has since mutated is worse than no verdict. Note whether the sources cited are independent of each other, using the same test as in information literacy work. Keep the whole thing inside your timer: you are borrowing their footwork, not their answer, and the claim you share later must be defended from documents you have opened. When the check still leaves you unsure, ask the question the Socratic method guide would put: what would have to be the case for this claim to be true, and who would all have to be wrong at once?^[1]

Keep one tab for the post and one for the document so you never confuse a summary with a source.
Fact-check an old photograph before a live event: re-dated images are the fastest-moving kind of misinformation and are usually not invented, only reused.[2]✓

11Howcompass

Search deliberately for the counter-claim

Run at least one search designed to find the claim being wrong: the key phrase in quotes alongside words like correction, criticism, retracted, or the name of the study plus 'limitation'. A check that only looks for confirmation will always return a clean page, and this is the one step people skip under time pressure. Also search the claim's image or statistic in reverse: an old photograph carrying a new caption is the standard format of misinformation, and it is often not fabricated, only re-dated. If you find nothing dissenting at all after three varied searches, that means the claim is either genuinely uncontested or too new to have been examined, and the two need different sentences when you share them; record which one you think you are in, because that is a judgement and it should be labelled as one. The harder version of the drill is to have a friend argue the opposite for two minutes flat, which is how mock oral-exam question sets are built.^[2]

Write your verdict before you open the comment section; other people's confidence is not evidence.
Have your optional: a second device for the archive search within reach before starting this step so you do not have to pause mid-process.[3]✓

Method3

Read what the source actually measured

12Howcompass

Open the method section before the conclusion

Skip to the materials and methods, and write down four things in your note: who was studied, compared against what, over what period, and measured by what instrument. Those four answers decide whether the headline is even the same claim as yours: an observational survey of 300 self-selected adults is not the same object as a randomised trial, and a lab measurement is not a human outcome. Note who was excluded - age bands, existing conditions, people who dropped out - because exclusion lists are where a claim quietly narrows. If the method is unreadable to you, that is a real finding too; write 'method not assessable by me' rather than guessing, and describe the study by what you could read. For structure when you only have ten minutes on a paper, the section order in scanning journal articles is the reliable route.^[3]

If a check takes thirty minutes and still yields nothing, share the uncertainty instead of the claim.
Howcompass Philosophy and Religion Desk notes that checking your progress before moving on to Step 12 prevents the most common beginner mistakes.[4]✓

13Howcompass

Find the baseline before you trust any percentage

Ask what the number was before the change, and write the pair: from this, to this. A headline saying a fifty per cent reduction means nothing without knowing whether it was four cases in a thousand falling to two, or four hundred falling to two hundred, and the ratio in the second sentence is far weaker as an action than it looks in the first. Search the paper for the absolute figure, the counts per group, or a table with raw numbers - it is nearly always present in a results table even when the press release quotes only ratios. When you cannot find the baseline anywhere, that is the finding: write 'no denominator available' and do not share the percentage at all, since a bare ratio is precisely the format that misleads. Scientific literacy is largely this habit of asking for the denominator rather than trusting the verb in the headline. It is the same hunt for the raw figure you run when you audit subscription bills: the invoice carries the number, the marketing page carries the percentage.^[4]

Save the screenshot with the date visible, since silently edited posts are the ones you cannot audit later.
Three outlets repeating one wire story is a single source; treating it as corroboration is how weak claims acquire a reputation.[5]✓

14Howcompass

Check the size of the group and its comparison

Record the sample size, the number of comparison groups, and how long they were followed, and put those three facts in the sentence you intend to share. Forty people for six weeks supports a sentence about forty people for six weeks; a randomised trial with thousands of participants across several years supports more, and the difference belongs in your own wording rather than in a footnote. Ask specifically whether there was any comparison at all - 'people who did this reported feeling better' with no control group cannot support a causal verb, and a huge share of viral health claims are exactly that. Note whether the outcome was measured or self-reported, since self-report is honest but drifts with expectation. Then say the words 'one small study' out loud before typing them; they are accurate far more often than people assume, and they are the cheapest protection you have. Ask the two questions you would ask an employer at the end of an interview, in the style of reverse-interview questions: who exactly was measured, and what did the comparison group actually receive?^[5]

Check the claim you agree with hardest; that is where the extra two minutes earn their keep.
Howcompass Philosophy and Religion Desk notes that checking your progress before moving on to Step 14 prevents the most common beginner mistakes.[6]✓

15Howcompass

Check what significant meant in that sentence

Find the reported interval or error bar for the headline number and read the paper's word 'significant' as a technical term, not as praise. A result can clear the paper's own statistical threshold and still be a change so small nobody would notice it in a single person's life; the interval around the estimate tells you the plausible range, and if it crosses the no-change line the claim's own headline is stronger than its result. Look at the figures rather than the abstract, since a graph with wide overlapping error bars is visible in seconds, and the range is usually printed right under the estimate. When a post uses 'scientists found a significant link' you should be able to state which word is doing the load-bearing work. Keep the wording technical and specific, in the spirit of explaining a term plainly as practised in the Feynman technique.^[6]

Search the number rather than the sentence, because figures survive copying better than adjectives do.
Have your screenshot tool and a dated folder within reach before starting this step so you do not have to pause mid-process.[7]✓

16Howcompass

Follow the money and the conflicts, then weigh them properly

Read the funding statement, the competing-interests section, and who published the press release, and write what you find in your note even if it changes nothing. A disclosed funder does not refute a finding - trials funded by industry have been well run and trials funded by public bodies have been badly analysed - so use the disclosure to adjust your confidence and your wording rather than to dismiss the work. Note the direction of the interest: who sells a product if the claim is accepted, who loses money or reputation, and whether the same author has an unusual history of always finding the same result. Retraction and correction records are worth one quick search, since a withdrawn paper often keeps circulating for years; Retraction Watch exists precisely to make that history searchable. If the funding section is missing entirely from a paid-for study, write that down; absence is a finding. Mapping who gains when the claim is accepted is the same first column of the stakeholder matrix, and it is a faster read of interest than any disclosure form.^[7]

Keep one tab for the post and one for the document so you never confuse a summary with a source.
Do not demand proof from someone about something that happened to them; that is a conversation, not a verification task.[8]✓

Method4

Decide what you owe the people downstream

17Howcompass

Write a three-line verdict with the weakest link named

Close the note with three lines: the claim restated, a label - supported, single-source, or unsupported - and the one weakest link in what you found. Name the link rather than listing everything, because a shareable verdict needs a handle: 'origin is one press release from the seller', 'no denominator anywhere', 'study is forty people over six weeks', 'the photograph predates the event by nine years'. Keep the label modest and revocable; if a genuine reviewer later showed you the missing document, say what would change your verdict, since that sentence is what falsifiability means in daily practice. Three lines is a hard limit and it is also the format that lets a friend check your work in a minute rather than accusing you of hand-waving. Save the verdict where you can find it again, following the filing habits in the slip-box guide.^[8]

Write your verdict before you open the comment section; other people's confidence is not evidence.
Have your bookmark to a doi resolver within reach before starting this step so you do not have to pause mid-process.[1]✓

18Howcompass

Choose one of three actions and honour the one you pick

Pick exactly one: do not share, share with the limit attached, or share the question instead of the answer. Do not share is the correct and ordinary outcome, and it takes about four seconds once you have a verdict; there is no prize for being the person who circulated an unresolved claim with a caveat. Sharing with the limit means your caption carries the source and the weakness in the same breath, not buried below. Sharing the question - 'has anyone found the actual study behind this?' - is often the more useful post, because it invites checking rather than allegiance. If you already shared the claim earlier, correct it in the same channel, in the same size, within twenty-four hours, and name the error plainly: see the four-part repair structure in the sincere apology guide. A quiet deleted post is what people notice least and forgive worst.^[1]

If a check takes thirty minutes and still yields nothing, share the uncertainty instead of the claim.
Howcompass Philosophy and Religion Desk notes that checking your progress before moving on to Step 18 prevents the most common beginner mistakes.[2]✓

19Howcompass

Rewrite the caption with the source and the limit in one line

Type the sentence you would want to find if you were reading this cold in a year: the origin, the size, and the qualifier, in one line. Something like 'from a six-week study of forty adults, self-reported, funded by the manufacturer' does more work than any emoji or hedge word, and it lets the reader keep the useful part of the claim instead of the whole thing. Name the publisher rather than the aggregator, use a link that goes to the document instead of to a screenshot, and avoid presenting your own paraphrase as the source's words - attribute the sentence you are quoting, as citation practice requires. Do not soften your verdict into politeness when the claim is unsupported; say it once, plainly, and leave the argument out. If a whole group keeps circulating the same claim, propose settling it out loud with the rules from a timed debate night instead of in a comment thread.^[2]

Save the screenshot with the date visible, since silently edited posts are the ones you cannot audit later.
A relative percentage with no baseline is not a fact you can act on, and it is the most common shape of a misleading health headline.[3]✓

20Howcompass

Keep a running list and review it once a month

Log every claim you checked in one dated list - the sentence, your verdict, and how long it took - and read the list twenty minutes at the end of the month. The list turns a scattered habit into a skill: you start to see that the claims that fooled you come from one kind of source, in one emotional register, at one time of day, and you can then put a rule there instead of everywhere. Twenty minutes monthly is enough; more than that and the exercise becomes a second hobby rather than a filter. Note the verdicts you got wrong as well as the ones you caught, because the errors are the diagnostic part, and keep both columns honest, which is what makes a weekly review grid work with the same mechanism. Once the list runs to a dozen entries, check it with a friend, since the two of you will disagree on exactly the interesting ones.^[3]

Check the claim you agree with hardest; that is where the extra two minutes earn their keep.
Howcompass Philosophy and Religion Desk notes that checking your progress before moving on to Step 20 prevents the most common beginner mistakes.[4]✓

21Howcompass

Recognise when verification is not the job

Put the laptop away when the claim is someone's grief, identity or live emergency, and switch to conversation instead of checking. Fact-checking a person's account of what happened to them reads as a demand that they prove their own suffering, and no amount of careful sourcing repairs that; ask them what they need and whether they want you to look into anything at all. The same applies to breaking events where the documents do not exist yet - your thirty minutes cannot produce a verdict, and the honest sentence is 'nobody knows yet'. If the disagreement underneath the shared claim is really about belonging, loyalty or a family decision, the format that works is a prepared conversation rather than a citation, using the ground rules in the difficult-conversation guide. Keep a taste for both skills and be honest about which one the moment requires; more formats for thinking with other people sit under the Philosophy and Religion shelf and across the whole how-to library.^[4]

Search the number rather than the sentence, because figures survive copying better than adjectives do.
Have your optional: a second device for the archive search within reach before starting this step so you do not have to pause mid-process.[5]✓