Media literacy, oriented

A post 50,000 people have seen may have passed no check at all, so reach says nothing about review. Every publication also has a payer, such as advertisers, a sponsor or an owner, who can shape what runs. That does not make a claim false; it marks where to check harder.

Three sorting moves come before any checking. First, split fact from opinion: a factual statement is one evidence could settle, and it stays one even when it is wrong, while an opinion rests on values. A 2019 US survey found nearly half its respondents did no better than chance at telling the two apart. Second, ask what the piece is for: informing you, selling to you or pushing a view. Third, find the label. Federal Trade Commission (FTC) staff say readers likely understand "Ad" but call "Promoted" ambiguous. The FTC treats native advertising, an ad dressed to match the stories around it, as deceptive when its format hides that it is an ad, even if each product claim is accurate. A creator paid in money or free products should disclose that tie in the post itself.

The trap is reading the content before the frame. Say a home site runs "Six blenders worth every penny," every link goes to one retailer, and a faint "Promoted" tag sits above the headline. The advice may be sound, but one company gains from every pick.

Rule: before you weigh a claim, name the kind of content, what it wants from you and who paid for it.

Lateral reading means opening fresh tabs to research a site instead of studying the site itself. When researchers watched professional fact-checkers, the checkers abandoned unfamiliar sites almost at once to learn who ran them, and reached better credibility verdicts than students or history professors. The page's own cues, the ones a checklist keeps you staring at, are cheap to fake: links to studies, an About page, polished design, a .org ending. An About page is the owners' own account, and impostor sites can mimic real outlets under almost-right addresses.

Some operations are built to pass the on-page test. Pink slime sites dress as local newspapers while running partisan material or corporate public relations, often from networks that churn out automated stories. A front group presents itself as ordinary citizens or impartial experts while the money comes from a backer with something to gain. Three unfamiliar sites carrying one article word for word are one source, just as ten links leading back to one post are.

Let funding set how hard you check, never the verdict; dismissing a claim only for who pays for it is the circumstantial ad hominem fallacy.

The trap is studying the page. Say a search lands you on "Valley Water Facts," with clean design, a mission statement about independence and a list of citations. Searching the name plus the word funding might show that a filter company pays for it.

Rule: when a source is unfamiliar, leave it and search its name before you read its claims.

Posts seldom arrive as clean claims. "The new parking rules are killing downtown" bundles a judgment with something checkable, such as whether more shops closed after the change. Write that part as one sentence and mark its strength: saying two things are linked is weaker than saying one causes the other, since both can climb because a third factor pushes them. One vivid story, however moving, tells you nothing about a rate.

Then test the reasoning. The claimant carries the burden of proof, and calling a claim true because nobody has disproved it is the appeal to ignorance. The more surprising a claim, the more evidence it needs. A claim is falsifiable when some possible observation would prove it wrong and you can say which; a claim that fits every outcome gives you nothing to test.

Conspiracy theories often lean on these same failures. Researchers describe five elements: a pattern tying events together, intent, a group of plotters, a threat and secrecy. Secrecy lets missing evidence be read as a cover-up, which is why such theories have been called self-sealing, closed to counterevidence. Real conspiracies do happen, and direct evidence such as records or confessions is what establishes them.

The trap is accepting a question's frame. Someone keeps asking why the water board buried its inspection report. Turn the question into the claim it smuggles in, "the board buried the report," and ask for evidence of that.

Rule: write the claim as one sentence, ask what would show it false, and ask the claimant for the evidence.

A strong feeling is your signal to slow down: in experiments, people nudged to trust their emotions believed more false headlines, though not more true ones. Trace the claim to where and when it first appeared, since old material often returns as new. Read beyond the headline; in one study of a medical journal's news section, 45 percent of headlines overstated the research. Read the sentences around any quote, too, because a cut can flip its meaning.

Searching has its own trap. In five experiments, people asked to search before judging false stories rated them true more often than people who did not search, mostly when low-quality sources filled their results. Pasting in a story's headline turned up unreliable pages far more often, and in a close look at one false article, so did adding one charged word from it. Describe the claim in your own neutral words, scan results before clicking, and watch for the small Ad label, since top spots can be bought.

A fact-check verdict covers the claim in the checker's wording, on the date of the check, and each outlet has its own rating scale. Because checkers select suspicious or potentially harmful claims, tallying their verdicts says nothing about how often anyone errs.

The trap is stopping at the share. Say a post reads "City ends recycling program." The article beneath it says one district paused pickup for two weeks.

Rule: pause, trace, read the whole thing and look for better coverage, spending effort in proportion to what is at stake.

In a study of Indian public WhatsApp groups devoted to politics, old photos stripped of context were the largest single category of misleading images, ahead of memes and doctored pictures. Such a fake needs no editing and its pixels are genuine, so test the caption's claim instead. A crop misleads the same way: a close shot of people yelling looks like a brawl until the full frame shows a crowd celebrating a goal.

A reverse image search uses a picture as the query and returns earlier copies, which can expose a caption calling an old photo new. No match proves nothing, since the original may never have gone online or been indexed. Shadows, maps and weather archives can also test where and when a scene was recorded.

In experiments, people were poor at spotting edits by eye and often could not locate them. In one set of experiments, photos generated by artificial intelligence (AI) could not be told from real ones, and a voice can be cloned from a short clip to fake a relative's emergency call. Detectors can miss newer generators, and compression can erase the traces they seek, so a "real" or "fake" score is weak evidence.

The trap is the pixel hunt. Say a photo of a toppled water tower circulates after this week's tornado, and you zoom in hunting for blurred edges. The image is clean, and it is from a storm years ago.

Rule: work outward from the file, to the earliest copy, the original poster and the date, before you study the pixels.

A count needs a denominator: 300 bike thefts is 6,000 per 100,000 residents in a town of 5,000, but 15 per 100,000 in a city of two million. A rate going from 10 percent to 12 percent rises 2 percentage points, which is also a 20 percent increase; both are true and answer different questions. "Doubles your risk" can mean 1 in 10,000 becoming 2 in 10,000, an absolute rise of 1 in 10,000. If nine households each make 50,000 dollars and a tenth makes 5 million, the mean is 545,000 dollars and the median 50,000. A poll at 48 percent against 46 percent, with a 3-point margin of error, cannot say which side is ahead.

Charts add levers. Start a bar chart's axis at 95, and a value of 98 gets a bar three times as tall as a value of 96. A series' start date can invent a trend or hide one. Grouping can flip a pattern: a courier can be more punctual than a rival on both short and long routes yet trail it overall, if the harder long routes make up most of its work. That reversal is Simpson's paradox.

The trap is base-rate neglect, ignoring how big each group was. Say a city has 1,000 cyclists, 900 of them in helmets; riders without one have a 10 percent injury risk, and helmets cut it to 5 percent. Then 45 helmeted riders get hurt and 10 bare-headed ones, so most of the injured wore helmets, and a post calling helmets useless has misread the numbers.

Rule: before you compare, ask out of how many, starting where, and grouped how.

The three terms sort content by truth and motive. Misinformation is false content passed on in error. Disinformation is false or misleading content created to deceive. Malinformation is genuine material, such as a private email, pulled out of context to mislead or harm. The same false post is misinformation when your cousin shares it and disinformation when its creator does, and posts rarely reveal motive.

Influence operations reuse a few mechanisms. Repetition makes a claim feel truer; across 182 studies, a review measured this illusory truth effect as small but reliable. Flooding, or firehosing, pours out so much content that other voices drown. Astroturfing uses networks of fake accounts to stage a grassroots surge. Campaigns hand narratives to local voices, often unwitting, so the origin vanishes, and they widen divisions a society already has.

The early fear that correcting a myth makes it stick has mostly not held up in later experiments, though in one election study beliefs had drifted back by the next survey wave, so a correction may need repeating. Prebunking, learning how a manipulation technique works before you encounter it, has real but limited measured benefits.

The trap is "it's just bots." Once a planted story catches on, many people spreading it are real and sincere. Say a neighbor reposts a claim that started in a fake-account network: she is not an agent, and the claim is still false.

Rule: judge whether the content is accurate first, and treat where a story began as a separate question from whether the person sharing it is honest.

Persuasion techniques work by moving you faster than you can check. Urgency, as in "offer ends at midnight," works because checking takes time, and federal consumer guidance names pressure to act right away as a scam warning sign. Loaded words tuck a verdict inside a description: "the board slashed library funding" and "the board reduced the library budget" describe one decision. Presenting the same options as gains or as losses can change what many people pick. Equal time for two sides can suggest the evidence is evenly divided when it is not. Likes, reviews and follower counts track popularity, not accuracy, and can be bought or faked.

Your own thinking supplies the rest. Confirmation bias gives claims you agree with a lighter check. Motivated reasoning starts from the conclusion you wanted and then gathers reasons, though researchers disagree about how it works. The hostile media effect makes even-handed coverage look tilted against your side. Still, skepticism is not cynicism. Pooling 67 experimental papers with participants from 40 countries, a meta-analysis found people were good at telling true news from false, and their more common error was dismissing true news, not believing false news.

The trap is checking only what you dislike. Say a post claims a rival town's office wasted a grant, and you share it in seconds, while a similar post about your own town gets a full search.

Rule: before you weigh evidence, name the answer you are hoping for, then run the same check whichever way the claim points.

In an analysis of UK university press releases about health studies, a third of those on correlational work implied more cause and effect than the paper showed, and news built on such releases mostly repeated the overreach. So find the study and ask what kind it is. A case report describes one patient. A randomized controlled trial assigns people to groups by chance, the design that best shows whether a treatment works, and a systematic review weighs many studies together. Ask who took part: results in animals may not carry over to people, and 30 young men in a trial may tell you little about older women.

Publication is a check, not a guarantee. A preprint is shared before peer review, and in peer review a few experts judge whether a paper seems sound without repeating the experiments; reviewed papers still get retracted sometimes. In an abstract, find the main outcome before the conclusion, which is the authors' interpretation and can carry spin. Funding seldom survives the trip to the news: in one analysis, 94 percent of journal articles reported it, against 9 percent of the news stories about them. Health-fraud warning signs include one product for many unrelated diseases, quick cures and words like miracle or secret.

The trap is trusting the newest study. Say a headline reports one small trial found a supplement eased headaches, while a systematic review of earlier trials found no clear effect. The review usually deserves more weight.

Rule: check the design, participants and funding in the paper itself, then weigh it against the earlier research.

Verification means checking a claim before reporting it, and it leaves visible traces. A careful report says where each fact came from, whether a document, a named official, an unnamed source or a news agency. A source may need anonymity to avoid retaliation, but then you rely on the outlet's judgment of that person, so give the claim less weight. If a story levels a serious accusation, look for the accused person's answer or a line saying they did not respond. Editorials and opinion columns argue a view without owing you a news report's balance, so judge their facts and reasoning.

Accountability shows after publication. A findable record of labeled corrections is better evidence than a site with no admitted errors. Updates are normal; a change with no note is the warning sign, and an archived copy lets you compare. In US law, defamation is a false statement of fact that harms reputation, and substantial truth is ordinarily a defense. A public official suing for defamation must also prove actual malice, meaning knowing falsity or reckless disregard of whether the statement was false. Under the First Amendment, the Supreme Court presumes heavily against prior restraint, government stopping a story before it runs; remedies such as a defamation suit generally come after.

The trap is reading a missing corrections page as a perfect record. Say one site has posted no correction in ten years, while another lists dozens of labeled fixes. Only the second lets you see how it handles mistakes.

Rule: trust reports that show where each fact came from, and outlets that show how they fix mistakes.

In the US, the states and over 10,000 local jurisdictions run elections, and rules differ between places. The Election Assistance Commission (EAC) directs voters to the local elections office for practical details, so check any deadline from a post on that office's own site. Some claims hinge on procedure: the EAC explains that election-night results are unofficial and that counting of mail, provisional and military ballots may continue, so totals can shift before results are certified. For federal races, ads and mass mailings from political committees must carry a disclaimer saying who paid. Separate a claim's checkable facts from its values, and check claims from your own side as hard as the other side's.

Records let you check claims directly. A primary source was created at the time, such as a court ruling, a budget or a meeting's minutes; a later account of it is secondary. Before reading a document for content, establish its author, intended reader, date and place. A gap in the record is not proof that nothing happened. The Freedom of Information Act (FOIA) lets almost anyone request existing records from federal agencies, though nine exemptions can keep some withheld.

The trap is trusting the summary. Say a post claims the school board voted to end late buses. The minutes show a motion to study the routes, tabled until spring.

Rule: for voting details, go to the election office; for any record, read the record itself and cite it precisely.

A ranked feed orders posts by the engagement software predicts they will draw, not by time, so absent from your feed is not absent from the world. In one audit of 806 users, engagement ranking amplified angry posts attacking the other political side, political posts those users did not prefer. Evidence that algorithms seal people inside filter bubbles is thinner and more mixed than popular accounts claim. AI chatbots can state errors fluently, so open the source an answer cites and check it says what the answer reports.

What spreads is not sorted by truth. In an analysis of millions of posts, negative news links were shared nearly twice as often. A verification badge confirms who runs an account, not whether its posts are accurate, yet in two experiments, even after badges were explained, people tended to read one as a sign of credibility. A screenshot travels without its link, date or author, and private groups pass claims along out of outside checkers' sight.

Your part is the share. In one online experiment, people asked first to explain how they could tell if a political headline was accurate reported less willingness to share false ones, with no drop for true ones. The effect was modest, and it covered stated intentions, not real posts. When you post, link the original, add the date and flag what is unconfirmed.

The trap is the quiet delete. Say you shared a wrong closure notice to a neighborhood group, then removed it. People who saw it never learn it was wrong.

Rule: before you share, ask how you know; after a mistake, make the correction as visible as the share.