Everything the adaptive question bank can teach and test in Media Literacy, from foundations through advanced practice. Work through it in order, or start practicing and let the questions find your level.
The same feed mixes posts from newsrooms, government agencies, companies, independent creators and accounts you cannot identify. Researchers describe them earning trust in different ways: institutions show degrees and brand names, while creators lean on feeling relatable. A cooking video, a health agency page and a brand's sponsored post can sit side by side. Before weighing any item, name what kind of publisher made it.
A gatekeeper decides what gets published. Through the 20th century that meant editors at newspapers and broadcasters, and newsrooms checked facts before a story ran. Online, anyone can reach a huge audience with no editor in between, so a post seen by 50,000 people may have been checked by no one. Never read a post's reach as a sign that someone reviewed its claim.
Someone pays for every publication, and the payer can shape the output. Researchers describe ad-funded news sites leaning on cheap, click-driven content; a health site owned by a supplement seller may favor supplements; a sponsored article serves its sponsor. None of that makes a given claim false, but it tells you where to look harder. Ask first: who pays for this, and what would they want me to believe?
Engagement means likes, shares, replies and time spent, and many feeds rank posts by how much of it they predict. One audit found a major platform's engagement ranking pushed more emotionally charged, hostile political posts than a time-ordered feed, posts users did not prefer. In a news game, higher engagement counts made players share dubious articles more and check them less. A big number shows popularity, not accuracy.
Many outlets republish or summarize stories someone else reported. A survey of misinformation research describes how a false story can spread this way: a small site picks up a social post without checking, others repeat it with 'reportedly,' each pointing to the last, until the origin is lost. Ten links that all trace back to one post are one source. Count independent sources, not copies.
Informing content tries to leave you with an accurate picture; influence content tries to get you to believe, buy or share. Persuasion is normal in ads and advocacy but needs different scrutiny. The federal cybersecurity agency describes disinformation tactics meant to make messages seem credible and stir people to act. A post titled 'What the research says' that ends in a discount code is selling. Ask what it wants from you.
Checking sources is a practiced set of moves, not a talent. In one college study, students knew a lot about how media work, yet only 7% checked elsewhere and judged correctly on any of four tasks, and they claimed to check more than they did. In another course, on one task, students who left a site to check it went from 3 of 87 before four one-hour lessons to 67 of 87 after. Practice the moves; do not assume you use them.
B. Sorting the kinds of content
A statement of fact is one that evidence could confirm or refute, and it stays a fact claim even when it is wrong; a statement of opinion rests on values. 'The library cut weekend hours in May' is a fact claim; 'the library is failing the town' is opinion. In a 2019 US survey, nearly half of respondents sorted such statements no better than chance. Label each sentence of a piece as fact or opinion before you check anything.
Paid content is supposed to carry a label, and the exact words matter. FTC guidance says readers likely understand terms like 'Ad' or 'Paid Advertisement,' that 'Promoted' is ambiguous, and that 'Sponsored by' may suggest a sponsor funded a story it did not shape. A brand logo alone is unlikely to be enough. When a small tag sits near a headline, read it before the first paragraph and ask who paid.
Native advertising is an ad made to match the articles, reviews or videos around it. The FTC treats it as deceptive when readers cannot tell it is an ad, because knowing that is likely to change how much weight you give it. Picture a decor site's '10 rugs worth the money' where every rug links to one store. Before taking advice, check whether one company gains from every recommendation.
Satire and parody are humorous content, and they can fool readers who miss the joke. Researchers tracking low-credibility sites note that satire which spreads widely is frequently taken as real news, and that fake-news sites often call their own content satire. A screenshot reading 'City Votes to Cancel Tuesdays' shows no site name. Find the original page before reacting, and judge the site by what it mostly publishes.
Universities and journals issue press releases about their own research, and news often builds on them. In a study of UK university health releases, a third of those about correlational research claimed a stronger cause than the paper; when they did, 81% of matching news stories exaggerated too, against 18% otherwise. A replication found the same link for causal claims. Trace a study story back to the release, then the paper.
User-generated content is material ordinary people post, like a phone video of a storm. It can show what no reporter saw, but no one vouches for when, where or what it shows. In one study of images in public political WhatsApp groups in India, about a third of the misleading ones were old photos reposted out of context, and AI can make fakes hard to spot. Treat a viral clip as a lead: confirm place, date and original poster.
Original reporting rests on evidence the outlet gathered itself: interviews, documents, data, being there. Aggregation summarizes what another outlet found, so it can be no stronger than that source. Misinformation research describes stories passing from site to site, each citing the last, until the first report is hard to find. When a story says 'according to reports,' click through until you reach whoever did the finding.
Every piece of content exists for a reason. The National Library of Medicine suggests asking what a site is for: informing people, selling something, or pushing one person's or group's views. A page called 'Sleep Science Explained' that links every paragraph to one mattress is selling. Purpose does not settle accuracy, but it tells you which claims need outside confirmation.
C. Evaluating a source
Lateral reading means opening new tabs to learn about a site or claim instead of studying the page itself. Researchers who watched professional fact-checkers found they left unfamiliar sites almost at once to ask who was behind them, and judged credibility faster and more accurately than students or history professors. Landing on 'Coalition for Healthy Lakes,' search the name before you read its articles.
An About page is written by the site itself, so it shows how the owners want to be seen. Groups write these pages to cast their own aims in a good light, and a site calling itself independent may not be. In one course study, 14 of 87 students leaned on About pages to judge a site before lessons in lateral reading, and 2 did after. Check what outside sources say about the owner.
A domain is the name part of a web address. Impostor sites can copy a real site's look but use a near-miss spelling or a different ending, such as .net instead of .com. Endings mislead too: in one course study, 23 of 87 students at first treated a .org ending as a sign of reliability. In the US, .gov sites belong to government bodies. Read the address letter by letter, then check the owner elsewhere.
A track record is how often a source has been right and how it acts when wrong. It sets your starting confidence, not the verdict: in one set of survey studies, headlines from distrusted outlets that people found plausible were mostly true when fact-checked. A correction that names the error and explains it is an observable sign of accountability. Use reputation to decide how hard to check, then check the claim itself.
Credentials signal expertise, which is why they get faked or stretched. A federal security agency warns that influence campaigns invent experts and think tanks with false credentials, and researchers note that students often take staff with advanced degrees as a mark of credibility. A physicist writing on nutrition is outside their field. Ask whether the training fits this subject, and confirm it beyond the person's own page.
Funding and ownership show where a source's interests lie. In one course study, a quick search on a science-sounding site's name could reveal funding from companies with a stake in its topic, which students who stayed on the site missed. But self-interest does not make a claim false; rejecting it for that reason alone is the circumstantial ad hominem fallacy. Let funding raise how hard you check, not decide the answer.
A transparent source shows how it works. The National Library of Medicine suggests looking for an editorial board, a review process, a content policy, writers' qualifications, cited sources and update dates. Accountability research adds that a real correction names the error, supplies the right information and explains how it happened. Policies are promises, so also check whether the site keeps them.
Checklist methods have you read the page closely, but its cues, such as polished design, a .org ending, links to studies or an About page, are easy to fake, and in course studies students leaned on them wrongly. Fact-checkers leave the page; in one civics course, students taught that move judged trustworthiness correctly more often than those in regular sections. Use the page for clues and other sources for the verdict.
D. Investigating who is behind it
A byline is only a name until you confirm it. Influence campaigns create fake personas, some with verified-looking accounts, and publish articles under them. Search the author's name with the topic: a real writer usually has past work, an employer and others who mention them. For a social account, a federal security agency suggests checking its age, photo, bio, follows and posting pattern. No trace anywhere is a warning sign.
Pink slime sites, named after a cheap meat filler, look like local newspapers but publish partisan content or corporate PR. Researchers describe them as often part of large networks with automated stories and undisclosed political ties, trading on trust in local news. A 'Riverton Daily Ledger' you have never heard of could be a real paper or one of these; its look will not tell you. Search the name to see who runs it.
A front group is an advocacy group that speaks as concerned citizens or neutral experts while a backer with a stake pays for it. One communication scholar argues such backers are sometimes hidden but often named in plain sight, and naming them does not end their sway. Imagine 'Parents for Safe Playgrounds' is funded by one equipment maker. Search the group's name with the word funding, and weigh its claims knowing who pays.
Astroturfing is many coordinated, often fake accounts posting one message so it looks like a grassroots wave while hiding who started it. A federal security agency suggests checking an account's age, photo, bio and activity. The FTC says a sudden burst of reviews, or a reviewer who posted only once, can sometimes mean fakes. Fifty near-identical replies in an hour call for checking the accounts before you count a crowd.
Influence operations often build networks of sites and accounts that look separate but are run together. A federal security agency describes fake sites that impersonate real outlets, and researchers describe local-news imitators run as networks with automated stories. If three unfamiliar 'local' sites ran one article word for word the same morning, treat them as one source. Search a distinctive sentence in quotes to find them.
Some people can speak only without their names, such as an employee reporting wrongdoing who fears retaliation. The SEC's whistleblower program accepts anonymous tips and bars retaliation, yet still screens each tip for quality. An unnamed source in a story is a similar trade: protection for them, less checking for you. Ask why the name was withheld and how the source would know, and weigh the claim by what else confirms it.
A conflict of interest means the person making a claim gains if you believe it, through pay, free products, a job or family ties. FTC guidance expects influencers to disclose such ties. In science news the disclosure often vanishes: in one study, funding appeared in 94% of journal articles but 9% of related news stories, and under 2% of releases or stories mentioned author conflicts. Look for it in the original.
E. Claims, evidence and reasoning
A post rarely hands you a clean claim. 'The new bus schedule is wrecking commutes' mixes a judgment with a checkable fact. Pull out the part evidence could prove or disprove, such as 'average waits rose after the March change.' Then note its strength, because 'linked to' and 'causes' are different claims. Write the claim in one sentence before you search, so you know what would settle it.
Evidence differs in how much you can check it yourself. A published dataset or the text of a report lets you look; a named expert's statement asks you to trust a person; 'people are saying' offers nothing to test. Studies have levels too: the National Library of Medicine notes that randomized trials give the clearest answer on whether a treatment works. For any claim that matters, ask for the document behind it.
The burden of proof is the duty to back up what you assert, and it sits with whoever makes the claim. An ad says a supplement doubles memory and dares doubters to prove it does not. That flips the burden. Logicians call one version the appeal to ignorance: treating a claim as true because no one has shown it false. Ask the claimant for evidence, and hold the claim as unproven until it arrives.
The less likely a claim is, given what is known, the more evidence it needs. Hume's test for miracle reports: accept one only if its being false would be more remarkable than the event. Even his critics grant that a miracle story needs more proof than a mundane one. A neighbor saying a road is closed needs little; a post saying a gadget ends power bills needs a lot. Ask which is likelier: the claim or an error.
Correlation means two things occur together; causation means one produces the other. Things can rise together because a third factor drives both, and randomly assigning people to groups, as trials do, helps guard against that. In one German medical journal's news pages, only 38% of flatly causal headlines rested on randomized trials. 'Tea drinkers live longer' may reflect who drinks tea. Ask how the study was designed.
An anecdote is one person's story; data tell you how often something happens across many people. Stories sway people more than they should: in four experiments, one account of a treatment failing shifted beliefs about whether it works, even when the story was uninformative or shown beside a chart of the statistics. A neighbor whose new roof leaked says little about that brand's leak rate. Ask: out of how many?
A claim is falsifiable if you can name an observation that would show it wrong. The philosopher Karl Popper argued that a theory able to explain every possible outcome is not strong but untestable. A post says a cleanse works, and if you feel worse, that proves toxins are leaving. Now every result counts as support. Ask the claimant what would show the claim false; if nothing could, it tells you nothing you can check.
A fallacy is an argument that seems sound but is not. A straw man distorts someone's view to make it easier to knock down. An ad hominem attacks the speaker instead of the claim; its tu quoque form rejects advice because the adviser does not follow it. A false either-or overlooks other options: 'Either we ban phones in class or no one will learn.' Name the move, then go back to the claim and its evidence.
F. The verification workflow
Strong emotion is a cue to slow down. In experiments, people induced to rely on their feelings believed more false headlines but not more true ones. Another experiment found that asking people to explain how they knew a headline was true or false lowered their intent to share false headlines without lowering true ones. A furious post says the town pool is closing for good: before sharing, ask how you know.
Claims change as they travel. One UK study found that for correlational health research, a third of press releases and two in five news stories worded findings more causally than the journal paper, and outlets can cite each other until the origin is lost. Click a story's links and run photos through reverse image search until you reach the first source. Then ask: does the original say what the share says?
The first article you see is one sample, picked by a feed. Search the claim itself and scan the results before clicking, a habit researchers call click restraint, since sites can push themselves up the rankings. If a bridge collapse is reported only by one unfamiliar site while established outlets say nothing, that silence is information. Look for the best coverage, not for reasons to keep the first.
Real material from years ago misleads when shared as if new. In one study of images in public political WhatsApp groups in India, old photos reposted out of context made up about a third of the misleading ones. A flood photo from 2015 posted during this week's storm is genuine but false in context. Check the original date, run a reverse image search, and take extra care with breaking news, when reliable reports are scarce.
Headlines compress and often overstate. A study of one German medical journal's news section found 45% of headlines exaggerated the research conclusions, and nearly half of the flatly causal headlines were softened in the article text below. 'Walking cures back pain' may top a story saying walking 'may ease' pain for some. Read down to the paragraph that says what the study actually found.
Quote mining is lifting a statement out of its context so it seems to mean something else. A city engineer says, 'People claim the bridge is unsafe, but our inspection found no problems,' and a post quotes her as saying 'the bridge is unsafe.' Every word is real; the meaning is reversed. Find the full transcript or document, read the sentences around the quote, and ask what point the speaker was making.
Summaries drift from their sources, so a strong check is reading the thing described. In one analysis of 33 dental health misinformation messages, investigators found no supporting study for most, and the rest leaned on real research pulled out of context. News also drops details the paper includes, such as who funded it. When a post says 'a new study proves,' find the study and compare its conclusion with the claim.
Not every claim deserves an hour. One model of how people judge credibility holds that they check harder when accuracy matters to them, and fact-checkers reached sound verdicts quickly by leaving a site and consulting trusted sources. A funny pet video you will not share needs little; a health claim before a decision needs more. If a search goes nowhere, restart with new terms rather than grinding on.
G. Fact-checking as a practice
Fact-checkers do not rate everything they see. They monitor what is circulating, choose which claims to check, sometimes by how much harm a false version could do, and test each against authoritative sources and earlier checks. Many real claims turn on cause and interpretation, not a single number. Before trusting a verdict, look for the trail: the exact claim, the sources consulted, and the reasoning linking them.
A verdict like "mostly false" sits on a scale an outlet designed. Some use five or six steps; others add labels such as miscaptioned or satire. When two major checkers rated matching claims, most disagreements came from different scales, focus or timing, not opposite conclusions. "Lack of evidence" means proof was missing, yet in one experiment readers took it almost as "false." Read a label as the summary of an argument.
Fact-checking networks hold members to written codes: be nonpartisan and fair, show the sources and method behind each verdict, and keep an open corrections policy. One global network has applicants independently verified against its code. Anonymous community notes, by contrast, hide their writers and methods. A code is a promise, so judge a checker by its pages: linked sources, a stated method, and visible corrections.
A statement of fact can be shown true or false with evidence, even when it is wrong: "the town has 40,000 residents" is checkable. "The town is too crowded" is an opinion resting on values. A claim that something could happen is harder, since a check reports what is, not what might be. A claim that fits every possible outcome cannot be tested at all. Ask first: what evidence would show this false?
A verdict applies to the claim as the checker worded it, on the day it was checked. "The city cut library hours" could be true for one branch and misleading for the whole system. When two major checkers rated matching claims differently, the cause was usually a different scale, focus or timing, not a real contradiction. So confirm the check tested the wording you saw, note its date, and read the sources it links.
Checkers pick suspicious or potentially harmful claims, so most published verdicts land on false or mostly false. Two major checkers overlapped on only about 6.5 percent of their claims. In one study of two health myths, debunking posts took a week to catch up with the falsehood, and a storyline can outlive the claims inside it. Treat fact-checks as careful rulings on selected claims, not a census of truth.
An early worry was that correcting a myth repeats it and makes it stick. Later experiments mostly failed to find this backfire: in two sets of three studies, corrections of claims people had never seen generally did not raise belief, even a week later. In one election study, a correction worked across political divides, but by the next survey wave beliefs had drifted back. Correct anyway, and expect to say it again.
H. Images, video and audio verification
A reverse image search takes a picture instead of words and returns pages where matching or near-matching copies appear. In one study, journalists checking shared images were told to start there, to see where each had already appeared online. A copy posted last year undoes a caption saying the photo is from today. No match proves nothing, though: the original may never have been posted or indexed.
The easiest fake needs no editing: take a real photo from an old storm or protest and caption it as today's. In one study of images shared in political group chats, old pictures taken out of context were the largest single type of image misinformation, ahead of memes and edited photos. Because the pixels are real, the image passes a test for editing. Test the claim the caption makes, not just the picture.
A crop removes everything outside the frame, and that can include the part that explains the scene. A tight shot of a few people shouting can read as a fight; the wider shot shows fans cheering at a match. Researchers note that fact-checkers struggle less with deepfakes than with crude edits like crops and with out-of-context visuals. When an image matters, look for the uncropped original before you accept its meaning.
Open-source investigators verify footage by geolocating it: matching what is visible in the frame against maps and satellite images. Dating it works the same way. The sun's direction and height change with place, date and time, so shadows narrow down when a photo was taken, and archives of past weather show whether it really snowed that day. When the scene and the caption disagree, trust the evidence in the frame.
Three ways to alter a photo: splice two images together, copy a patch within one photo to hide or repeat something, or erase an object and fill the gap. In experiments, people were poor at telling original from edited photos and often could not locate the edit, and did poorly at spotting errors in shadows and reflections. In one study, erase-and-fill edits were hardest to catch. An oddity you notice is a lead, never proof.
Synthetic media is content generated or heavily altered by algorithms, including AI. In one set of experiments, AI-made photos of both invented people and real celebrities could not be told apart from real photographs. Voices can be cloned from a short clip posted online, and scammers use clones to fake a relative's emergency call. A realistic look or familiar voice is not evidence; verify through a channel you already trust.
Detection tools learn the traces left by particular generators and may miss newer ones. Compression and resizing can wipe out those traces, and modern cameras' built-in AI processing can add look-alike traces to real photos; one tampering detector flagged harmless added text. NIST calls detection a cat-and-mouse game. Treat a "real" or "fake" score as a weak clue, and settle it by tracing where the file came from.
Provenance is a file's origin and history: who made it, when, and what was changed. A record of it can travel with the file as metadata or a watermark, but metadata is often stripped when files are shared online, and anyone copying a file can alter it. So work outward: find the earliest copy, the account that posted it, and whether the original source stands behind it. Origin answers what pixels cannot.
I. Numbers and statistics in the news
A count tells you how many; a rate divides that count by the size of the group it came from. "300 bike thefts" means one thing in a town of 5,000 and another in a city of two million: on the same scale, that is 6,000 versus 15 thefts per 100,000 people. Rates let you compare places, years and groups of different sizes. When you meet a big number, ask: out of how many?
When a rate moves from 10 percent to 12 percent, it rose 2 percentage points, the plain difference between the two. As a percent change it rose 20 percent, because the 2-point gain is a fifth of the starting 10. Both are correct, but they answer different questions, and the bigger-sounding one is not more true. Check the starting level, and say which kind of change you mean.
Relative risk compares the risk in one group with the risk in another. If a habit raises a condition from 1 in 10,000 people to 2 in 10,000, the risk doubled: a relative risk of 2, or a 100 percent increase. Yet the absolute risk rose by just 1 in 10,000. Both statements are true, but only the second tells you how much one person's chances changed. When you see a ratio, ask for both rates out of the same total.
The mean adds every value and divides by how many there are; the median is the middle value, with half above and half below. On a street where nine households earn 50,000 dollars and one earns 5 million, the mean income is 545,000 dollars while the median is 50,000. When a few values sit far from the rest, the mean drifts toward them. Ask which summary was used.
A poll measures a sample, not everyone, so its estimate carries sampling error, summarized as a margin of error. Larger samples shrink it. If 48 percent favor a new park and 46 percent oppose it, with a margin of 3 points, the poll cannot say which side is larger. The margin also leaves out other errors, such as slanted questions or people who refused to answer. Read it as a range.
Survey answers respond to how a question is asked. In one experiment, people rated a material less favorably when it was called "plastic" than when it was called "polymer." In another, a general risk rating shifted toward whatever specific questions came just before it. Loaded terms, two questions packed into one, and agree-or-disagree formats can add distortion. Before trusting a poll number, read the exact question.
Any series with ups and downs can rise or fall depending on where you start and stop. Measure a town's crime from an unusually quiet year and a normal year looks like a surge; start from a peak and the same data show a decline. A study of anti-vaccine posts citing statistics found cherry-picked evidence, such as chosen countries or periods, in about half. Ask why a chart starts where it does, and check the longer record.
Say 900 of a city's 1,000 cyclists wear helmets, which halve injury risk: 10 percent of bare-headed riders get hurt and 5 percent of helmeted ones, so 10 injuries versus 45. Most of the injured wore helmets, yet calling helmets useless ignores the base rate, how common each group was to begin with. A study of anti-vaccine posts using statistics found this error in 58 percent. Before comparing counts, ask how big each group was.
J. Charts and visual argument
Bar charts compare values by bar length, so their vertical axis should start at zero. Start it at 95 and a gap between 96 and 98 makes one bar three times as tall as the other. In one study of 130 vocational students, truncated bars misled readers, with no sign that graph skills protected them; another study found a truncated line chart had no measurable effect on opinions. Read the axis numbers before you judge a gap.
A dual-axis chart plots two measures with different units, each against its own scale on one side. Because the maker sets both ranges, stretching or squeezing either one can make the lines climb together, cross or split, with no change to the data. Even lines that truly track show only that two things moved at once, not that one caused the other. Read both scales, and ask what each line does on its own.
When a chart uses pictures, the eye compares their areas. Draw a coin stack twice as tall to show twice the value, keep its shape, and it covers four times the area, so the gain can look far bigger than double. Tilting a pie into 3D can make front slices look larger than back ones. In one study, height-scaled icons misled readers while a 3D pie made no measurable difference. Compare the printed numbers, not the shapes.
A courier can beat its rival on short routes and on long routes yet post a worse on-time rate overall, if most of its jobs are the hard long ones. This reversal, called Simpson's paradox, can appear when you combine groups with different mixes. The opposite trick splits data into whichever subgroups look best. When a chart shows a total, ask what the breakdown shows; when it shows a breakdown, ask why those groups.
A choropleth map shades each area by its value after sorting values into a few bins. The binning method changes the picture: splitting into equal-sized groups can put two nearly identical counties in different colors, while "natural breaks" can leave very different numbers of counties in each color. Rates for small areas can swing on a few cases. Read the legend first, and ask what a different split would show.
A single line climbing upward says little until you know what normal looks like, what similar places did, and how precise each point is. Estimates from samples come with ranges, and a wide range calls for caution. Yet a scan of 69 Canadian COVID-19 dashboards found only one that showed uncertainty visually. In one experiment, the story around a chart swayed opinions more than its axis. Ask: compared with what, and how sure?
The small print answers what the picture cannot: what was measured, in what units, from which source, for which places and dates. A note that a later year used a new method can turn an apparent jump into an artifact of counting. A federal statistics agency warns that data from different sources can differ in procedures, timing and question wording. Read the notes before the bars, and compare only like with like.
Pictures persuade by seeming to show the claim. In one experiment with AI-made images, false headlines were believed more when the picture looked realistic and seemed to supply evidence; in a preliminary result, rating the images first slightly lowered belief in the false ones. A file photo of empty shelves beside a shortage story implies the shortage is here and now. Ask what the image proves about the specific claim.
K. Misinformation, disinformation and malinformation
These terms sort content by truth and intent. Misinformation is wrong but shared by mistake, like an honest reporting error. Disinformation is false or misleading and made to deceive. Malinformation is real, such as a genuine private email, used out of context to mislead or harm. One false post can be misinformation when a relative shares it and disinformation when its maker does. Check accuracy first; intent comes second.
Calling something disinformation is a claim about motive, and motive rarely shows in a post. Even researchers split on how much intent should count: in a survey of 150 misinformation experts, views on intentionality diverged. A post you suspect was planted may come from a sincere person, and an honest error can be reposted by someone who knows it is false. Check the claim and its context, which you can verify.
Rumors thrive when events are uncertain and official news is slow. A claim appears, spreads, and repetition alone makes it feel truer. Fact-checkers tracking one disaster rumor watched its numbers grow by the day, and when officials denied it, the story turned into a tale of a cover-up instead of fading. Treat early claims in a crisis as provisional, and be wary of a story that adapts to every correction.
A data void is a search term with little credible information behind it, such as breaking news or a term that has fallen out of use. Campaigns can seed a fresh phrase on social media, nudge people to search it, and let their own pages fill the results. One study found chatbots cited dubious sources mainly on narrow questions that mainstream coverage had not answered. Search for the event itself, not the slogan.
Coordinated inauthentic behavior means many accounts acting together while hiding who runs them. Campaigns post the same message from numerous fake accounts so it looks like a grassroots wave, a practice called astroturfing. Personas get invented histories and AI-generated profile photos to seem real, and fake engagement makes posts look popular. Watch for bursts of similar wording from new or thin accounts.
In a hack-and-leak, someone steals real documents and releases them, sometimes through a third party, at a moment of their choosing. Because the files are genuine, they feel like proof, yet the leaker decides which pages you see and how they are framed. Officials classify facts used out of context to mislead as malinformation. Ask who gained access, why now, what was left out, and whether the pieces mean what the framing says.
Prebunking borrows the vaccine idea: a weakened dose of a manipulation technique, explained, before you meet the real thing. Short videos naming tactics like scapegoating helped adults 45 and older in 12 EU countries spot manipulation, with small effects. A 19-second ad improved recognition for months. Yet in feeds mixed with real posts, one inoculation's effect on engagement vanished. Learn the techniques, then keep checking.
A typical correction has three parts: the claim, a clear label that it is false, and an account of what is actually true. Across four experiments, beliefs shifted about as much whether the myth or the fact came first, and a version that recast each myth as a plain true statement did about as well. Repeating a myth to people new to it generally did not strengthen it. Get the clear, true account to the people who need it.
L. Propaganda and influence operations
In 1937 an institute studying propaganda defined a set of devices. Name-calling pins a bad label on an opponent. Glittering generalities wrap a cause in virtue words like freedom or progress. Transfer borrows the glow of respected symbols, such as a flag. Bandwagon says everyone is already joining. Each offers a feeling or a crowd instead of evidence, so translate the message into a plain claim and ask what supports it.
Repetition makes true and false claims alike feel truer. The leading explanation is that a familiar claim is easier to process, and that ease feels like truth. A 2026 review of 182 studies found this illusory truth effect small but reliable, and one study saw the biggest jump at the second exposure. Judging accuracy at first sight weakened it, perhaps only for about two days. If a claim sounds familiar, ask where you heard it.
Flooding, also called "firehosing," means pumping out so much content, often from fake accounts, that other voices are drowned out. The aim may be less to persuade than to leave people feeling nothing is true. One study found state outlets sent a hostile audience a wide, self-contradicting spread of claims, but sent a friendly audience a few consistent ones. When you meet a flood, step back to a few reliable sources.
Some news outlets are funded or controlled by a government, which can limit their independence. Platforms sometimes label such accounts, but a label helps only if people notice it: one study found a prominent label reduced the effect of misinformation, while another found small tags had no measurable effect, apparently because few noticed them. Look up who funds and controls an outlet, then judge each claim on its evidence.
A narrative from a foreign campaign gains credibility when a local influencer, activist or writer repeats it. Campaigns target such people, who are often unaware they are repeating a planted story, and the original source drops out of view. Some operations have even recruited real freelance writers for fake news outlets. When a striking claim arrives from a familiar voice, trace it back: who said it first?
Influence campaigns often work by amplifying disputes people already feel strongly about. One study of a state-backed troll operation found it courted several opposing communities at once, feeding each one outrage about its rivals through a mix of facts, inflammatory opinion and falsehoods. When content makes you furious at the other side, ask who benefits from that anger, whichever side you are on.
Once a planted narrative takes hold, many people spreading it may be real and sincere. Researchers studying one campaign found state media and suspected fake personas entangled with sincere activists and journalists. That makes "it's just bots" a poor explanation, and calling every believer an agent is unfair and often wrong. Separate where a story began from the honesty of the person sharing it.
M. Persuasion and emotional manipulation
Strong emotion can push you to share before you check. Across four experiments, people told to rely on their feelings believed more false headlines, but not more true ones. A survey linked anger with seeing false health claims as credible, and an analysis of millions of posts found links to negative news shared nearly twice as often. When a post makes you furious or afraid, treat that feeling as a cue to verify first.
A curiosity-gap headline hints at a surprise but withholds it: "You will not believe what this town found under its library." It works by making you feel you need the missing piece. Across nearly 9,000 headline tests, vagueness won clicks only up to a point, after which too little information lost readers. Before clicking, guess what the story probably says, then see whether it delivered more than that.
Some words describe and judge at once. "The council rammed through a scheme" and "the council approved a plan" report the same vote, but the first already says it was wrong. Propaganda researchers list loaded language, strong words with emotional connotations, as a standard technique, and in one experiment even swapping "plastic" for "polymer" shifted attitudes. Restate the claim in neutral words and see what remains.
Framing is the choice of which side of a fact to show. Offer a sure option and a gamble described by what each saves, and people tend to take the sure thing; describe the same options by what each loses, and many switch to the gamble. A large replication found the effect even when every outcome was spelled out. When a choice is framed one way, restate it the other way before you decide.
Balance is fair when both sides have similar evidence. When they do not, a 50-50 debate misleads: one engineer who trusts a bridge design facing one who doubts it looks like a split field, even if 40 of 42 independent reviews approved it. In experiments with university students, a warning about false balance before such a debate reduced its harm to attitudes. Ask not how many sides spoke, but how much evidence each has.
Persuasion can run through the speaker's character as well as the argument, so a trusted face can carry a weak claim. Ask three things. Is the expert's field the topic at hand? Are they paid by or tied to whoever benefits? Federal ad guidance says such ties should be disclosed. Is the authority named at all? In one study, false health posts more often cited vague sources like "experts," while debunks more often named agencies.
High numbers suggest many people independently vouched for a post, but they may not be independent, or real. In a news game, players shown high engagement counts were more likely to like and share low-credibility posts and less likely to fact-check them. A 2024 federal rule bans buying or selling fake reviews and fake followers or views for commercial use. Judge the claim; treat a big number as the start of a question.
"Act now," "offer ends at midnight," "do not hang up": urgency works because checking takes time. Federal consumer guidance lists pressure to act immediately as a sign of a scam, and regulators have flagged countdown timers on offers that were never really time-limited. Some philosophers treat this kind of pressure, short of force, as manipulation. Slow down; honest sellers, the FTC notes, give you time to decide.
N. How your own thinking gets in the way
Confirmation bias is the pull to favor information that fits what you already believe. In one diary study during a national election, the most common reason people gave for calling a post false was that it clashed with what they knew, yet a fact-checker rated only 10% of their picks false and 37% true. Run the same check on a claim you like as on one you dislike.
Motivated reasoning means processing information so it lands on a conclusion you already preferred. Researchers disagree about how it works: one account says people more inclined to think things through use that effort to defend their side, while other evidence found that more deliberate thinkers were better at rejecting false headlines that suited their politics. Before you weigh evidence, name the result you are hoping for.
Identity-protective cognition is resisting facts that threaten a group you belong to, often by using your reasoning skill to argue against them. Picture a home-brewing club proud of its slow method, shown a blind taste test where judges could not tell it from a quick one: members pick at the test instead of weighing it. Ask how you would judge the same evidence if your group had no stake in it.
The availability heuristic means judging how likely an event is by how readily instances surface in memory. A review of health scares says steady coverage of a harm makes cases easy to recall, which can lead people to overestimate it. The evidence is mixed: one experiment testing whether personal stories trigger the heuristic found no sign of it. Before judging a risk, look up how often it actually happens.
The illusory truth effect is the tendency to rate a statement as truer after you have met it before. A meta-analysis of 182 studies found the effect small but reliable, for true and false statements alike, and weaker when people judged accuracy the first time they saw a claim. So rate a claim on first sight, and treat a feeling of familiarity as no evidence at all.
The hostile media effect is the tendency for partisans to see even-handed coverage as slanted against them. In two experiments, partisans rated neutral news as biased when it came from an outlet they saw as leaning the other way, though this did little to change what they were willing to share. When a story feels unfair, check what it actually says before blaming the outlet.
The third-person effect is judging that unwelcome messages work on others more than on you. A reanalysis of advertising studies found a clear gap for undesirable ads but the reverse for desirable ones, which people said swayed them more. In one US survey, people who thought misinformation swayed others more than themselves were less satisfied with democracy. Assume a message that could fool your neighbor could work on you too.
Checking a claim is not the same as doubting it by reflex. A meta-analysis of 67 experimental papers, with participants in 40 countries, found people told true from false news well, and when they erred, they more often rated true news false than believed false news. So there is more to gain from accepting true news. Aim to accept well-supported reports as readily as you reject weak ones.
O. Algorithms, feeds and search
On a ranked feed, software orders posts by predicting what you will engage with, not by when they were posted. Ranked feeds can also add posts from accounts you never chose: one study found a large platform showing people images from pages they did not follow. Your feed is a sample picked for you, so missing from your feed is not missing from the world.
Engagement-based ranking promotes posts predicted to draw clicks, likes and shares. In one audit of 806 users, it amplified angry posts hostile to political opponents, posts users said made them feel worse about the other side, and users did not prefer the political posts it chose. Ranking by what users said they wanted cut the anger but risked more one-sided content. A post that angers you may be ranked high for that reason.
Personalization is tailoring what each user sees, usually by predicting what that person will engage with. An audit using 100,000 automated test accounts found a video platform recommended content matching each account's political leaning, though some moderate and cross-cutting videos still appeared. Before assuming everyone saw what you saw, compare your feed with someone else's.
A filter bubble is the idea that personalized feeds wall you off from other views. In four experiments with nearly 9,000 people on a site built to mimic a real video platform, slanted recommendations had little short-term effect on opinions, though long-term effects were not ruled out. A computer simulation showed echo chambers can also form when people quit mixed groups, with no algorithm involved.
Recommendation drift is the worry that each suggested video leads somewhere more extreme. An audit with 100,000 automated accounts found no meaningful rise in ideological extremity deeper into recommendation chains, but a growing share of picks from conspiratorial and other problematic channels. That share stayed small, at most 2.5%, yet over a third of accounts met it. Notice where a chain of suggestions has taken you.
Platforms handle false content in different ways. Removal deletes a post or account, the most restrictive step. A label leaves the post up and adds context in plain view. Downranking leaves it up but shows it to fewer people, which is harder to notice. One study estimated posts drew about 46% fewer reposts than expected once a crowd-written note was attached. A post you cannot find may be downranked, not deleted.
AI tools can produce hallucinations: plausible, fluent statements that are fabricated or wrong. Letting a tool search the web can improve accuracy, yet errors persist because the model predicts likely next words rather than checking facts. A small audit also found chatbots sometimes cited unreliable sites on topics where credible sources were scarce. Open the source an answer cites and confirm it says what the answer claims.
P. Social platforms and how things spread
Sharing is not sorted by truth. In 2.7 million posts from news outlets and members of Congress, posts about the other political side were shared at about twice the rate of posts about their own. In another study, links to negative news were shared 1.91 times as often. Two experiments found belief-confirming claims shared more than clashing ones, while novelty did not predict sharing. Pause on posts you most want to share.
A screenshot or forwarded image arrives without its link, date, author or surrounding thread, so you cannot see what it first showed. In one study of images shared in public political chat groups, old photos reshared with a new story were the largest kind of image misinformation, about 34%. Before trusting one, look for the original post or an earlier copy of the image.
Reach measures attention, not accuracy. The FTC tells endorsers to disclose a material connection to a brand, such as payment or free products, and to place it in the post itself, since one only on a profile or behind a 'more' link is easy to miss. In two experiments, once the cues were explained, a verified credential made posts seem more credible even when it was unrelated to the topic. Check the disclosure and the fit.
Encrypted messaging groups pass content along by forwarding and through phone contact lists, out of public view. Community leaders interviewed in one US study described these closed apps as a testing ground where narratives are tried out before moving to open platforms; much of what circulated was video. Treat a forwarded message as unsourced until you find where it started, and check it outside the group.
Engagement bait is a post designed to harvest reactions, comments and clicks. A study of 125 pages posting AI-made images found spam pages using clickbait to send people to low-quality sites, and scam pages selling products that did not exist. The feed showed these images to people who did not follow the pages, and many commenters seemed not to realize they were fake. Ask who gains from your click.
Platforms use badges to signal that an account is authentic. In two experiments, after the badges were explained, people still tended to treat a verified identity as a sign the message was credible, mixing up who is speaking with whether they are right. In another experiment, health posts were rated no more accurate with a badge than without. Judge the claim on its evidence; a badge at most tells you who posted it.
Replies under a post can carry a correction, but they can also mislead. In a study of fake AI images, many comments treated the images as real while a few warned others. In an experiment in three countries, short user corrections had small, often non-significant effects on belief in false posts, and adding a fact-check link made no clear difference. Scan replies for a specific, checkable correction, then verify it yourself.
Q. Conspiratorial thinking
Researchers describe five elements in conspiracy theories: a pattern linking many events, agency (it was intended), a coalition of plotters, a threat, and secrecy. Secrecy makes them hard to refute, since missing evidence can be blamed on a cover-up. One philosopher adds that such theories lean on odd clues rather than direct proof. Ask what evidence would count against the theory.
Some conspiracies have been proven real. What separates a documented plot from a conspiracy theory is evidence: direct proof such as records or a confession, often brought out by whistle-blowers or the media. One account adds that genuine whistle-blowers admit the limits of what they know. Ask of any claimed plot what direct evidence exists.
Proportionality bias is assuming that the size of a cause must match the size of its effect. Researchers describe it, along with the urge to see events as intended, as a bias that helps generate conspiracy theories. If a citywide blackout traces to one overloaded transformer, the small cause can feel wrong. Remember that accidents and chains of small errors can produce large events.
A loaded question builds in an assumption no one has shown, like asking why the council hid the audit before anyone has shown it was hidden. A string of such questions lets a speaker imply a charge without stating it, which dodges the duty to defend a claim you make. Turn each question into the statement it implies, then ask for evidence for that statement.
An argument from ignorance treats a claim as true because nothing disproves it. Conspiracy theories add a twist: the lack of evidence is read as proof the plotters hid it. Such theories have been called self-sealing, meaning immune to counterevidence. If a park closure is called a secret land deal and no records show a deal, that absence is not support. Ask what finding would change the believer's mind.
A theory can come bundled with a community. In interviews with four former believers, loneliness and hard life events helped draw them in, and the online community gave them identity and belonging while making it hard to leave. A key reason they left was a clash between their own experience and the group's doctrine. When someone defends a theory, consider what the group gives them.
Engaging with someone's own reasons can shift a belief. In one experiment with 554 people, those who described a theory they held and the evidence behind it, then debated it with an AI model, lost more confidence than a control group. The drop was larger when they thought the partner was human, which the authors tied to how neutral the partner seemed. Ask what convinced them, and respond to that.
R. Science and health information
Coverage often goes beyond the research it reports. In one analysis of UK university press releases, a third of those on correlational studies made stronger causal claims than the paper, and over a third of those on animal or cell studies implied human results; when a release did this, most related news stories did too. Find the study and check whether it measured cause or only association, and whether it studied people.
A preprint is a complete, public draft of a study that has not yet passed peer review, in which experts check whether the data support the conclusions. Its results may change after review. In one US study, only about one in five people could define the term. Critics note that review itself can be inconsistent and biased. Treat a preprint as provisional and a published paper as checked, not proven.
A case report describes one patient and can flag something new, but it is anecdotal. Observational studies record what happens without intervening. In a randomized controlled trial, people are assigned to groups by chance, which gives the clearest test of whether a treatment works. Systematic reviews look across many studies. Before trusting a health claim, find out which kind of evidence it rests on.
Small studies are more prone to chance results, and larger ones can be more reliable. Who was studied matters too: results in animals may not hold in people, and people who differ from you in age or sex may respond differently. A trial of 30 young men may tell you little about older women. Check how many took part and who they were before applying a result to yourself.
One new study rarely settles a question. Studies can disagree because they used different designs or asked different questions, and a result that breaks with earlier work needs more studies to confirm it. When many studies reach the same conclusion, that agreement helps show a finding is reliable. Ask how a new result fits earlier research, and look for a systematic review.
Funders can sometimes profit when a study turns out a certain way, so it helps to know who paid. That rarely reaches you through the news: in one analysis, 94% of journal articles stated their funding but only 9% of related news stories did, and authors' conflicts of interest reached under 2% of stories. Funding is not proof of bias, but read the paper's funding and disclosure sections.
The FDA lists warning signs of health fraud: one product that cures many unrelated diseases, quick cures, personal testimonials in place of evidence, words like miracle or secret, and claims that doctors are hiding it. In one study of 309 health claims in videos by professionals, 62% rested on little or no evidence, and the weakest drew more views. Ask what study supports a claim.
S. Political and election information
US elections are run by states and more than 10,000 local jurisdictions, and the rules differ from place to place. The Election Assistance Commission points voters to their local elections office for practical voting details, and says its own state summaries should be checked against state and local sources. If a post gives a voting deadline, confirm it on your election office's site.
Campaign advertising exists to persuade, so read it as advocacy rather than neutral information. For federal races, the Federal Election Commission requires political committees' public ads and mass mailings to carry a clear disclaimer naming who paid and whether a candidate authorized the message. Read that line first, then check the ad's factual claims against an independent source.
A poll estimates opinion from a sample of people, so its numbers carry a margin of error that signals how reliable they are. A gap between two figures may be chance unless a test that accounts for that margin says otherwise. Polls that use random samples with larger samples and response rates tend to be more accurate than self-selected ones. Read the method and the margin before the headline.
Some election claims hinge on how voting is run. Election night results are unofficial: the federal Election Assistance Commission explains that mail, provisional and military ballots may still be counted, so totals can shift before the canvass, and in most states an audit, ends in certified results. In one diary study, only 4% of flagged claims were checked against authoritative sources. Check claims with your election office.
One essay on AI notes that fact-checkers struggle less with deepfakes than with crude edits, such as cropping, and visuals stripped of their context. Fake quotes spread too: in one study of political chat groups, memes carrying fake quotes or statistics made up 30% of image misinformation. If a council member's clip ends at 'we should close the library,' find the full recording to hear what came next.
A factual claim can be checked; a value judgment weighs what matters more. One essay on misinformation argues that public disagreement is shaped by values and identity, not only by false information. Two neighbors can agree a new road saves ten minutes yet split over whether that is worth losing a park. Separate the checkable part from the priorities, and watch for words like wasteful that mix fact and judgment.
A primary source is an original record made at the time, such as meeting minutes, a budget or a filing; a post about it is a secondary account. If a post says the school board voted to cut buses, look for the motion and the vote in the minutes. FOIA, the Freedom of Information Act, covers federal agencies: many post records online and take written requests. Check whether a record is already public first.
Using a source's politics as a stand-in for accuracy is a shortcut that misleads in both directions. In one experiment, people fell for non-political false headlines more often when they came from a publisher on their own side, and an inoculation game reduced this. A meta-analysis found people more skeptical of news that clashed with their politics. Check claims from allies as hard as claims from rivals.
T. Journalism standards and ethics
Verification is checking whether a claim holds up before reporting it. In 1920, a press critic contrasted the rules for trial testimony with news, which reaches the public with no test of reliability. Journalists interviewed about unreviewed studies described extra skepticism and clear labels, but many said they lacked time or expertise to vet them fully. Look for signs of checking: documents, named sources, stated limits.
A careful report tells you who supplied each fact: a named official, a document, a wire service or an unnamed source. As early as 1920, a press critic argued that every news item should state whether it came from a news agency, a reporter or a press bureau. When a source is unnamed, you are trusting the outlet's judgment of that person. Prefer reports that say how they know, and weigh unnamed claims more lightly.
To judge a report's independence, ask whether the people it covers or whoever paid for it could have shaped it. One quality checklist for science journalism counts a stated conflict of interest as a mark of integrity. Yet disclosures often get lost: in one analysis, conflicts noted in about a quarter of journal articles reached under 2% of press releases and news stories. Look for who paid and who stands to gain.
A story can be true and still cause needless harm through the details it includes. Guidelines for reporting on suicide, for instance, advise against describing the method; a recent review of 15 studies found coverage followed that advice more often after such guidelines appeared, though effects on suicides were unclear. Ask whether a detail, name or image serves the reader or only exposes someone.
Balance means presenting the relevant views fairly, but giving a strong view and a weak one equal space can make the evidence look evenly split. A review of health scares describes equal coverage leading audiences to assume equal evidence. In three experiments, readers discounted a dissenter without expertise, but two of the three samples largely ignored a vested interest. Ask whether each side has comparable evidence.
During breaking news, information arrives piecemeal and many rumors stay unverified for a long time. A study of 330 rumor threads across nine news events found people tended to back unverified rumors, false ones took longer to resolve than true ones, and even news organizations posted confident-sounding claims that later proved unverified. Note what is confirmed, and wait before sharing the rest.
An embargo is an agreement not to publish before a set time. A 2006 journal editorial explained that science reporters often learn of studies from press releases sent about a week before publication, which gives them time to report and leads outlets to publish the same day, while journals gain coordinated publicity. A wave of near-identical stories may trace to one release, so read the study.
News reports aim to tell you what happened; editorials and opinion columns argue for a view. A study of newspaper editorials notes they are not bound by the norms of balance and objectivity that apply to news reporting. As early as 1920, a press critic proposed a news service with editorial matter strictly kept out. Check the label, and judge an opinion piece by whether its facts are accurate and its reasoning sound.
U. Corrections, accountability and press freedom
Published work sometimes gets things wrong, so ask what an outlet does next. One European fact-checking code asks members to fix known errors with a note in the same article and to keep a findable corrections policy and record. PubMed links an erratum notice to the article it corrects. Before you trust a site, look for its corrections page: a steady log of labeled fixes tells you more than a site that never admits one.
Stories are updated as news develops, and that is normal. The problem is a change with no note. One fact-checking code asks members to flag even added information in an update note. When a name in a citation is corrected, PubMed can keep the old form searchable and shows a notice with both. If a page you relied on now says something different with no note, compare it with an archived copy saved earlier.
A charge from one side is half a story. Journalism textbooks from the early 1900s already told reporters to let the accused speak, not only the accuser, and to check a one-sided claim with the other side or label it as one-sided. So when a story makes a serious accusation, look for the target's response or a note that they did not reply. If neither appears, weigh the charge as unanswered.
Defamation is a false factual claim that harms reputation. In US law, substantial truth is ordinarily a defense, and calling a remark opinion does not shield it if it reasonably reads as a factual claim. A public official who sues must also prove actual malice: knowing falsity or reckless disregard of whether it was false. Accurate but harsh criticism of a mayor is not defamation. This is general information, not legal advice.
FOIA, the Freedom of Information Act, lets almost anyone request records from federal agencies. Agencies must release them unless one of nine exemptions applies, such as personal privacy or national security, and black out protected parts. FOIA covers existing records only: an agency need not create a document or answer questions. When a story cites records released under FOIA, you can often request them yourself.
The First Amendment says "Congress shall make no law ... abridging the freedom of speech, or of the press." The Supreme Court treats prior restraint, government stopping publication in advance, with a heavy presumption against its validity. Remedies for unlawful speech, like a defamation suit, generally come after publication. A story blocked in advance never reaches you; one that runs can still be challenged.
Access shapes what you read. PubMed lists citations and abstracts for free but holds no full texts, so a search there shows you the summary first. Paid content is also hard to preserve: the Library of Congress says its web archive cannot collect material that requires a subscription. Before relying on a claim, look for a free full text, such as a copy in PubMed Central, and say so when you read only the abstract.
V. Search and finding better information
The words you type decide what comes back. In five experiments, people asked to search online before judging false news stories were more likely to rate them true than people who did not search, mainly when their results held low-quality sources. Queries that pasted a story's headline or link surfaced unreliable results far more often. So describe the underlying claim in your own neutral words, and try a second phrasing.
Good search tools let you narrow results instead of scrolling. PubMed filters by publication date and by article type, such as Review or Clinical Trial, and lets you tie a term to a single field. The National Archives catalog filters by type of material, such as textual records or sound recordings. You can also screen results by address: one ending in .gov belongs to a US government organization. Narrow first, then read.
The first results on a search page are not always the most reliable. The FTC warns that some scammers and shady firms buy their way to the top, and advises spotting the small Ad or Sponsored labels and scrolling past them. One study found clinics selling scientifically unsupported cancer treatments bought search ads keyed to the questions patients type. Check the label before you weigh the claim.
Words that only a false story uses can steer a search toward pages that repeat it. In one study's close look at a single false article, queries that added one charged word from it turned up unreliable results far more often than those without it. The authors tie this to data voids: corners of the web where low-quality sources back each other up. Search the event in plain words, like 'plant closing reasons', not the accusation.
Much good evidence sits inside databases with their own search boxes. PubMed indexes over 40 million biomedical citations. Chronicling America holds millions of US newspaper pages published through 1963. The National Archives catalog describes federal records, though only a small share are digitized. Even the Library of Congress's web archive cannot collect database content that needs a search, so go to the database itself.
Web pages get edited without notice, and some disappear. Web archives such as the Library of Congress's save copies meant to show how a site looked at a given moment, and most sites it collects are captured more than once. Compare a capture from before a change with one from after. The Library's archive is incomplete, though: its staff choose what to collect, and it cannot capture some content, such as pages behind a paywall.
Search results vary. One study found that six search engines returned very different sources for the same query at the same moment, and that even users of one engine could see different results. So note the exact words, the tool, the date and what you found. PubMed's search history records each query with its result count and time, and can be downloaded. A permanent link to the exact record beats memory.
W. Primary sources and documents
Primary sources are first-hand evidence made at the time under study, such as letters, photos and official records. Secondary sources look back to explain or interpret events, usually from some distance in time or place. A court ruling is primary; a history of it written decades later is secondary. Newspapers from the period can count as primary sources too. When an account and the record disagree, go read the record.
The National Archives teaches one routine for any document: meet it, observe its parts, make sense of it, then use it as evidence. Observing means answering who wrote it, who received it, when and where. Only then ask what it says and why the author wrote it. An internal memo and a press release from the same day can describe one event very differently, and knowing which you hold changes how you read it.
Every source is made at a moment, under pressures, for an audience. The Library of Congress and the National Archives both ask readers of a primary source what was happening when it was created and who it was meant for. A letter written knowing an official might read it, or a company report written for investors, says what that setting allowed. Read the silences too, and ask what the author had reason to leave out.
A typed version of an old document is a transcriber's reading of it. In writing from before the mid-1800s, a long s looks like an f, so 'fome' may really be 'some'. National Archives guidelines tell volunteers to type words exactly as written, misspellings included, and to mark unreadable words as illegible. A translation adds a second layer of choices. When one word carries a claim, check it against an image of the original.
Provenance is a record's history of ownership and custody. One US archival principle, continuous custody, holds that records should stay with the organization that made them or pass directly to an archive, to protect their integrity. A document that surfaces with no account of where it has been may still be real, but it lacks that support. Ask who held it, since when, and how it reached the person showing it to you.
Archives hold what someone chose to record and later chose to keep. Official records come from organizations at work, and archivists appraise which have lasting value; others may be destroyed. The National Archives notes its catalog cannot list every person named in its records. Silence in the record does not prove nothing happened. Ask whose view is missing and where else it might survive, such as letters or oral histories.
A citation is a promise that a reader can find what you saw. The Library of Congress gives two reasons to cite primary sources fully: to credit the creators and publishers, and to let others find the source quickly and correctly. It adds that citing makes you think about where a source came from. 'Census records show' is weak; the table name, the year and a stable link let anyone check the claim.
X. Scholarly and expert information
A title signals expertise only within a field. The FTC's Endorsement Guides apply the same idea to ads: a person presented as an expert must really have expertise in the subject they endorse, and a bare title like doctor can imply more than the person has. A physician quoted on bridge safety is an outsider there. Ask what this person has studied or done on this exact question, and how they reached their view.
Before a journal publishes a paper, a few experts judge whether it seems sound. A 2026 opinion piece stresses what review leaves out: nobody repeats the experiments, and reviewers seldom rerun the numbers. Mostly they judge whether it seems believable and consistent. Peer-reviewed papers are still sometimes retracted, for pervasive error or data no one can reproduce. Treat 'peer reviewed' as a minimum bar, not proof.
Some outlets mimic academic journals but skip the checks. The FTC alleged that one publisher claimed rigorous peer review and prominent editors, while many papers got little or none and listed editors had not agreed to serve; a court later ruled for the FTC. NIH's warning signs include unclear fees, aggressive requests for papers, false editorial boards and suspicious review. Check a journal's review process before citing it.
A single study is one turn in a long conversation. Its reference list points back to the earlier work it built on. PubMed also lists newer papers that cite it, though that list may be incomplete, and links published comments that support or challenge it. Reading forward shows how later work responded. A finding that nobody has repeated is less settled than one that independent labs have reproduced.
Journals retract papers for pervasive error or for data that cannot be supported or reproduced, and publish errata for smaller fixes. In PubMed, a retracted article is labeled and linked to its retraction notice, and each erratum is linked to the article it corrects. PubMed does not say whether a retraction came from honest error or misconduct. A paper still circulating online may be one its own journal has withdrawn.
An abstract sums up a study's question, methods, results and conclusion; the conclusion is the authors' reading of their results, which other experts may not share. Watch for spin: wording that plays up benefits when the main outcome showed no significant difference. One review found spin in the titles or abstracts of 35 of 66 allergy and immunology trials whose main results were not significant. Find the main outcome first.
Coverage that pairs one critic with one supporter can make a field look split when it is not; this is called false balance. Counting helps. In one large survey of scientists, about 93 percent agreed that science has shown beyond reasonable doubt that a virus causes COVID-19, so some 7 percent did not. A list of a thousand dissenters can still be a small share of a big field. Ask what share of relevant experts agree.
Y. Advertising, sponsorship and influence
Native ads take on the look of what surrounds them, such as a mattress seller's paid article on sleep habits laid out like a site's own health stories. The FTC treats an ad as deceptive when its format misleads people about its commercial nature, even if every product claim in it is true. The closer an ad matches the real content, the more likely it needs a clear label. Ask who paid for it before you weigh what it says.
Under the FTC's Endorsement Guides, a needed disclosure must be clear and conspicuous, meaning easy to notice and easy to understand, and online it should be unavoidable. FTC staff suggest plain words such as Ad or Paid Advertisement, call Promoted ambiguous, and note that Sponsored by can suggest a funder who did not shape the content. A label seen only on a profile page or after tapping More fails that test.
Some review sites and creators earn a commission when you buy through their links. That does not make the advice wrong, but FTC staff say such a connection should be disclosed clearly and close to the recommendation. FTC staff note that the bare phrase affiliate link may not tell readers the writer gets paid, while paid link or a plain statement about commissions can. Weigh the advice knowing who profits.
A post can look like a personal tip yet be paid promotion. FTC staff guidance says influencers should disclose their ties to a brand, whether money, a job, or a personal or family link, and free products count too. The disclosure belongs in the post or video itself, not only on a profile or buried in hashtags. Vague tags such as spon or collab fall short. With no disclosure, you cannot tell whether a tie exists.
Reviews can be bought, written by insiders or posted on a site the seller quietly runs. A 2024 FTC rule bars paying for reviews that must take a set positive or negative stance, and bars passing off a company-controlled site as independent. The FTC also advises checking for a sudden burst of reviews, reading a reviewer's other reviews, and accepting that fakes are hard to spot by eye. Start with impartial expert reviews.
For a small paper, local ads can be a lifeline, and that money can bend coverage. In interviews in rural Alberta, Canada, a former publisher said a business pulled its weekly full-page ad after the paper refused to retract a story; another ran preview stories only for businesses that bought ads, unlabeled. MedlinePlus applies the same logic to health sites: one funded by a company may lean toward its products. Ask who pays.
Websites and apps track what you do with cookies, pixels and device identifiers, and advertisers use that data to show ads matched to your interests or location. An FTC staff report found that targeted ads made up most of the revenue at many large social media companies it studied, a reason to collect vast amounts of data. A message may reach you because of your profile, not because it is news. Ask why you, and why now.
Z. Sharing responsibly and local news
Sharing puts a post before people who might never have seen it. In one online experiment, people who first had to explain how they knew whether a political headline was accurate then rated themselves less willing to pass on the false ones, while their wish to share true ones held steady. The effect was modest and measured intentions, not real posts. Before you share, ask yourself the same thing: how do I know this is true?
A bare screenshot strips out what readers need to judge it. MedlinePlus tells readers to look for the sources behind information and the date it was written or updated, so give your readers both: link the original, not a copy, and add the date. If a claim comes from one side only, say so, as a 1911 reporting textbook told reporters; flag unconfirmed claims too. 'County post, 9 a.m.; cause unconfirmed' takes seconds.
Corrections can work: in one experiment, people who saw a false post corrected in the replies believed it less than those who saw it uncorrected. But a fix only reaches people who see it. So post your correction where the first post went, say plainly what was wrong, and link a better source. If you edit or delete the original, say that you did, so readers know what changed.
Correcting a relative can feel risky, so some people stay quiet. In one UK study, people close to someone spreading misinformation often preferred to raise it face to face; some used humor or talked around the topic. A social media experiment found facts mattered more than tone: neutral, warm and rude corrections cut onlookers' belief in a myth equally, though rudeness made the exchange seem less civil. Bring facts, kindly.
A news diet has two dials: breadth and pace. Breadth means more than one outlet and more than one kind, such as local and national, so a claim can be checked elsewhere; MedlinePlus advises seeing whether other reliable sources say the same thing. Pace means how often you check. The CDC suggests taking breaks from news and social media, since a constant stream of bad news can wear on you. Set both on purpose.
A 2024 study notes that over a quarter of US newspapers closed between 2005 and 2020, and that earlier work tied such losses to lower civic engagement. In browsing data, it found little sign that people in news deserts, places without a local paper, read more low-quality sites, but some sign they read more national news, which other work suggests covers local issues less. If a local paper still covers your town, follow it.
Many websites earn money by hosting ads. One study found that ads, often placed automatically by ad platforms, help fund websites that publish misinformation, and that most company decision-makers surveyed did not know their ads ran there. Your choices feed the same system: link original reporting, not copies, pay for outlets you rely on if you can, and report scams and fake reviews at ReportFraud.ftc.gov.
Keentune is not affiliated with or endorsed by the organizations whose documentation informs these maps.
All exam, test, and product names and trademarks are the property of their respective owners and are used here for identification and reference only. Keentune is independent study practice — not affiliated with, authorized, or endorsed by any of these organizations.