Fake News and Misinformation Detection: Thinking Critically

The Forward That Fooled a Whole School

It is 9:40 at night. A message lands in your class WhatsApp group: "URGENT!! CBSE has postponed all board exams by two weeks due to a technical error in question papers. Please inform everyone immediately. FORWARD TO ALL SCHOOL GROUPS BEFORE IT'S TAKEN DOWN." No logo, no circular number, no link to any official page — just capital letters, an exclamation mark doing a lot of work, and forty unread replies already piling up. By 10 pm it has reached three other sections of your school. By 10:30, a worried parent has called the class teacher. By the next morning, the school has to send an actual official notice just to deny a rumour that never had any official source in the first place.

Nobody in that chain was lying on purpose. Your classmate who forwarded it believed it. The friend who sent it to your classmate believed it too. Somewhere upstream, though, someone typed those sentences knowing they were false, or exaggerated a half-true rumour into a fully false one — and by the time it reached you, it had travelled through so many phones that it looked less like one person's claim and more like "something everyone is saying." That gap — between how confident a message sounds and how little evidence actually backs it — is the entire subject of this chapter. Detecting misinformation is not about having a good instinct for lies. It is a procedure: a set of checks you run, in order, every time, the same way you'd run a program.

Three Words That Are Not Synonyms

People use "fake news" loosely, but researchers who study information disorder — notably Claire Wardle, who built one of the frameworks used by fact-checking organisations worldwide — split it into three precise categories, and the difference between them matters because it changes how dangerous the content is and how you should respond to it.

• Misinformation is false information shared by someone who believes it is true. Your classmate forwarding the fake exam notice is spreading misinformation — they are wrong, but not malicious.
• Disinformation is false information created and shared by someone who knows it is false, specifically to deceive. The original author of the fake exam notice, if they invented it deliberately, produced disinformation.
• Malinformation is information that is actually true but is shared with intent to cause harm — for example, leaking someone's real private medical report to embarrass them. It's factually accurate and still damaging.

Notice that "fake" isn't even the defining feature of all three — malinformation is true. What defines this whole category is the mismatch between the information's effect and the sender's honesty about their intent. Most of what circulates on WhatsApp and Instagram in India is misinformation: ordinary people re-forwarding something they didn't invent and didn't verify.

The Six Disguises

Wardle's framework also catalogues the specific forms false or misleading content takes, because "spot the fake news" is too vague a skill to practise — you need to know what you're looking for. Six of these show up constantly in Indian social media:

• Fabricated content: entirely invented — a quote no one said, a report that doesn't exist, a "news" website built purely to host one false story.
• Manipulated content: genuine media that has been edited — a photo cropped to remove context, a video slowed down or sped up to change what it shows, an old clip re-uploaded with a new caption.
• Imposter content: a fake account or fake website impersonating a real, trusted source — a Twitter handle styled to look like a news channel, a domain name one letter off from a real newspaper's.
• False context: genuine, unedited content shared with a false explanation — a real photo from a 2019 flood in Bihar recirculated in 2026 captioned as "happening right now."
• False connection: when the headline, image, or caption doesn't actually match or support the content underneath it — classic clickbait, where the shocking headline oversells a mundane story.
• Satire or parody: content that isn't trying to deceive at all — sites that clearly label themselves as comedy — but gets mistaken for real news when a screenshot is shared without that label attached.

The important skill isn't memorising these six labels for an exam. It's recognising that "is this fake?" is really five or six different, more answerable questions: Was this invented? Was this edited? Is this account who it claims to be? Is this real content in a false setting? Does the headline match the story? Is this a joke being read straight? Each of those questions has a concrete way to check it, which is what the rest of this chapter builds.

Why Lies Outrun Corrections: The Mathematics of Forwarding

Here's a question that looks like arithmetic and is actually the reason misinformation feels unstoppable. Suppose you receive a WhatsApp message and, within an hour, forward it to 10 people. Suppose each of those 10 people also forwards it to 10 people within the next hour, and so on. How many people have seen the message after n such rounds?

This is a simple function: reach(n) = 10ⁿ. Let's trace it round by round, the way you'd trace a loop in a program:

• Round 0 (the original sender): 10⁰ = 1 person
• Round 1: 10¹ = 10 people
• Round 2: 10² = 100 people
• Round 3: 10³ = 1,000 people
• Round 4: 10⁴ = 10,000 people
• Round 5: 10⁵ = 100,000 people
• Round 6: 10⁶ = 1,000,000 people

Six rounds of forwarding — which, in a busy family or school network, can genuinely happen within a single evening — is enough to reach a million people. That's the nature of exponential growth: it looks almost flat for the first few steps and then rockets upward, because each round multiplies by the same factor as every round before it, rather than just adding a fixed amount. The diagram below plots exactly this: notice how the curve is barely off the ground for rounds 0 through 2, then suddenly climbs near-vertically by round 4. Most of the "explosion" happens in the last one or two rounds, which is exactly why a rumour can feel like it "came out of nowhere" — it didn't; it was quietly doubling and tripling in the background the whole time.

This isn't just a hypothetical. In 2018, researchers at MIT (Vosoughi, Roy, and Aral, publishing in the journal Science) studied roughly 126,000 stories spreading on Twitter over more than a decade and found that false news reached its first 1,500 people about six times faster than true news did. False stories weren't just as fast as true ones — they were reliably faster, because they tend to be more novel and more emotionally charged, and both novelty and emotion make people forward things quickly without pausing to check them. This is also precisely why WhatsApp, after misinformation forwarded across India contributed to serious incidents of mob violence in several states in 2018, restricted forwarding in India to a maximum of 5 chats at a time and added a "Forwarded many times" label to messages that had travelled far from their original sender — a direct, engineered response to the exponential-forwarding problem you just calculated above.

Building a Credibility Score: Turning Judgment into an Algorithm

"Use your judgment" is not a useful instruction, because judgment under time pressure is exactly what misinformation exploits — a message designed to make you feel urgency is a message designed to stop you from thinking. The fix is to convert judgment into a fixed procedure you run before you feel the urge to forward, the same way a program runs the same checks on every input regardless of how convincing that particular input looks.

Here is a credibility-scoring function, written the way you'd write it in Python for your Computer Science practicals. Each check is a boolean (true/false) fact about the message, and each contributes a fixed number of points — positive if it supports trustworthiness, negative if it's a red flag:

def credibility_score(source_reputable, has_evidence,
                       corroborated_by_others,
                       uses_emotional_language,
                       has_date_and_author):
    score = 0
    if source_reputable:
        score += 3
    if has_evidence:
        score += 3
    if corroborated_by_others:
        score += 2
    if has_date_and_author:
        score += 1
    if uses_emotional_language:
        score -= 2
    return score

Let's trace it on two real inputs, the way you'd trace a function on a CS exam.

Message A — the fake exam-postponement forward from the opening story: unknown original source (not CBSE, not the school) → source_reputable = False → contributes 0. No circular, no PDF, no link attached → has_evidence = False → 0. The three friends who also sent it all got it from the same chain, not from independently checking with the school → corroborated_by_others = False → 0. No author name, no date → has_date_and_author = False → 0. "URGENT!!", "FORWARD TO ALL", all-caps → uses_emotional_language = True → −2. Total: 0 + 0 + 0 + 0 − 2 = −2.

Message B — an official press release posted on a verified government or news organisation's own website, with data tables cited, reported independently by two other newspapers, carrying a named author and publish date, written in plain factual language: 3 + 3 + 2 + 1 + 0 = 9.

print(credibility_score(False, False, False, True, False))
# Output: -2

print(credibility_score(True, True, True, False, True))
# Output: 9

Now turn the raw number into a decision. The maximum possible score is 3+3+2+1 = 9 (every positive check true, no emotional red flag); the minimum is −2 (every positive check false, emotional language present). A simple three-way classifier:

def interpret(score):
    if score >= 6:
        return "Likely credible - verify big claims anyway"
    elif score >= 2:
        return "Uncertain - investigate before sharing"
    else:
        return "High risk - do not forward"

interpret(-2) returns "High risk - do not forward" — exactly what should have happened to Message A before it ever left the first phone. interpret(9) returns "Likely credible" — as it should. The value of writing this as a function isn't that a real newsroom literally runs this exact code (professional fact-checkers use more sources and more nuance); it's that turning "does this feel true?" into five yes/no questions and a sum forces you to actually check each one, instead of letting the loudest, most urgent-sounding claim win by default.

The SIFT Method: What Professional Fact-Checkers Actually Do

The five checks above tell you what to look for. They don't tell you how to check it in under a minute on your phone. For that, researcher Mike Caulfield developed a four-step protocol called SIFT, and it's worth learning by name because it's short enough to actually run every time:

• Stop. Before reading further or reacting, pause. Do you already know this source is reliable or unreliable? If you don't know, don't share yet.
• Investigate the source. Who is telling you this, and why should you trust them on this specific topic? A cricket commentator's opinion on a vaccine carries no more weight than yours does.
• Find better coverage. Open a new tab and search for the claim itself, separately from the article. If it's true and important, multiple independent outlets will have covered it. If you find nothing except copies of the exact same post, that's a warning sign.
• Trace claims, quotes, and media to the original context. If a photo or quote is being used as evidence, find where it originally came from. A quote "trimmed" from a longer statement, or a photo from a different year or country, is one of the most common tricks in false-context misinformation.

Steps 3 and 4 point at something researchers at Stanford's History Education Group discovered when they compared how professional fact-checkers evaluate a webpage against how students and even some historians do it. Students tend to practise vertical reading — staying on the page, scrolling up and down, judging it by how professional the layout looks, how confident the writing sounds. Professional fact-checkers instead practise lateral reading — they leave the page almost immediately and open several other tabs to check who runs the site, what other outlets say, and whether the organisation has a track record. Lateral reading consistently outperforms vertical reading, for a simple reason: a page's own design and tone tell you nothing reliable about its accuracy, because a skilled liar can make a fake page look exactly as polished as a real one. The evidence you actually need lives outside the page, not on it.

When the Evidence Is a Photo or Video: Reverse Image Search and Deepfakes

Tracing a claim to its original source is straightforward with text — you search for a sentence. It's harder with an image, because you can't type a picture into a search box the same way. Reverse image search tools solve this by converting the image into a compact numerical fingerprint — a summary of its patterns of edges, colours, and shapes — and comparing that fingerprint against billions of already-indexed images to find visually similar or identical ones. This is exactly how people discover that a photo "from an accident yesterday" is actually seven years old and from a different country: the fingerprint matches an image that was indexed years earlier, and that earlier appearance carries the true date and location.

This tool becomes essential once you account for deepfakes — video or audio synthetically generated by AI models trained on real footage of a person, capable of showing someone saying or doing something they never actually said or did. Deepfakes break a rule most of us apply unconsciously: "I'll believe it if I can see it happening." That rule was reasonable when convincingly faking a video required a film studio's budget. It is no longer reasonable, because the tools that generate convincing synthetic video are now widely accessible. The SIFT step of tracing media to its original context matters more, not less, in a world with deepfakes — because your eyes alone can no longer reliably do the job.

India's Fact-Checking Ecosystem

You don't have to build every check from scratch every time. India has a working fact-checking infrastructure worth knowing by name. The PIB Fact Check unit, run by the Government of India's Press Information Bureau, specifically verifies claims about government schemes, policies, and announcements — useful for exactly the kind of "official circular" rumour in this chapter's opening story. Independent organisations such as Alt News and BOOM investigate viral claims across politics, health, and current events, and both are signatories to the International Fact-Checking Network's Code of Principles, a set of standards (transparency of sources, transparency of funding, a public corrections policy) that IFCN-certified fact-checkers commit to following. Checking whether a fact-checker has already covered a viral claim is often the fastest possible version of "find better coverage" — someone may have already done the tracing for you.

Three Misconceptions Worth Unlearning

Misconception 1: "It has thousands of likes and shares, so it must be true." Popularity measures how emotionally engaging or surprising a message is, not how accurate it is — and as the exponential-forwarding math above shows, a message needs no truth at all to reach huge numbers quickly, only a forwarding factor and a few rounds. Share count is evidence about the message's virality, not its veracity. These are different quantities that happen to share no relationship with each other.

Misconception 2: "I can tell a fake photo or video just by looking at it." This was closer to true a decade ago, when manipulation left visible seams. Modern manipulated and AI-generated media is frequently built specifically to survive casual visual inspection. Treat "it looks real" as providing zero evidence either way, and rely on tracing and reverse search instead of your eyes.

Misconception 3: "A professional-looking website is a credible one." Imposter content thrives on exactly this assumption. A fake news domain can be built in an afternoon with a masthead, a clean layout, and a name one character away from a real outlet's (a common trick is swapping ".com" for ".co", or inserting a hyphen). Professional appearance is a design choice, cheap to copy; it says nothing about who owns the site, how it's funded, or whether its claims have ever been checked.

Worked Practice: Score a Real Case Yourself

Try running the full procedure on this scenario before reading the analysis: a message circulates claiming "Eating a certain fruit combination causes instant poisoning — a doctor confirmed 3 deaths this week, please forward to save lives," with no doctor named, no hospital named, and no city named, forwarded to you by two separate school friends who each got it from different family WhatsApp groups.

Run SIFT first: Stop — this triggers a strong fear reaction, which is itself a signal to slow down. Investigate the source — there is no source; it's an anonymous forward with no attributed author. Find better coverage — a lateral search for the specific claim turns up nothing from any hospital, health ministry, or news outlet, only more copies of the identical forwarded text. Trace the claim — "a doctor," with no name, cannot be traced to anyone; this is a hallmark of fabricated content.

Now score it: source_reputable is False (anonymous forward) → 0. has_evidence is False (no named doctor, hospital, or report) → 0. Here is the important trap: your two friends both sending it feels like two confirmations, but check where each one came from — different family groups, yes, but neither is an independent original source; both are just further copies of the same anonymous forward. corroborated_by_others must mean corroboration from a source that investigated independently, such as a health authority or news outlet, not more copies of the identical rumour — so this is False → 0. has_date_and_author is False → 0. uses_emotional_language is True ("save lives," urgency) → −2. Total: −2, classified as high risk — do not forward, and if you're concerned, check the actual food-safety claim on a recognised medical or fact-checking source before acting on it at all.

Summary

Misinformation, disinformation, and malinformation are distinct: honest error, deliberate deception, and true-but-harmful disclosure. False content typically arrives disguised as one of a few recognisable forms — fabricated, manipulated, imposter, false-context, false-connection, or misread satire — each with its own tell. Forwarding chains grow exponentially (reach(n) = kⁿ), which is why rumours can reach huge audiences within hours and why real research has found false news spreading measurably faster than true news online. A fixed, five-question credibility score — reputable source, evidence, independent corroboration, absence of emotional pressure, visible author and date — turns a vague feeling into an auditable procedure, and SIFT (Stop, Investigate the source, Find better coverage, Trace to the original) gives you the concrete steps to answer each question, with lateral reading (checking other tabs) consistently beating vertical reading (judging the page by its own look). Reverse image search and awareness of deepfakes extend the same tracing principle to photos and video. India has working fact-checking infrastructure — PIB Fact Check, Alt News, BOOM — that has often already investigated a viral claim before you encounter it.

Check Your Understanding

1. A message reports a real, unedited photograph from an actual 2021 protest, but captions it as happening in your city this week. Which of Wardle's six categories does this fall under, and why does the photo being genuine not make the post accurate?
2. Using credibility_score, compute the score for a message from an anonymous forward (not reputable), with a linked government PDF as evidence, corroborated independently by one news outlet, written in a calm factual tone, but with no author name or date given. Show each term before summing.
3. Explain, using the reach(n) = kⁿ model, why a forwarding factor of k = 5 reaches far fewer people after 4 rounds than k = 10 does — compute both values for n = 4 and compare.
4. A friend says, "I always check if a video looks fake before believing it." Identify which misconception from this chapter this reflects, and explain what check they should be running instead.
5. Explain the difference between lateral and vertical reading, and describe why a fact-checker deliberately leaves a suspicious page almost immediately after opening it.
6. Why does receiving the same rumour from three different friends NOT count as three independent corroborations in the credibility-score algorithm? What would count as genuine corroboration instead?

Think About It

Think about this: How would you explain fake news and misinformation detection: thinking critically to a friend who has never seen a computer? What real-world analogy would you use? Imagine you had to build a system using these concepts — what would be your first step? Try this: before moving on, write down three things you learned and one question you still have.