On March 20, 2023, a photograph began circulating on Twitter and Reddit. It showed Pope Francis wearing an enormous white puffer jacket — the kind of oversized coat you might see on a street in Seoul or Tokyo, not on the head of the Catholic Church. The image was sharp. The lighting was convincing. The Pope's face looked exactly right.
Within 24 hours, millions of people had seen it. Many shared it with captions like "the drip Pope" and "finally, a Pope with style." A significant number of those people believed it was real. It was later confirmed to be a Midjourney-generated image — created by a Chicago-based shoe designer named Pablo Xavier who said he made it "just for fun" while on a trip.
What made it spread wasn't that it was technically perfect. It wasn't — look closely and the rosary beads blur oddly. What made it spread was that it fit something people wanted to believe: that a beloved, elderly figure could be cool and surprising. The fake worked because it understood its audience before it understood its tools.
There is a shortcut to understanding any kind of deception: try to create it yourself. Security researchers do this constantly. They build fake phishing emails so they can recognize real ones. Forensic scientists study how paint ages artificially so they can spot forged paintings. Fraud investigators learn how counterfeit currency is made before they look for it in circulation.
This module works the same way. In each lesson, you are going to take apart the construction process of AI-generated fakes — not just what they look like when finished, but how the choices get made: what topic to pick, what emotional hook to use, what details to include to make something feel real, and which details always give things away.
Here's the thing that Pablo Xavier understood intuitively: a good fake isn't primarily a technical achievement. It's a persuasion strategy. The AI image generator is just a tool. The real decisions are human ones: What story do I want people to believe? What makes them more likely to believe it? What details make them less likely to question it?
Once you understand how a fake is constructed — the choices, the hooks, the deliberate details — you will never look at suspicious content the same way again. You will automatically ask: "What decision was made here? Why this image, this headline, this quote?" That question is the most powerful detection tool that exists.
Every piece of AI-generated misinformation — whether it's a fake image, a fabricated news article, a deepfake video, or a synthetic audio clip — requires three decisions that happen before any AI tool is even opened. These aren't technical decisions. They're strategic ones.
Pablo Xavier's Pope image hit all three perfectly. The target was people who liked the Pope and found him endearing. The hook was humor and surprise. The credibility anchors were the realistic face, the Vatican-appropriate white color scheme, and the recognizable rosary. He didn't plan this consciously — but the image worked because it instinctively got all three right.
When you learn to see these three decisions in a piece of content, you're not just spotting one fake. You're reading the strategy behind it. That skill transfers to every fake you'll ever encounter.
In 2023, the same Midjourney tool that made the Pope's puffer coat was used to generate images that looked like Donald Trump being arrested by police — images that spread widely before the arrest actually happened, and which many people initially mistook for real photographs. The person who made those images claimed they were artistic commentary on current events. Were they?
Here's the ethical question this module will not resolve for you: Is there a meaningful difference between creating a fake to understand fakes, and creating a fake that might escape into the wild?
In this module you will design fake content — on paper, with a lab AI, without generating actual images or audio. But the knowledge you build here is the same knowledge someone would need to make real fakes. Knowing how to pick a credibility anchor is useful to a media literacy teacher and useful to a political operative at the same time. The knowledge doesn't come labeled with instructions for how to use it.
Sit with that. We're not going to give you a clean answer, because there isn't one.
Most people who share misinformation are not trying to deceive anyone. They share because the content hit one of the three decisions effectively — they were the target, they felt the hook, they trusted the credibility anchors. You can now see the structure behind content that other people experience as emotion. That's a real and consequential advantage.
In this lab, you won't generate any real fake content. Instead, you'll design the strategy for a hypothetical fake — working out the target, hook, and credibility anchors — and then immediately analyze why that strategy would work or fail. Your lab partner will push back, probe your choices, and ask you to defend them.
This is how professional media forensics researchers think. They reverse-engineer the intent before they analyze the artifact.
On October 5, 2024, a post began circulating on X (formerly Twitter) showing what appeared to be a FEMA (Federal Emergency Management Agency) document. It stated that FEMA was redirecting hurricane relief money from states hit by Hurricane Helene to fund migrant housing. The document looked real: it had the correct FEMA header formatting, official-looking department codes, and a specific dollar amount — $750 per day per migrant.
That single number — $750 — did more work than anything else in the document. It was specific. Specific numbers feel like they came from somewhere real. Round numbers ($700, $1,000) feel estimated. Odd, specific numbers feel like they were pulled from an actual spreadsheet.
The document was fabricated. FEMA confirmed this. But the post was viewed more than 10 million times, and within 48 hours, multiple U.S. senators had referenced its claims in statements. Representative Marjorie Taylor Greene posted it to her official social media. The number $750 had traveled further than any retraction ever would.
There's a concept in psychology called the Concreteness Effect — meaning our brains treat specific, concrete information as more trustworthy than vague, abstract information. This is generally a useful instinct. If someone says "I saw a dog," you get less information than if they say "I saw a black Labrador with a torn left ear outside the 7-Eleven on Fifth." The second version has details that feel like firsthand knowledge.
Fake content creators — whether human or AI-assisted — exploit this instinct relentlessly. They include:
Odd, specific numbers. Not "millions of dollars" but "$2.3 million." Not "hundreds of people" but "312 people." The specificity signals that someone counted.
Named, real institutions. Not "a government agency" but "FEMA" or "the FDA" or "Johns Hopkins University." Real institution names carry borrowed credibility.
Plausible timestamps. Not "recently" but "on Tuesday, October 3rd." Specific dates suggest a paper trail exists, even if it doesn't.
Authentic-looking formatting. Headers, department codes, logo placement — these signal bureaucracy, which signals process, which signals oversight, which signals truth.
When you see a very specific claim — a precise number, an exact date, a named official — your instinct says "this is detailed, so it's probably true." Flip that instinct. Ask instead: "Is this specific detail something I can verify, or is it just specific enough to feel true?" Verifiable specificity is evidence. Unverifiable specificity is a trick.
In 2022, a network of accounts on Twitter began posting what appeared to be excerpts from internal Pfizer documents. The documents were formatted with precise-looking pharmaceutical terminology, clinical trial reference numbers, and patient cohort sizes down to the individual. They circulated widely in communities skeptical of COVID vaccines.
What made them feel real wasn't the content — it was the texture of the formatting. Pharmaceutical documents have a particular look: dense tables, specific adverse event codes, cautious bureaucratic language. The fakes replicated that texture almost perfectly. People who had never read a real pharmaceutical document in their lives recognized the visual pattern of "official document" and trusted it.
This is what's called format mimicry: copying the visual and stylistic conventions of a trusted genre to borrow its authority. A document that looks like a government report benefits from every real government report anyone has ever seen. The credibility is borrowed, not earned.
AI tools have made format mimicry dramatically easier. A language model can generate text that sounds like a scientific paper, a legal brief, a government memo, or a news article — with appropriate jargon, paragraph structure, and citation style — in seconds. The content may be completely false, but the genre signals are authentic.
When you design a fake (even on paper), the most powerful choices you make are the credibility anchors: the specific number, the real institution name, the document-like formatting. These are where most of your persuasive power comes from.
Here's the uncomfortable implication: removing these elements makes a fake immediately obvious. If the FEMA document had said "a lot of money per migrant per day," it would have been ignored. The $750 figure is what made it a political event.
Now apply this when you're reading. When you feel convinced by something, ask: what specific detail is doing the convincing? Can I actually verify that detail, or am I just experiencing the feeling of specificity? That feeling is the trick. The trick works even when you know about it — which is why understanding it is so important.
There's an ethical question hiding here too: if a journalist uses a specific number they can't fully verify in order to make a true story more compelling to readers, is that different from a fake using a specific number to make a false story feel real? Both are exploiting the concreteness effect. The story is different, but the mechanism is identical. Where exactly is the line?
Most people who encountered the FEMA fake felt convinced by the $750 figure and never examined why. You now know the mechanism: specificity signals authenticity. You can now catch yourself in that feeling and ask the right question — not "does this feel true?" but "can this specific claim be traced to a real source?"
Your lab partner will give you a fake piece of content to analyze. Your job is to identify every credibility anchor in it — the specific numbers, institution names, dates, formatting details — and then explain whether each one could actually be verified or is just providing the feeling of specificity.
Then flip it: pick one of those anchors and explain how you'd remove or replace it to make the fake either more convincing or more obviously fake. You'll be making real editorial decisions about what makes false content work.
In April 2018, a video went viral showing a group of children in the United Kingdom being taught a song that included Arabic phrases. The video was captioned with a claim that the children were being forced to recite Islamic prayers at a state school in violation of UK law. It was shared by tens of thousands of people, including several high-profile political commentators.
The children were actually at an Arabic language class — an entirely voluntary after-school activity, completely legal, with parental consent. The video was real. The caption was fabricated. A real video, real children, real voices — and a completely invented meaning attached to all of it by a single line of text.
Channel 4 News and the BBC both ran fact-checks within 48 hours. But the corrections were shared a fraction of the number of times the original was. The outrage had already done its work. Parents were calling the school. The headteacher received threats. A real institution was damaged by a fake caption.
Here's something most media literacy guides get wrong: they treat emotional reactions to misinformation as a failure of critical thinking — as if you should have been more careful, more skeptical, less emotional. This is not accurate, and it's not fair.
Your emotional system is fast. Your analytical system is slow. This isn't a bug — it's how human cognition works. Psychologist Daniel Kahneman called these System 1 (fast, automatic, emotional) and System 2 (slow, deliberate, analytical) thinking. System 1 evolved to keep you alive. If you had to analytically evaluate every threat before reacting, you'd be dead. The problem is that modern misinformation is designed to trigger System 1 so hard that System 2 never gets a chance to engage.
The children's video worked because it triggered moral outrage — one of the most potent System 1 triggers that exists. When you feel that your community's children are being harmed or disrespected, the analytical brain goes offline. This is not unique to any political group, age, or country. It is universal human neuroscience.
Researchers at MIT studying social media found in 2018 that false news spreads six times faster than true news on Twitter, and that the primary driver is emotional novelty — false stories are more emotionally surprising than true ones. This means the emotional hook is not just a feature of misinformation. It's the competitive advantage misinformation has over truth.
Misinformation researchers have identified a specific set of emotions that reliably accelerate sharing behavior. When you design a fake — even on paper — you're choosing from this menu:
Professional fact-checkers are trained to recognize when they're feeling one of these emotions while reading. That feeling is now a signal — not to dismiss the content, but to slow down and engage System 2 before sharing. The question is whether that training can scale to billions of ordinary readers. Most researchers are not optimistic.
When you design a fake with an emotional hook, you're making a concrete, reversible decision: you choose which emotion, and you build content that triggers it. This is the most important thing to understand — the emotion is a design choice, not an accident.
And that means it can be identified and named. Once you name the emotion a piece of content is targeting, something strange happens: the content becomes slightly less effective on you. Not because you've become unemotional, but because you've moved some of the processing from System 1 to System 2. You haven't stopped feeling outrage — but now you're also thinking about why someone wanted you to feel outrage right now, about this topic, in this format.
This is the real skill this module is trying to build. Not "don't feel things." That's not possible, and it's not desirable. But: notice the emotion, name the design choice, and then decide whether to share.
Here's the ethical tension this lesson will not resolve: if a news organization uses a headline designed to trigger moral outrage in order to get people to read a true, important story — is that different from a fake using moral outrage to spread a false one? The mechanism is identical. The content differs. Does the content difference make the tactic acceptable?
From now on, every time a piece of content makes you feel a strong emotion before you've finished reading it, you have a new reflex available to you: name the emotion, identify the design choice, and ask what someone wanted you to do with that feeling. Most people will never develop that reflex. You just did.
In this lab, you'll bring a piece of content — a headline, a social media post, a video description, or something you've seen recently that felt emotionally charged — and run it through an emotional autopsy. Identify which of the six spreading emotions it's using, how that emotion was engineered into the content, and what the sharer was likely hoping you'd do after feeling it.
Your lab partner will challenge your analysis and propose alternative readings. This is not a gotcha exercise — sometimes the most emotional content is also the most accurate. The skill is identifying the mechanism, not assuming the content is false.
In September 2022, a story spread rapidly claiming that a group of Arizona election workers had been photographed destroying mail-in ballots before they could be counted. The post included what appeared to be a photograph of ballots being shredded, with an official-looking envelope visible in the frame. It was shared by thousands of accounts before election officials responded.
What actually happened: Maricopa County election officials identified the photograph within hours. The image showed workers processing damaged ballots according to standard procedure — damaged ballots are duplicated onto clean ballots so they can be machine-read, and the damaged originals are then destroyed in a controlled manner. The "evidence" in the photo — the envelope, the shredder, the uniforms — were all real. The interpretation was completely wrong.
But what stopped this particular fake from traveling further wasn't a fact-check. It was one election worker who had seen the original image months earlier in a legitimate training document about proper ballot processing — and immediately recognized the photograph from that context. She knew the content because she had studied it. That knowledge is what stopped the spread at a critical moment.
This election worker had something that a general fact-checker didn't: she knew the domain. She could look at the content and immediately identify what was true (the images) and what was fabricated (the interpretation). She could do in seconds what would take a journalist hours — because she had direct knowledge of how the thing being faked actually worked.
This is the defender's advantage that this entire module has been building toward. When you understand how fakes are constructed — the three strategic decisions, the credibility anchors, the emotional architecture — you start to see content the way that election worker saw the photograph. You don't just see the result; you see the construction choices. And construction choices leave marks.
Every fake leaves traces of the decisions that were made to create it. These traces are called exposure points — and they are different from the visual artifacts and technical glitches that AI detection tools look for. Exposure points are strategic failures: places where the design choices reveal that someone was trying to persuade rather than inform.
A credibility anchor that can't be verified. An emotion that arrives before you've finished reading the first sentence. A specific number that appears nowhere else on the internet. A document that looks exactly like a real document type but has no traceable origin. These are not accidents — they are the fingerprints of construction decisions.
In the previous lessons, you designed fake content strategies — on paper, in concept, with your lab partner. Now the exercise flips. You're going to take one of those strategies and systematically expose it: find every weakness, every point where a careful reader could have caught the deception.
This is called a red team audit. In cybersecurity, red teams attack systems their own organization built — not because they want to damage it, but because finding the vulnerabilities yourself is far better than having an adversary find them. Journalists, government agencies, and tech companies all use red team audits. You're about to do one on a fake.
For any fake content you designed (or can imagine designing), run through these audit questions:
1. Can any credibility anchor be traced? If you used a real institution name, does the fake content match what that institution actually does or says? A search of FEMA's actual published policies would immediately disprove the $750 figure.
2. Does the emotional hook arrive before the evidence? Real news stories establish facts before conclusions. Fakes almost always lead with the emotional conclusion. If the anger arrives before the evidence, that's an exposure point.
3. Is the source traceable? Who first published this? Can you find an original source for the claim, or does it only exist in shares? Misinformation often has no traceable origin — it appears fully formed in social media without a primary source.
4. Does the content require you to believe something that benefits someone specific? Who gains if people believe this? Real news events benefit from truth. Fakes serve specific interests — political, financial, reputational. Identifying who benefits is a key audit step.
Across this module, you have built something that is not a checklist and not a detection tool. You've built a way of seeing. You understand the strategic decisions that go into fake content. You can identify credibility anchors and test them. You can name the emotional mechanism a piece of content is running. You can audit your own fake to find its exposure points.
These skills were developed by working through the construction process — not by memorizing a list of warning signs. Warning sign lists go out of date as AI tools improve. Construction knowledge doesn't. The strategic decisions behind a fake image in 2023 are the same decisions behind a fake pamphlet in 1935. The tools change. The human choices don't.
At an institutional level — in newsrooms, government agencies, and social media policy teams — the people making decisions about what content gets labeled, restricted, or removed are using versions of exactly these frameworks. They're asking: what was the intent here? What strategic decisions were made? What's the exposure point that confirms this was designed to deceive? You now understand the reasoning behind those decisions, which means you can evaluate them critically — not just trust that institutions are getting them right.
The final ethical question this module leaves with you: you now know enough to build a convincing fake — the strategy, the anchors, the emotional hook. You also know how to expose one. These are the same knowledge. The person who uses this knowledge ethically and the person who uses it to deceive may be making choices that look identical from the outside. What's the difference? Is it just intent? And is intent enough?
AI tools will keep improving. The images will get sharper, the audio will get cleaner, the documents will look more official. But the three strategic decisions — target, hook, credibility anchor — will never change, because they're not technical. They're human. You now understand the human layer. That understanding will outlast every version of every AI tool that exists today.
This is the full-circle lab. You'll take the fake content strategy you developed in Lab 1 — or design a new one now — and systematically audit it using the four red team questions from Lesson 4: Can credibility anchors be traced? Does emotion arrive before evidence? Is the source traceable? Who benefits from people believing this?
Your lab partner will argue the other side — defending the fake's effectiveness wherever you claim it has an exposure point, and challenging you to find stronger vulnerabilities. The goal is to leave with a complete picture of why a fake works and exactly where it fails.