Deepfake Meme Reputation: How a Viral Meme Reached 31M Views

Here's something that should genuinely unsettle you: a video can change how you think about a real person, even after you know the video is fake.

Not because you're gullible. Not because you missed the label. Because that's just how human memory works. And if you've ever shared a funny AI meme without thinking twice, you've already felt this effect from the inside, you just didn't have a name for it.

31M Watched a Deepfake Reputation Crisis. Then Shared It Anyway.

Earlier this year, a video of footballer Erling Haaland appeared to show him catching a glimpse of himself on a restaurant TV screen and doing a gloriously awkward double-take. It racked up 31 million views on X in a matter of days.

There was one problem. The original footage came from a Chinese comedy skit. Someone had swapped the face using AI, a deepfake (a video where one person's face is replaced with someone else's using artificial intelligence). AFP, one of the most well-funded fact-checking operations on the planet, needed days to trace the clip back to its real source.

By then, it didn't matter. The meme had already done its work.

And here's the part that catches most people off guard: a huge number of the people who kept sharing it already knew it was fake. They shared it because it felt like Haalandspontaneous, a little weird, weirdly charming. The correction was a footnote. The image was the story.

The Thing Nobody Warns You About: Continued Influence

Psychologists have a name for what happened to those 31 million viewers. They call it the Continued Influence Effectthe documented fact that false information keeps shaping how people think, even after they've been explicitly told it's wrong. This article is part of a series, start with Age Verification Api How It Works.

This isn't a theory. Communications Psychology (Nature) published peer-reviewed research specifically on AI-generated deepfake videos, finding that transparency warnings, "this is fake" labels, did not prevent viewers from being influenced by what they saw. Their perceptions of the person in the video shifted anyway.

A separate review in AI & Society confirmed the same pattern: disclaimers reduce but do not eliminate the effect. And research published in ScienceDirect found that deepfakes can shift attitudes and impressions even in viewers who actively questioned whether the content was authentic.

So the "just label it fake" solution, the one that feels obvious, the one platforms rely on, is genuinely incomplete. Not useless. Incomplete.

---

Why Your Memory Keeps the Deepfake File

Here's an analogy that actually captures this. Imagine a friend tells a story about you at a party, completely made up, but it makes you sound hilarious. You correct it immediately. "That never happened." But by then, forty people heard the original version. Each of them now has two mental files on you: the correction, and the story. When they're deciding what kind of person you are, both files are open. The false story doesn't vanish just because it was debunked. Worse, it colors how they interpret your real behavior later, "oh, classic, just like that time you supposedly did X."

That's continued influence. And that's exactly what happened to Haaland's public image, except at the scale of 31 million people, not forty.

The reason it hits especially hard with celebrities and athletes comes down to something researchers call parasocial relationshipsthe one-sided emotional bonds fans feel toward people they've never met. (It's the reason you feel like you know a footballer, even though they have no idea you exist.) Fan engagement research found that 31% of Gen Z sports fans now feel more connected to individual athletes than to their teams. When you're that invested in a person, you are actively hungry for more content about them. An AI meme that fits your mental image of them feels like discovery, not deception. You share it before the critical part of your brain has finished loading.

"AI memes are not just jokes, they are quiet acts of narrative construction. Every share votes for a version of a person that the person themselves may never have approved." NSS Magazine, on the Haaland deepfake phenomenon

Why People Share Deepfakes: It's Completely Understandable

Most people think the danger of an AI-edited video is deception, someone trying to fool you into believing something false. So naturally, the safety check most people run is: can I tell if this is real? Can I spot the blurry edges? Does the face move weirdly? Is the lighting off? Previously in this series: Instagram Turned On A Setting That Lets Strangers Make Ai Ph.

That framing makes total sense. It's the obvious question. It's also not the most useful one.

When detection is the only lens you use, you treat the problem as solved the moment you spot the fake. "I knew it was AI. I shared it as a joke. No harm done." But the Haaland case makes clear that sharing something you've already identified as fake can still quietly reshape how millions of people perceive a real human being, without their consent, without their input, and without any way for them to stop it.

The speed gap makes this worse. AI can generate a convincing face-swap in seconds. A credible fact-check takes days. By the time the label arrives, the content has already reached most of the people it's ever going to reach. As Yellow.com reported in their breakdown of the systemic risk, the real threat isn't one person creating a fake, it's millions of everyday users amplifying content faster than any institution can respond.

Nobody runs a detection check before hitting share. And even when they do, a "fake" label doesn't stop the continued influence effect from doing its thing.

The Better Check: Three Questions, Not One

This is where it gets practical. At CaraComp, working with facial comparison every day, the lesson we keep coming back to is this: accuracy isn't the only question worth asking about an image. Whether you're evaluating evidence or just deciding what to repost on a Tuesday night, the same three-part check applies.

Next time an AI-edited video of a real person lands in your feed, funny, flattering, or otherwise, run this before you share:

1. Is it clearly labeled? Not buried in a thread. Not mentioned in the comments. Right there on the content itself, where someone who sees a screenshot or a repost will still know what they're looking at. If it isn't labeled, adding that context yourself before sharing is the single highest-impact thing you can do. Up next: That Enter Your Birthday Box Is Dead Heres What Actually Che.

2. Did the person consent? Some public figures actually lean into AI memes of themselves, Haaland's camp has largely let this one run, and there's a reasonable argument it's boosted his personal brand. But "they haven't complained yet" isn't the same as consent. A real person's face, voice, and mannerisms are being used to build a version of them they never approved. That matters even when the result is flattering.

3. Could it mislead people about who they really are? This is the sneakiest one. You don't need malicious intent for an AI meme to leave millions of people with a subtly wrong impression of a real human being. Even a sweet, funny clip adds a new file to the mental folder people keep on that person. Ask whether the version of them in this video is fair, not just entertaining.

The Haaland meme is, in many ways, a best-case scenario. Funny content. A public figure whose team leaned in rather than fought back. A fact-check that eventually landed. And still, 31 million views, a permanent addition to how the world imagines him, and no moment where he got to say yes to any of it.

Now think about the next viral AI clip, where the person isn't famous enough to have a PR team, where the content isn't flattering, and where the fact-check never comes. The only thing standing between that person and 31 million wrong impressions is the few seconds you take before hitting share.

A video doesn't have to fool you to change the story. That's the part nobody tells you, and now you know it.

What Makes a Video Meme Spread So Fast

A video meme moves faster than almost any other kind of content online because it asks nothing of the viewer. You don't have to read anything or think hard, you just watch, laugh, and tap share. That low friction is exactly why the Haaland clip hit 31 million views before anyone official had a chance to weigh in. The format itself is the accelerant, regardless of whether the video is real or generated.

Meme Templates and the Deepfake Pipeline

Most viral deepfakes don't start from scratch. They borrow familiar meme templates, the reaction shot, the double-take, the awkward glance, because audiences already know how to read them instantly. When an AI face-swap slots a real person's face into a template people recognize, it feels less like a fabrication and more like a punchline. That familiarity is part of why the Haaland meme spread so effortlessly: the format did the persuading before anyone even asked if the video was true.

How a Meme Generator Changes the Math

A meme generator used to mean swapping captions onto a still image. Now the same idea applies to full video, where a generator can put a real face into a wholly different scene in minutes. This matters because the barrier to making a convincing deepfake meme has dropped from "technical skill required" to "a few clicks," which means far more of these clips will reach your feed going forward. The tools got easier; the three-question check from this article did not get any less necessary.

Video Meme Culture and Consent

A video meme built around a real person raises a consent question that older, static memes rarely did. When someone edits a photo, the joke usually stays obviously fake. When an AI system generates a video meme, the joke can look and sound exactly like the real person, which blurs the line between parody and misrepresentation. That blur is precisely why the second question in the checklist, did the person consent, matters so much for deepfake meme content specifically.

The Haaland deepfake meme worked partly because it was funny and partly because it was harmless-seeming. But a meme doesn't have to be mean to reshape how people see someone. Every video meme that goes viral adds one more layer to the public's mental picture of that person, whether or not the person ever agreed to be part of the joke.

Spotting a Meme Before You Share It

Before you repost any meme involving a real person's face, take ten seconds to look for a label, a watermark, or a caption disclosing that it's AI-generated. If none exists, that's your cue to either skip the share or add the context yourself. A meme without a label passes the confusion on to the next thousand people who see it, and most of them will never trace it back to the original video.

The deepfake algorithm behind clips like the Haaland video works by mapping one face onto another person's expressions and movements frame by frame. Understanding that mechanism doesn't make a meme less funny, but it does explain why the results look so convincing even to a careful eye. Making deepfakes this realistic used to require a studio's worth of equipment; now it takes a laptop and an afternoon.

Not every use of this technology is playful. There are documented cases where people abuse deepfakes to harass or humiliate someone, which is a very different situation from a viral joke about a footballer's double-take. Deepfakes created for cruelty deserve a different response than deepfakes created for comedy, even though the underlying deepfake video technology is identical in both cases. Creating deepfakes with good intentions doesn't erase the fact that the same tools can be turned toward harm.

An ai-generated image or clip doesn't announce its own origin story, which is exactly the problem. The wrapped meme format, a real person's likeness wrapped around someone else's performance, asks viewers to do the work of separating fact from fabrication, usually with no clear signal telling them where to start. That's why the labeling habit described earlier in this article matters so much: it puts the signal back where the viewer can actually see it.

None of this means you should stop enjoying a genuinely funny deepfake meme. It means treating the format with the same care you'd give any other claim about a real person, because that's exactly what it is, dressed up as a joke.