AI Prank on TikTok: the Homeless Man Hoax and a Safe Version

The AI prank on TikTok is a hoax built on one AI-generated picture. You take a photo of your living room, ask Gemini or Snapchat to add a stranger to it, text it to a parent with something like “he knocked, so I let him in”, then post their reply. The best-known version, the “AI homeless man prank” of fall 2025, ended with real 911 calls in the US, the UK and Ireland. Below: where it came from, how to do a version that won’t bring the police to your door, and how to turn it into TikTok views.

Where did the AI prank on TikTok start?

Know Your Meme traces the first video to September 25, 2025. TikToker @princessdasha123 shared an iMessage chat with her mom about a man eating her snacks in the kitchen. The next day @levy.gabriella posted the same setup with her dad, and her video passed 1.9 million views in six days.

After that it moved fast. On September 29, @sarahmccullough_ hit 5.6 million views in three days. On the 30th, @lovedumplings666 reached 5.9 million views in two. The picture barely changed from one video to the next: a scruffy, tired-looking man on the couch or asleep in the bed. Hence the name people searched for, “AI homeless man prank”.

The first images came from the photo editing tool that Snapchat+ gives its paying subscribers. Once The Tab published a step-by-step guide on October 1, 2025, most people switched to Gemini, Google’s image generator. The trend crossed the Atlantic within days. This Swiss couple nearly had the police show up:

@aleundseija

Er het fast de polizei ahglüte willer ned dehei gsi isch😂🫠 #homelessman #ai #prank #switzerland

♬ Originalton – Ale & Seija 🤍

How does the prank work?

There are four parts. A real photo of your home, so the setting holds up. A stranger added by AI, with lighting and shadows that match. A boring text message that makes the scene believable. And finally the screenshot of the panicked answer, cut into a video with some ironic music.

The trouble is that it works too well. On October 6, 2025, Salem police in Massachusetts put out a statement: several parents had believed the intrusion was real and called 911. Officers responding had no idea it was a joke and treated the call as a burglary in progress. By October 10, NBC News had counted warnings from Oak Harbor (Washington), Round Rock (Texas), Brown County (Ohio), Dorset Police in the UK and Ireland’s Garda. Two juveniles were charged in Ohio. “This behavior is not a ‘prank’, it is a crime,” the Brown County sheriff’s office wrote.

ABC News covered the police warnings here:

@abcnews

The viral “AI homeless man prank,” which involves artificially generated images of a “homeless man,” has prompted police warnings over safety concerns, with officials calling it “stupid and potentially dangerous.” #news #homelessmanprank #prank #police

♬ original sound – ABC News

TikTok responded by adding an “AI-generated” label to the videos NBC flagged. Its Community Guidelines already require creators to label AI content that shows realistic scenes or people.

Why did it spread so fast?

Because the reaction is real. Viewers aren’t watching an AI image, they’re watching a dad type “CALL THE POLICE” in all caps. That kind of raw panic keeps people watching to the end, and TikTok pushes videos people finish.

It was also easy. You didn’t need editing skills: one photo, one prompt in Gemini, three screenshots. Anyone could copy the format in ten minutes, which is usually what turns a video into a trend.

The backlash came quickly. Salem police said the prank “dehumanizes the homeless”, and plenty of commenters agreed. In the UK, Dorset Police sent officers after “a call from an extremely concerned parent”, according to the BBC. If the point of your video is to grow your account, an angry comment section is not the audience you want.

A year on, the format survives in gentler forms: the “builder prank” (a workman who ripped his trousers and asks to borrow a pair, per The Tab on October 3, 2025), the shirtless plumber, and in September 2026 the ancestor photo prank, where the fake picture is of a long-lost ancestor.

How do you do an AI prank without getting in trouble?

1. Pick someone who knows your sense of humor: a partner, roommate or sibling. Skip older or anxious parents. They are the ones who call 911.
2. Drop the intruder: no stranger in the bed, no threatening figure. Keep it absurd and harmless, like the builder with ripped trousers, a giant dog on the sofa or a pop star in your kitchen.
3. Take your photo: daylight, a tidy room, landscape framing. The sharper the photo, the more real the addition looks.
4. Generate the image: upload the photo to Gemini and describe the addition precisely (“add a builder in work clothes standing by the door, natural light”). Run it two or three times and keep the best.
5. Send, then reveal quickly: two or three messages at most, then say it’s a joke. If they mention calling anyone, stop right away.
6. Ask before posting: the screenshot shows a family member’s name and reaction. Blur their number and profile picture.
7. Turn on the AI label when you post, as TikTok’s rules require. An unlabeled video can be removed or shown to fewer people.
8. Work the first hour: reply to comments and pin the funniest one. If you want to give the video an early push, a pack of TikTok likes added right after posting helps the algorithm test it on a wider audience.

Which hashtags should you use?

Videos in this format run under #aiprank, #prank and #pranktiktok. NBC counted more than 1,200 videos under #homelessmanprank on October 10, 2025, but that tag is now tied to police warnings, so leave it alone. For the newer versions, #builderprank or #ancestorprank are safer.

Since August 2025 TikTok only accepts five hashtags per video. Use two for the trend and one or two for your niche. Post in the evening, when your audience is on their phones, and compare your analytics after a week.

Can a brand use it?

Yes, but only the harmless version. A brand that fakes a break-in gets a PR mess and, in some places, a visit from the police. The ones that pull it off go absurd: a sofa from the new collection appearing in a customer’s living room, a courier turning up with twelve boxes of croissants, the store mascot sitting in a bathtub.

Two rules. The reveal happens inside the video, never in the comments. And the AI label stays on. A company account caught fooling its audience loses more than it gains. For a new account, a base of TikTok followers also makes the profile look credible when a video starts to travel.