Pain point: your group wants a funny TikTok or Reel tonight. Someone says "let's film something." Then everyone freezes, because writing a scene feels awkward, slow, and somehow more work than scrolling.
Fix: let ChatGPT write a short script. You read it once, film one deadpan take, and put this text on screen early: we had ChatGPT make us a script.
That is the whole trend. It is popular because it solves a real creator problem: funny content without a writers' room. Below you get the intent behind the joke, a 5-minute workflow, a big pasteable prompt pack, filming rules that protect the humor, and the mistakes that kill it.
What people are actually trying to do
Most people who search this trend are not looking for "AI theory." They want:
- A script they can shoot tonight
- Lines that sound a little weird (so deadpan works)
- A format that works with 2 people, not only a big friend group
- Clear rules so the first take does not flop
The user pain is social, not technical. Nobody wants to be the person who "kills the vibe" with a boring idea. ChatGPT becomes the safe third party that wrote the awkward lines - so your group can commit without feeling self-conscious.
Do this in 5 minutes
- Open ChatGPT (app or web).
- Paste one prompt from the pack below (match your headcount).
- Read the dialogue once. Assign lines. Do not rewrite the weird parts.
- Film one vertical take, totally serious.
- Add on-screen text in the first 2 seconds: we had ChatGPT make us a script - then post to TikTok or Reels.
Keep it 30-60 seconds. Flatter delivery usually wins.
Mini setup that makes night-of filming easier
- Put phones on Do Not Disturb
- Stand in a quiet corner with even light (window light or one lamp is enough)
- Decide who holds the phone before you open ChatGPT
- Agree on one rule: nobody breaks character, even if a line is ridiculous
If you spend more than five minutes "improving" the script, you are already drifting away from the trend.
Why this works (so you do not mess it up)
The joke is not "AI is random." The joke is commitment.
ChatGPT often writes lines that sound oddly formal, too dramatic, or slightly off for a normal friendship. You say those lines like a courtroom drama. Viewers laugh at the collision - then laugh again when they notice AI wrote it.
Three ingredients make the format land:
- Trivial conflict, serious tone (dishwasher, thermostat, bill split)
- Dialogue-only scripts (easy to memorize in one pass)
- Label honesty (the on-screen text sets the joke frame)
If you laugh mid-line, polish the script into natural speech, or cut the take into pieces, the trend usually dies. The awkward sincerity is the point.
Copy-paste prompts (use these)
Each prompt asks for dialogue only so you can read and shoot fast. Paste, generate, assign speakers, go.
2 people (best starter pack)
Write a 40-second two-person scene where one friend breaks up with the other over how they load the dishwasher, but treat it like a real breakup. Dialogue only, label speakers, no stage directions. End on the most dramatic line possible.
Write a 45-second two-person scene where roommates negotiate the thermostat like a hostage situation. Dialogue only, label speakers. End with an unexpected twist.
Write a under-60-second scene where a customer tries to return an obviously used item and the worker refuses with escalating corporate politeness. Dialogue only, two speakers. End with an outrageous customer demand.
Write a 40-second two-person scene where one person asks to borrow $7 for coffee and the other responds like this is a high-stakes financial negotiation. Dialogue only. End with a ridiculous condition.
3-4 people (group energy)
Write a under-60-second scene where three friends stage an intervention because a fourth friend is texting their ex too much. Every argument is overblown. Four speakers, dialogue only, label names. End with a ridiculous ultimatum.
Write a under-60-second family dinner scene where someone confesses a tiny secret (wrong turn signal years ago) and everyone reacts like it is shocking news. Four speakers, dialogue only.
Write a under-60-second scene where a friend group splits a restaurant bill to the exact cent like a legal case. Four speakers, dialogue only.
Write a under-60-second group chat read-aloud scene where friends argue about who left someone on read, but deliver every line out loud like a drama. Four speakers, dialogue only, no stage directions.
Work / everyday (very shareable)
Write a under-60-second first-day job scene where an icebreaker goes wrong because of one oversharing answer. Three coworkers, dialogue only, no stage directions.
Write a under-45-second mock performance review where a manager gives formal feedback about how someone stirs coffee. Two speakers, dialogue only. End on the most bureaucratic line.
Write a under-50-second scene where two coworkers argue about the office microwave etiquette like it is a company policy crisis. Dialogue only. End with a fake "escalation to HR" line.
Prompt upgrades when the first draft is bland
Ask ChatGPT for a second pass with one of these add-ons:
Regenerate the same scene, but make one speaker oddly formal and add one absurd detail (a fake statistic, a horoscope argument, or a corporate merger metaphor). Dialogue only.
Same speakers and conflict. Cut 20% of the words. Keep the weirdest line. Dialogue only.
Rewrite with stronger ending. Last line should feel like a soap-opera cliffhanger about a tiny household problem. Dialogue only.
Prompt tip: keep ChatGPT's weird wording. If a line feels unnatural, that is usually the funny part.
How to deliver lines (this is where most videos fail)
Reading is not enough. Delivery is the difference between "cute idea" and "I need to share this."
- Speak a little slower than normal
- Keep eye contact with the other person, not the camera (unless you are solo)
- Do not smile after your own joke
- Match energy: if one person goes theatrical, the other should stay stone-faced
- Leave a half-second pause before the last line
Solo version tip: hold your phone at eye level, glance at the script once, then look into the lens and deliver like you are in a serious PSA about a ridiculous topic.
Duo tip: stand shoulder-to-shoulder or face each other in a tight frame. The smaller the physical space, the funnier the seriousness often feels.
Filming checklist (save this)
- Vertical phone, eye-level, everyone in frame
- One read-through only (do not over-rehearse)
- One continuous take (no jump cuts)
- Deadpan faces the whole time
- On-screen text in the first 2 seconds
- 30-60 seconds total
Optional polish that still fits the trend:
- Soft natural light
- Quiet room (no TV in the background)
- Caption the final line if the audio is messy
- Keep original audio - do not drown it in loud music
What good vs flat looks like
Good: a dishwasher "breakup" where both people sound devastated, and the last line is something ChatGPT would never say in real life, delivered with total sincerity.
Flat: the same scene, but someone giggles, says "wait, that is stupid," then rewrites the line into normal speech. The audience never gets the AI collision.
Good: a bill-split argument that ends with a fake legal threat about appetizers.
Flat: a five-minute script with too many characters and no ending.
Use that as your quality bar before you post.
Mistakes that kill the joke
- Breaking character
- "Fixing" AI lines so they sound natural
- Scripts longer than a minute
- Skipping the "ChatGPT wrote this" text
- Overacting instead of sincerity
- Filming five takes until the energy dies
- Choosing a conflict that is actually mean about a real person in the group
Keep it kind. Roast situations, not people who did not consent.
A sample night-of workflow (so it feels doable)
Here is a realistic 12-minute loop you can follow with friends:
- Minute 0-2: pick headcount and one conflict from the prompt pack
- Minute 2-4: generate in ChatGPT, skim once, assign names
- Minute 4-6: one quiet read-through (phones down, no rewriting)
- Minute 6-9: film take one
- Minute 9-10: if someone broke, film take two only - then stop
- Minute 10-12: add the on-screen text, post, move on
Stopping after two takes matters. The trend rewards freshness more than polish. If you chase a "perfect" version, you usually sand off the AI weirdness that made it funny.
Caption ideas that match the format
- we had ChatGPT make us a script
- ChatGPT wrote this. We stayed serious.
- AI wrote the fight. We just performed it.
- dishwasher discourse, ChatGPT edition
Keep captions short. Let the video carry the joke.
Variations that still feel like the same trend
- Duo confession: two people, one tiny secret, huge reactions
- Customer service: corporate politeness vs absurd demand
- Roommate diplomacy: thermostat, dishes, shared fridge
- Office theater: microwave rules, standup meetings, fake HR
- Family table: tiny confession, oversized drama
If you invent a new variation, keep the three constants: short runtime, deadpan delivery, and clear "ChatGPT wrote this" labeling.
FAQ
What is the "we had ChatGPT make us a script" trend?
Friends ask ChatGPT for a short scene, then act it out deadpan in one take and label it as AI-written. The humor comes from serious delivery of slightly unnatural dialogue.
Do I need a big group?
No. Start with two people. Many of the strongest versions are duos.
What if the script is not funny?
Regenerate with a more specific trivial fight, or add one absurd detail. Do not rewrite the awkwardness away.
Can I use Instagram Reels too?
Yes. Same vertical one-take format and the same on-screen text.
Do I need premium ChatGPT?
Usually no. Any ChatGPT chat that can return a short dialogue script is enough for this format.
Should I tell ChatGPT to "be funny"?
You can, but "be funny" often makes it try too hard. Better: give a tiny conflict + serious tone + short runtime + dialogue only.
Soft next step on Toolbit
Want more AI tools for scripts, short video, or creative workflows? Browse Toolbit.ai, the AI tools directory, and categories.
Practical guide to the public "we had ChatGPT make us a script" TikTok/Reels format (2026). Keep it kind - roast situations, not people who did not consent.
Toolbit AI - Team · Last updated: September 11, 2026
