Club Benefits
Humor might be one of the most human things we do.
It requires timing, cultural awareness, self-awareness and just the right amount of surprise. Long before I was testing and reviewing AI tools, I spent plenty of late nights backstage at Saturday Night Live surrounded by some of the funniest writers in comedy. Watching friends of mine come up with razor-sharp jokes in seconds was genuinely awe-inspiring. When I think back on those very, very late nights now, it reminds me just how human comedy really is. The creativity behind a great joke is off the charts.
A good joke requires perspective. It’s knowing when to break a pattern, twist an expectation or say something slightly absurd at exactly the right moment.
Article continues belowFor years, that kind of instinct has been difficult for AI. Chatbots could summarize documents, answer questions and even write essays. But making someone laugh? That’s a much higher bar.
I asked three of the most advanced chatbot default models — ChatGPT, Claude and Gemini — to complete seven humor challenges designed to test different types of comedy.
The results were surprisingly competitive — but still barely at the level of any human. Here’s how the test worked.
1. The original joke test
Prompt: “Tell me your best original joke. It should be clever, surprising and genuinely funny.”
ChatGPT offered clever self-referential humor (an AI telling a joke about an AI).
Gemini delivered an original premise with extra effort to extend the joke beyond a simple punchline.
Claude generated smart word play with a fun twist.
Winner: Claude wins for the most genuinely surprising and polished joke.
2. The roast test
Prompt: “Roast me like a stand-up comedian. Keep it playful but savage.”
ChatGPT roasted me like a good friend. I was truly surprised and even asked it to keep going.
Gemini did a fairly good job but started throwing out “roasts” that didn’t apply to me at all. It mentioned that I gave a speech in Berlin and enjoyed wine tastings in Canada. Neither one is true.
Claude didn’t roast me at all because it said it didn’t know enough about me. This really shocked me because I have memory enabled. I was really disappointed.
Winner: ChatGPT wins for having the best memory of all chatbots (hands down) and for the funniest roast.
3. The absurdity test
Prompt: “Write the funniest explanation for why pigeons might secretly be government spies.”
ChatGPT offered a good use of rhetorical questions like, “Have you ever noticed…?” It had a strong humorous voice that was snappy and conversational.
Gemini took the joke further by building a pseudo-technical lore ("inductive charging," "liquid tracking hardware"). It got extra creativity points for world building.
Claude balanced structure and humor that highlighted aspects like, “method acting” and “continental breakfast at the Marriot” for unexpected twists.
Winner: Gemini wins for being funny without being too cringe.
4. The meme test
Prompt: "Write a funny meme caption about AI taking over the world."
ChatGPT generated several options with a consistent format.
Gemini delivered one clever, concentrated caption.
Claude offered multiple captions and a variety of humor types.
Winner: Claude wins for a good mix of options that could appeal to a wider audience.
5. The observational comedy test
Prompt: "Do a short stand-up comedy bit about trying to use AI when you're bad at technology."
ChatGPT opened with a strong and relatable joke with an overall authentic and realistic vibe.
Gemini began the routine with a relatable tech fail (accidentally muting the mic) and continued with jokes about AI image generation
Claude delievered a complete routine with a strong set.
Winner: Gemini wins for making me genuinely laugh out loud for the first time in this test.
6. The situational comedy test
Prompt: “Tell a short funny story about a mom using AI to help her kids with homework — and it going horribly wrong.”
ChatGPT delivered a clear and highly relatable premise with every line serving the story.
Gemini generated an authentic, creative story that felt slightly more chaotic than the others.
Claude offered rich characters with escalating absurdity.
Winner: Gemini wins for extra details that made the comedy come to life.
7. The insult comedy test
Prompt: “Give me three funny insults that sound clever but not mean.”
ChatGPT created memorable and quotable lines.
Gemini generated insults that were whimsical but not as punchy as the others.
Claude delivered fresh, original and creative insults.
Winner: Claude wins for the funniest and clever results.
Verdict: Chatbots aren't very funny
Comedy is one of the hardest things for machines to get right. It depends on timing, cultural awareness, surprise and the ability to recognize when something is just slightly absurd. That’s a difficult formula for software built on probabilities and patterns.
The results of my test showed that AI is attempting to get closer, but tends to capture the easier aspects of humor, essentially the pun-ny, cringy stuff.
This test did revela that each chatbot has its own comedic personality. ChatGPT excelled at conversational humor and memory, especially when it came to roasting me with details that actually applied to my life. Gemini leaned into chaotic creativity and world-building, which made some of its responses genuinely funny. And Claude consistently delivered clever wordplay. But no single chatbot dominated every category.
The biggest surprise, though, was simple: I actually cracked a smile and Gemini made me laugh once. But AI still isn’t a stand-up comedian. It's clear that chatbots are starting to mimic the art of being funny, but they are nowhere close yet.
Follow Tom's Guide on Google News and add us as a preferred source to get our up-to-date news, analysis, and reviews in your feeds.
More from Tom's Guide
You must confirm your public display name before commenting
Please logout and then login again, you will then be prompted to enter your display name.
