Battle of the AIs: my 9-year-old judged ChatGPT, Gemini and Claude

Step-by-step process

The test has one rule at its core: every AI gets the same prompt and the kid judges the output. You are not testing which one sounds smartest to you. You are testing which one lands with an 8-year-old.

Step 1: Start with a real question

Use a question your kid asked on their own. Ours came up because of the flight home. A real question gives you honest feedback: your kid wants the answer, so their attention tells you the truth about each output.

Step 2: Set the ground rules

Five rules kept the test fair:

• Free tier on all three. It matches what most parents have.
• Fresh chat in each app, incognito where possible, so no memory or custom instructions help one app.
• Identical prompt, pasted word for word.
• First answer stands. No re-rolls, no "make it better."
• Blind judging. Call them A, B and C. Kids have brand opinions too.

Step 3: Paste the prompt

The exact prompt I gave all three:

My 8-year-old son asked me how airplanes fly. We're flying from Colombia to Ireland next week and he's really curious about it. Explain it in a way he'll really understand, and make me something visual to help — whatever you think works best.

Note the open brief at the end. I did not tell them what to make. The gap between what each one chose to make is the whole story.

Step 4: What ChatGPT made

ChatGPT worked logged out, in an incognito window. Zero barriers, the only one of the three.

The answer was a script for me to read to him, with lines like "imagine you're sticking your hand out the car window." Solid explanation. It corrected the number one misconception head-on: engines don't hold the plane up, wings do.

The "something visual"? An ASCII diagram of the four forces, drawn with keyboard characters and emoji. Fine on a screen. Nothing a kid would call a picture.

It also suggested the classic blow-over-a-strip-of-paper experiment. More on how that went below.

Step 5: What Gemini made

Gemini refused to answer until I signed in with a Google account. Logged out, you get a login wall, not an answer.

Once in, it gave a tidy four-forces explanation in a teacher voice. It ignored the Colombia-to-Ireland detail completely. Then came the strange part. The answer said "you can use the image below as you explain" and later added "if you look closely at the image, you'll see the wings are curved on the top."

There was no image. I scrolled. I captured the full page, top to bottom. Text only. Gemini described its own visual aid, built part of the lesson around it, and never made it.

💡 Parent Insight: This is the hallucination lesson in miniature. Picture yourself reading the answer aloud, telling your kid to "look closely at the image," and pointing at a wall of text. Check the output before the kid sees it. Every time.

Step 6: What Claude made

Claude (free tier, account required) read "make me something visual" and built an interactive web page in its side panel. Not a description of a visual. The visual. (This is the same trick behind the three Claude Artifacts projects we tried before.)

The page had a wing-angle slider he could drag to tilt the wing and watch the lift change, a "Take off!" button, and a facts section about our actual flight: the Bogotá-to-Ireland arc, cruising higher than Everest, crossing the Atlantic. It closed with a tip to pick a window seat over the wing.

One more detail worth knowing: the page works offline. He gets to play with it on the flight itself.

💡 Parent Insight: ChatGPT and Gemini both handed me a script and homework. Claude handed my son a toy. That difference, being told something versus doing something, decided the whole test.

Step 7: Let your kid judge

Show each output as-is. No fixing, no explaining on the AI's behalf. Then ask three questions per app:

• "Now you explain it to me." Comprehension is the killer metric.
• "Was that fun or boring?"
• After all three: "Which one would you show your friends?"

His reactions, in order:

• Gemini: confusing. The answer kept pointing at an image he could not find, because it did not exist.
• ChatGPT: the paper experiment lost him. "Hold it under your mouth so it hangs down" left him holding a sheet of paper with no idea where to put it. Adult-obvious is not kid-obvious.
• Claude: he dragged the slider himself, made the plane take off, and then explained lift back to me. Hands-down winner, his call, no hesitation.

His verdict on the winner, word for word:

"I can't wait until takeoff and I bags sitting at a window! Hopefully we will be near the wing so I can take a closer look"

Read that back against Claude's answer. The window seat over the wing tip came straight from the page he played with. Comprehension turned into anticipation. He was not repeating a fact. He was making a plan.