ChatGPT vs Gemini vs Claude Make Rocket League

By Minimunch · Tech · 199.8K views · 9:14

The teardown in brief

What's working

• Natural authentic reactions create genuine entertainment — your vocal energy spikes at real moments of surprise (0:54 'that's a nice shot', 4:59 'woah look at how smooth that is') feel unscripted and make the viewer feel like they're experiencing discovery with you, not watching a scripted review
• Clear visual comparison structure — the three-AI format gives the video natural chapters and built-in stakes (which AI wins?), making it easy for viewers to track progress and stay oriented
• Claude section delivers satisfying gameplay climax — the final 3 minutes (6:00-8:21) shift from testing to actual competitive play, which gives the video an emotional peak beyond just 'this one is better'. The close match against the bot creates real tension

What's costing attention

• Front-loaded sponsor before proving value — asking viewers to sit through a 20-second product explanation before showing ANY results breaks the viewing contract. The audience clicked for AI comparison, not a tutorial on GenSpark features
• Repetitive problem-fix cycles without variation — the ChatGPT section has 4-5 iterations of 'bug discovered → fix requested → test again' that follow identical beats. By iteration 3, the viewer can predict the pattern, which kills engagement
• Gemini section lacks entertainment value — unlike ChatGPT (incremental improvement) or Claude (best results + gameplay), the Gemini section is just 'this is broken' for 57 seconds with no arc, no humor, no payoff. Bad results can be entertaining if framed as comedy or compressed into a quick montage, but drawn-out mediocrity is just boring

The first 30 seconds

ChachiT, Gemini, and Claude. Today, we're putting these AIs against each other to see who can make the best Rocket League clone from scratch. And we're going to be building all of these games inside of GenSpark AI. Genpark is an all-in-one AI workspace 2.0 with features like super agent, AI slides, AI sheets, AI develo

The concept is clear within 5 seconds ('testing AIs to make Rocket League clone') and you reaffirm the thumbnail promise immediately, which is good. But you then spend 20 seconds (0:05-0:25) on GenSpark feature breakdown BEFORE showing any results or action. For a tech/gaming audience expecting immediate comparison, this sponsor wall delays the actual content start to 0:36. The hook isn't confusing — it's just SLOW. You're asking for patience before proving the video delivers. Predicted drop: 28% by 30 seconds (vs 15-20% baseline for a strong Tier 1 hook). The first action moment (ChatGPT game appearing) doesn't arrive until 0:42. With a cold open showing Claude's best results for 10 seconds, then cutting back to 'here's how I made this,' the drop would be 18-20% instead.

Where viewers drop

0:05 — Sponsor Before Action (critical)

The video opens with a 3-second premise, then immediately pivots to 20 seconds of GenSpark sponsor explanation before showing ANY results or action. The viewer clicked for AI competition, not a product pitch. They're still deciding whether to invest 9 minutes, and you're asking them to sit through a commercial first.

Why it matters — For a tech/gaming audience aged 16-25, this creates a 25-30% drop in the first 30 seconds on top of the baseline packaging drop. High-energy audiences have near-zero patience for pre-content setup. The sponsor feels like a gatekeeper, not a natural integration.

1:07 — Repetitive Iteration Fatigue (moderate)

The ChatGPT section becomes a loop: test → problem discovered → fix requested → test → same problem or new problem → fix again. This happens 4-5 times (goal doesn't work, squares everywhere, black screen, goal still doesn't work, multi-point bug). By the third iteration at 1:30, the pattern is predictable and the viewer knows another cycle is coming.

Why it matters — Repetition is the #1 retention killer platform-wide. When the viewer can predict the next beat ('he's going to find another bug and ask AI to fix it'), they start checking their phone. You lose 5-8% extra per repetitive cycle beyond the second one. This section runs for 2+ minutes in a predictable loop.

3:34 — Gemini Dead Zone (moderate)

The entire Gemini section (3:33-4:30) is 57 seconds of 'this is terrible' reactions. The game doesn't work, the physics are broken, the opponent is buggy. There's no arc — just disappointment. Unlike ChatGPT where each iteration IMPROVED something, Gemini never gets good. The viewer experiences this as dead runtime watching something fail.

Why it matters — Viewers tolerate struggle if there's PROGRESS. Struggle without progress feels like wasted time. This section has no payoff and no forward momentum — it's just proving Gemini is bad. For a comparison video, the 'loser' section needs to be brief or at least entertainingly bad. This is just flat bad, which isn't entertaining.

8:49 — Outro Sponsor Hijack (mild)

The winner is declared at 8:48 ('Claude is the obvious winner'), which is the natural emotional climax and end point. But instead of ending there or adding a brief personal sign-off, the video extends for another 25 seconds of GenSpark feature listing (unlimited 2025, model names, etc). The story is OVER at 8:48, so everything after feels like an ad that's keeping you from clicking away.

Why it matters — The outro drop is the steepest in any video — once viewers sense the main content is done, they leave fast (10-20% drop in 30 seconds). Adding a sponsor read AFTER the emotional conclusion accelerates this. You're asking viewers to stay for a commercial when they've already got their payoff. End-screen click-through rate also drops because they're leaving during the sponsor, not the actual outro.

How the video is built

• 0:00 Setup & ChatGPT Test — Hook, sponsor intro, and full ChatGPT iteration cycle — building and debugging until the game works
• 3:23 Gemini Test (Brief Failure) — Gemini attempt fails quickly — broken physics, bad AI, no improvement
• 4:31 Claude Test & Gameplay Climax — Claude produces the best game, leading to competitive gameplay against AI bot with close match tension
• 8:49 Outro & Sponsor Close — Winner declared, sponsor feature list, sign-off

What any creator can steal

• The sponsor blocks action for 20 seconds before the viewer sees ANY results
• ChatGPT section has 4 identical iteration cycles that feel predictable by 1:30
• Gemini section is 57 seconds of flat failure with no arc or entertainment value
• Stakes disappear for 90 seconds during ChatGPT iteration cycles
• Outro extends 25 seconds past the emotional climax with sponsor feature listing
• Establish a 'montage rule' for repetitive processes: show the first iteration in full to establish the pattern, then compress subsequent identical cycles into 15-20 second montages set to music. This works for any video where you're iterating on a process (debugging, testing variations, making multiple attempts). Saves 60-90 seconds per video without losing the 'it took effort' narrative.