🗂️ The production workflow

Every Short will follow this exact pipeline — no deviation. The workflow is fixed before the experiment starts so results are comparable across all 5 videos.

Topic selection via ChatGPT Ask ChatGPT to identify 5 high-retention AI-niche Short topics with a question hook format. All 5 topics locked before filming begins — no mid-experiment swaps based on early performance.
Script generation (30-second format) ChatGPT writes a 30-second script: 3-second hook → 22-second body (3 punchy points) → 5-second CTA. Word count target: 75–85 words. One iteration, no manual editing.
AI voiceover via ElevenLabs Paste script into ElevenLabs, select a consistent voice persona used across all 5 Shorts to build channel identity. Free tier audio downloaded. Watermark stays in (experiment rule).
Visuals — VidGen Studio + RunwayML VidGen Studio handles the main composition. RunwayML Gen-3 used for any motion sequences or transitions that need generated video. First-pass output accepted — no cherry-picking generations.
Thumbnail generation via Canva AI + Midjourney Midjourney generates a hero image based on the topic. Canva AI overlays the title text. Same font, same colour palette across all 5 to build visual brand consistency.
Upload with AI-written title, description & tags ChatGPT writes the YouTube title (A/B hook format), description (SEO-optimised, 200 words), and 10 tags. Uploaded manually — but copy-pasted from AI output with zero edits.

🎬 The 5 Shorts — planned topics

Topics chosen before the experiment starts. Locked in. Each one is a standalone 30-second Short with a question hook designed to maximise watch-through rate.

Short 01 · Week 1
"What AI tool is replacing Google Search right now?"
🔒 Pending
Hook strategy: Open with a claim that challenges viewer assumptions ("You're still using Google wrong..."). Target niche: productivity-curious audience already interested in AI. Expected to perform best for reach since the topic is broadly searched. Thumbnail: split-screen Google logo vs AI chat interface.
Script prompt (to be used)
Write a 30-second YouTube Shorts script on the topic: "What AI tool is replacing Google Search right now?" Hook in first 3 seconds — it must make the viewer feel they're missing something. 3 punchy body points, each one sentence. End with a question to drive comments. 80 words max. No filler. Spoken word only — no stage directions.
Tools: ChatGPT (script) → ElevenLabs (VO) → VidGen Studio (composition) → Canva AI (thumbnail)
Views: — · Likes: — · Comments: — · Retention: —
Short 02 · Week 1
"I built an app in 10 minutes with AI — here's what happened"
🔒 Pending
Hook strategy: Story-led with a surprising outcome to drive completion. This format ("I tried X") consistently performs well on Shorts. The AI will script a fictional but plausible story of a first-time user building with Google Antigravity — based on documented real outputs from the review. No fabricated claims.
Script prompt (to be used)
Write a 30-second YouTube Shorts script in the "I tried X" format about using an AI coding tool to build an app in under 10 minutes. Make it feel personal and surprising — not a product ad. Hook in 3 seconds, 3 punchy outcomes, end with a "would you try this?" CTA. 80 words max.
Tools: ChatGPT (script) → ElevenLabs (VO) → RunwayML (motion clips) → VidGen Studio → Canva AI
Views: — · Likes: — · Comments: — · Retention: —
Short 03 · Week 1
"3 AI tools that actually save you money (not just time)"
🔒 Pending
Hook strategy: Value-led list format. "Save money" appeals to a slightly broader audience than pure productivity content. This Short intentionally targets a different viewer profile than 01 and 02 to test whether the channel algorithm recommendation differs by topic. All three tools referenced are ones I've personally reviewed.
Script prompt (to be used)
Write a 30-second Shorts script: "3 AI tools that save you money, not just time." Each tool gets one sentence — name it, and say exactly what it replaces (and what that costs). Hook: a stat about how much people waste on software they don't need. CTA: "which one surprised you most?" 80 words max.
Tools: ChatGPT (script) → ElevenLabs (VO) → VidGen Studio (split-screen format) → Canva AI
Views: — · Likes: — · Comments: — · Retention: —
Short 04 · Week 2
"This AI thing is scary good — and nobody's talking about it"
🔒 Pending
Hook strategy: Mystery + exclusivity format. Designed to test whether the algorithm favours vague-but-intriguing hooks over specific-but-searchable ones. The content will cover a genuinely under-discussed AI capability (likely AI memory in agents or real-time reasoning). Not clickbait — the content must deliver on the headline.
Script prompt (to be used)
Write a 30-second Shorts script about an underrated or misunderstood AI capability that most people don't know about yet. Hook: "Nobody is talking about this AI thing, and it's kind of terrifying." Three sentences of substance. End: "would this worry you or excite you?" Do not be vague — the content must be specific. 80 words.
Tools: ChatGPT (script, with web browsing for current AI news) → ElevenLabs → RunwayML → VidGen Studio
Views: — · Likes: — · Comments: — · Retention: —
Short 05 · Week 2
"I let AI manage my whole week — this is what happened"
🔒 Pending
Hook strategy: Continuation of the personal experiment format from Short 02. If that Short performs well, this one should benefit from the same viewer profile. The "week managed by AI" story ties directly to Experiment 01 (Instagram 7-day), giving real data points to reference — the only Short with a cross-experiment tie-in.
Script prompt (to be used)
Write a 30-second Shorts script: "I let AI plan and manage my entire week — tasks, emails, content, even my meals. Here's what actually happened." Hook in 3 seconds. Three honest outcomes — include one thing that went wrong. CTA: "would you try a week like this?" 80 words, spoken only, no stage directions.
Tools: ChatGPT (script) → ElevenLabs (VO) → VidGen Studio → Canva AI (thumbnail)
Views: — · Likes: — · Comments: — · Retention: —
"The question isn't whether AI can make a YouTube video. It's whether YouTube's algorithm can tell the difference — and whether viewers care."

📊 What I'm tracking

YouTube Studio metrics, pulled 48 hours and 7 days after each upload. Tracking both absolute numbers and ratios:

Views — total and from Shorts feed vs external. Watch-through rate — the single most important metric for Shorts reach. Like rate (likes ÷ views). Comment rate — an indicator of content that provokes reaction. Subscriber conversion — views that result in a new subscriber. Impressions click-through rate — how compelling the thumbnail and title are before the video starts.

All raw numbers will be published in the results table once this experiment is live. No cherry-picking or excluding underperforming Shorts from the data.

⚠️ Known limitations going in

Brand-new channel disadvantage. A channel with zero subscribers starts with essentially no algorithm trust. Results may be artificially low compared to what an established faceless channel could achieve with the same content. I'll note this explicitly in the verdict.

ElevenLabs watermark on free tier. The audio will have an audible watermark. This may hurt retention if viewers find it jarring. Documenting whether watermarked AI voices affect watch-through rate is itself a useful data point.

VidGen Studio is still early. Based on my experience building VidGen Studio in the Antigravity experiment, the toolchain is powerful but has rough edges. I'm expecting at least one Short to have a quality issue I have to work around within the experiment rules.

❓ Frequently Asked Questions

What niche will the channel be in?

AI tools and productivity — the same niche as this site. It's the niche I know best and can write accurate scripts for without relying on the AI to be factually correct about domains it might hallucinate in. It's also a niche with genuine search demand on YouTube Shorts right now.

Will you show the actual channel?

Yes — when the experiment goes live, I'll link the channel directly from this page so you can watch the Shorts in their native context, see the real subscriber count, and compare that to the analytics I share. No hiding the channel if it underperforms.

Why YouTube Shorts and not long-form?

Shorts have a faster feedback loop — a video can get its first meaningful views within 24–48 hours, whereas long-form videos often take weeks to get distributed. For a 14-day experiment, Shorts give more data points. Experiment 03 (Micro SaaS) may include a long-form companion channel if the Shorts results are strong enough to justify it.

What counts as success for this experiment?

I'll be defining success tiers in the final verdict: Worth it = 1,000+ combined views across 5 Shorts with >40% average watch-through rate. Conditional = 300–1,000 views, showing the algorithm is distributing the content but retention is weak. Not worth it = under 300 views total, suggesting the AI-generated content isn't being served by the algorithm at all. These are pre-set thresholds — I'm not moving them after seeing results.

Can I replicate this experiment myself?

Yes, and I'd encourage it. All prompts used in each Short are published on this page before the experiment starts, so you can run the same pipeline on a different niche and compare results. If you do, share your numbers — I'll reference them in the verdict. The more data points across niches, the more useful the conclusion.