Prompts to ship short-form tech content (YouTube Shorts, TikTok, Reels) without burning out. Hook-driven scripts, thumbnail copy, and the ruthless 'is this title clickbait or actually good' check.
Honest note — Algorithms change quarterly. These prompts encode the structural rules that have stayed stable (hook density, retention curve, CTR baseline) — they don't chase the latest 'YouTube secret hack' trend. Re-test if your channel data starts disagreeing with the suggestions.
When your retention drops 50% by second 3 because your hook is 'Hey guys, today we're going to look at...'.
Claude 4.7 Opus (2026-05)GPT-5 (2026-05)
Generate 25 first-second hook lines for a YouTube Short / TikTok on this tech topic:
Topic: <TOPIC>
Target viewer: <BEGINNER_DEV | SENIOR_ENG | NON_TECHNICAL | CAREER_SWITCHER>
What the video actually shows: <VIDEO_PAYOFF>
Rules:
- Each hook ≤ 9 words (must be sayable in <1.5s).
- No 'Hey guys', no 'In this video', no 'Today we'll'.
- At least 5 hooks each in these patterns:
* Contrarian ("Stop using X, do Y instead")
* Surprise-stat ("99% of devs get this wrong")
* Question-cliffhanger ("Why does <X> actually work?")
* Outcome-tease ("This 1 line saves $X/month")
* Identity-callout ("If you're still doing X, watch this")
After the 25, mark your top 3 with the reason (e.g., 'highest curiosity gap', 'targets the algo's clicked-but-didn't-finish signal').
TipTest 3 hooks across 3 uploads — the algorithm will tell you which pattern your audience responds to. Then double down on that pattern for the next month.
When your scripts are 90s long and you keep losing viewers at the 30s mark.
Claude 4.7 Opus (2026-05)GPT-5 (2026-05)
Write a 60-second YouTube Shorts / TikTok script on this tech topic. The script must keep retention >70% to second 50.
Topic: <TOPIC>
Value payoff (what the viewer learns): <PAYOFF>
My tone: <SOLO_NERDY | SOLO_CONFRONTATIONAL | INTERVIEW | SCREEN_RECORDING_VOICEOVER>
Structure (each beat is 1 line, with timing):
[0-2s] HOOK — promise the payoff or call out the target viewer.
[2-7s] STAKES — name what they lose if they don't know this.
[7-20s] SETUP — the wrong way / common belief, with 1 concrete example.
[20-40s] PAYOFF — the actual answer / technique, with a specific number or before-after.
[40-50s] SECOND HOOK — an unexpected twist or 'but here's the catch'.
[50-58s] LOOP CLOSE — a sentence that makes the viewer want to rewatch, OR a CTA that loops to the start.
[58-60s] CTA — follow / comment / link (only ONE — never multiple).
Word count: 130-150 words total (matches 60s at natural pace). No filler. Read it aloud to check timing.
TipIf a beat doesn't earn its seconds in retention, cut it. The 'STAKES' beat is the most-skipped — only keep it if you can make it visceral.
When you've got a candidate title and you're not sure if it's high-CTR or just shameless.
Claude 4.7 Opus (2026-05)
Audit this YouTube title for the clickbait-vs-curiosity tradeoff.
Title candidate: <TITLE>
What the video actually shows: <ACTUAL_CONTENT_SUMMARY>
Target viewer: <TARGET_AUDIENCE>
Score the title on:
1. **Curiosity gap (1-10)** — does it open a loop the viewer needs to close?
2. **Truth alignment (1-10)** — does the video actually deliver on the title's promise? (10 = perfect match, 1 = full bait-and-switch).
3. **CTR ceiling estimate** — given the topic + the gap, what's the realistic top CTR (e.g., 8-12%)?
4. **Retention risk** — will viewers who clicked feel cheated by second 15? Yes/No + reason.
Then rewrite the title 3 ways:
- **Honest**: same curiosity gap, zero exaggeration.
- **Edge**: maximum CTR without crossing into bait.
- **Bait**: what shameless looks like, for contrast (label it so I don't use it).
Final verdict: ship original / ship Honest / ship Edge.
TipThe 'Bait' variant exists so you can SEE where the line is. Ship Honest or Edge — never Bait. Bait kills channel trust over time even if it spikes one video.
Thumbnails need text that's readable at phone-feed size (~80px wide). This prompt forces the constraint.
Claude 4.7 Opus (2026-05)GPT-5 (2026-05)
Generate 10 thumbnail text overlays for this video. Each overlay = 2-4 words MAX (readable on a phone feed at ~80px wide).
Video topic: <TOPIC>
Thumbnail visual: <DESCRIBE_THE_IMAGE — e.g., my face shocked, a code screenshot with red arrow, a before-after split>
Tone: <CURIOUS | URGENT | OUTRAGE | TUTORIAL>
Rules:
- 2-4 words. Hard limit. If you can't say it in 4 words, the visual must say it.
- Numbers when possible ("3 mistakes", "$0 to $10k", "Day 1 vs Day 30").
- The text complements the visual — never describes it. If the thumbnail shows a shocked face, the text shouldn't say 'shocking truth'.
- No ALL CAPS unless the word is a number or acronym.
For each overlay, give me a 1-line reason it works (which curiosity loop or pattern interrupt it triggers).
TipRun the top 3 overlays through TubeBuddy A/B test rotation. The algorithm picks the winner faster than your gut does.
When you're stuck for what to film next. Your audience has already told you — you just haven't read your comments carefully.
Claude 4.7 Opus (2026-05)GPT-5 (2026-05)
I'll paste my recent video comments. Find me 5 high-potential video ideas hiding in them.
My channel niche: <NICHE>
My typical video length: <SHORTS | 5_MIN | 10_PLUS>
Comments:
<PASTE_50_TO_100_COMMENTS>
For each of 5 ideas, give me:
1. **The seed comment(s)** — quote them.
2. **The video title** (one specific, hook-forward title).
3. **Why it's high-potential** — pick from: (a) multiple viewers asked the same question, (b) someone described a problem they'd pay to solve, (c) controversy or disagreement in the comments, (d) someone wrote a long detailed comment (= the topic engages emotionally).
4. **Realistic CTR + retention forecast** for your niche (low/mid/high — be honest, not optimistic).
5. **What format** (talking head, screen recording, interview, etc.).
Ignore generic positivity ("great video!") — only mine for content signals.
TipRun this monthly. Your comment section is the world's cheapest market research.
How to use these prompts
Each prompt has placeholders in <ANGLE_BRACKETS> — fill them in before pasting. Copy the prompt with the button, paste into Claude, ChatGPT, Gemini, or any chat-UI'd LLM.
Why "model tested" dates matter
LLMs improve and regress with every release. A prompt that worked on Claude 3.5 may need rewriting for Claude 4. The dates show when each prompt was last verified — anything older than 6 months should be re-tested before depending on it.
Found a better prompt?
Hit contact and share — we keep prompts that beat ours.