4 Prompts I Use to Find Viral Video Questions BEFORE I Open Any AI Tool

The short answer

Most creators open ChatGPT and ask “give me viral video ideas.” That’s why 99% of AI videos look the same — and why they tank.

I do the opposite: I find the question first, then I use AI to execute. Here are the 4 prompts I run in sequence before I touch any generation tool. Copy-paste them, swap the bracketed parts with your niche, and you’ll have a video concept in under 15 minutes that someone is actually searching for.

The same chain produced the reel that broke 1.4M views on my channel.

Table of contents

1. Why the question matters more than the prompt 2. Prompt 1 — Question discovery 3. Prompt 2 — Angle refinement 4. Prompt 3 — Hook generation 5. Prompt 4 — Visual concept 6. How to chain them properly 7. Real example: from question to shipped reel 8. FAQ 9. TL;DR

Why the question matters more than the prompt

The dirty secret of viral AI videos: the model is the easy part. The hard part is finding something worth filming.

I made 42 AI videos that averaged 89 views each before I figured this out. Same tools, same prompts, same edit pipeline. The difference between #42 (89 views) and #43 (1.4M views) wasn’t a better prompt. It was that I finally stopped asking “what looks cool?” and started asking “what question is my audience already typing into TikTok at 1am?”

Once you have a real question someone is asking, the rest of the workflow is almost mechanical. The 4 prompts below are how I find that question reliably.

These work in any AI assistant — ChatGPT, Claude, Gemini. I’ve tested all three. Slight variations in output but the framework holds.

PROMPT 1 — Question discovery

“` Act as a search-intent researcher for [your niche — e.g. AI video creation for beginners, sustainable fashion, indie game dev]. List 10 specific questions that beginners are asking right now on Reddit, YouTube comments, and Twitter, BUT where the top 3 Google/YouTube results are either outdated (older than 6 months), too technical, or miss the practical “how do I actually do this today” angle.

For each question, give me: (1) the exact phrasing a beginner would type (2) why current answers fail (3) the search volume signal (low/medium/high) based on how often you’ve seen it asked

No generic questions. No “what is X” basics. Find the questions where the top result is YouTube videos with under 5K views or blog posts from 2023. “`

PROMPT 2 — Angle refinement

“` Take this question: “[paste question from Prompt 1]”

Give me 5 different video angles to answer it, ranked by virality potential.

Each angle must include: (1) the core promise in 8 words or less (2) the contrarian or surprising element that makes someone stop scrolling (3) who this angle is NOT for (so the right audience self-selects) (4) the format that fits best (tutorial / breakdown / reaction / case study / list)

Reject any angle that sounds like a generic “how to” — I want the angle that makes people text the link to a friend. “`

The “text to a friend” criterion is also non-obvious. Viral content isn’t just clicked — it’s shared. The angle that gets a friend-share is the angle that wins the algorithm right now.

PROMPT 3 — Hook generation (first 3 seconds)

“` Write 10 spoken hooks for a short-form video with this angle: “[paste angle from Prompt 2]”

Constraints: (1) max 12 words per hook (2) no questions to camera, no “in this video,” no “let me show you” (3) each hook must create a curiosity gap OR make a specific claim with a number (4) written to be SPOKEN out loud in 2-3 seconds (5) mix 3 styles: bold statement, pattern interrupt, and counterintuitive truth

After the list, mark which 2 hooks would work best on TikTok vs YouTube Shorts and why. “`

The bold statement / pattern interrupt / counterintuitive truth mix matters because algorithms favor variety in your account’s content profile. If all your hooks are pattern interrupts, your channel feels samey. The mix builds range.

PROMPT 4 — Visual concept (the one image)

“` For a video with this hook: “[paste hook from Prompt 3]”

Describe the ONE visual that should appear in the first frame — the image that, if someone saw it as a thumbnail or scroll-stop frame, would make them watch.

Give me: (1) the scene in cinematic terms (subject, action, environment, lighting, camera angle) (2) the emotion it should provoke in under 1 second (3) what’s in the frame that contradicts expectations (4) a ready-to-use prompt I can paste into Higgsfield / MidJourney / Veo / Kling to generate it

No abstract concepts — describe what the camera literally sees. “`

The “contradicts expectations” requirement is doing the heavy lifting. A frame that matches expectations is a frame that gets scrolled. A frame that contradicts expectations earns the first second.

How to chain them properly

The four prompts must run in this exact sequence. Don’t combine them into one mega-prompt. The output quality drops sharply when you do.

Why sequential matters:

• Prompt 1 outputs 10 questions. You pick 1.
• Prompt 2 takes that 1 question and outputs 5 angles. You pick 1.
• Prompt 3 takes that 1 angle and outputs 10 hooks. You pick 1-2.
• Prompt 4 takes that 1 hook and outputs 1 visual concept.

Each step narrows the funnel. By the time you arrive at Prompt 4, the visual concept is already locked into something specific, opinionated, and findable by people who care about the question.

If you skip Prompt 2 and jump from question to hook, you get generic hooks. If you skip Prompt 3 and jump from angle to visual, you get cinematically beautiful frames with no narrative pull. Trust the chain.

Real example: from question to shipped reel

Here’s the chain I ran for my “1.4M views with zero cameras” reel:

• Contrarian element: most tutorials assume you film something first
• Not for: people who want fast/easy stock-footage AI
• Format: case study + tutorial hybrid

I generated that frame in Higgsfield Soul + did 3 follow-up frames in the same visual language. The reel was 47 seconds. 1.4M views in 12 days.

The framework didn’t make me a better filmmaker. It made me ask a question worth filming.

FAQ

Does the prompt chain work with ChatGPT, Claude, and Gemini equally well?

Yes, with slight differences. Claude tends to be most willing to give contrarian angles in Prompt 2. ChatGPT tends to be most useful for visual descriptions in Prompt 4. Gemini is middle-of-the-road. I run all three on the same input when I have a big video coming and pick the best output from each prompt.

How long does the full chain take?

10-15 minutes if you’re focused. The first time you run it, expect 25-30 minutes — you’ll spend extra time evaluating outputs. Once you’ve done it 5 times, you’ll be at the 10-minute mark.

What if Prompt 1 doesn’t return good questions?

Two fixes: (a) be more specific in the niche definition. “AI video creation” is too broad. “AI video creation for indie filmmakers on under $200/mo” is better. (b) ask for questions in a specific medium: “specifically questions asked on Reddit r/[subreddit] in the past 30 days.”

Do I need to subscribe to all the AI tools mentioned in Prompt 4?

No. Prompt 4’s job is to give you a ready-to-use prompt. You can paste it into any tool — your existing MidJourney, Veo, Kling, or aggregator like Higgsfield. The prompt is portable.

Can I use this for long-form YouTube (not just Shorts)?

Yes. The chain works for any video format. For long-form, modify Prompt 3: change “spoken hooks” to “spoken first 30 seconds — the cold open” and remove the 12-word constraint. The rest stays the same.

What if my niche has no obvious “viral” angle?

Some niches don’t have viral potential — that’s fine. Viral is one metric; ranking on Google, building authority, and converting to email/sales are equally valid. The chain still works for those goals; just replace “virality potential” in Prompt 2 with “conversion potential” or “ranking potential.”

Is there a paid version of this with more prompts?

Yes — the 4 prompts are step 1 of a 7-step system. The remaining 3 steps (editing prompts, hook templates, posting framework) are inside The 1M Views Blueprint ($27, 52 pages).

TL;DR

• The question matters more than the prompt. Spend 15 minutes finding it before touching any AI tool.
• 4 prompts, run in sequence: discovery → angle → hook → visual.
• Each step narrows the funnel. Don’t combine them.
• Real result: this exact chain produced a 1.4M-view reel on my channel.
• Works with any AI (ChatGPT, Claude, Gemini) and any video tool (Veo, Kling, MidJourney, Higgsfield).
• Total time: 10-15 minutes once you’re trained.

Get the prompts as a 1-page PDF (free)

If you want to keep this handy as a reference, the 4 prompts are available as a 1-page PDF download.

👉 Get The 4 Prompts (free, 1 page)

For the full 7-step system, that’s inside The 1M Views Blueprint ($27, 52 pages).

Related reading

• Sora Is Dead — Here’s What I’m Using Instead
• Veo 3.1 vs Kling 3.0 vs Seedance 2.0
• How to Make AI Videos That Actually Get Views in 2026
• YouTube’s New AI Monetization Policy 2026

Cinematic 16:9 YouTube thumbnail. Confident 28-year-old man with a dark well-groomed beard and black square-frame glasses, pointing his pinky finger toward his open mouth to show iced-out diamond grills engraved with the text "PROMPT INSIDE". Wearing an orange camo Stone Island jacket (Stone Island badge on the arm) over a black tee, a gold Cuban-link chain with a diamond cross, and a gold Rolex Day-Date. Dark luxury study / library background with floating glowing note cards numbered 1, 2, 3 and 4. Photorealistic, warm cinematic lighting, high contrast.