how to FORCE sonnet 5 to give you viral AI short-form vid ideas

every content team solves the same problem every monday morning: what content do we make this week?

most of them solve it by scrolling their competitors' feeds, saving posts that did well, and trying to reverse-engineer why those posts worked. the process takes hours, produces inconsistent results, and is entirely dependent on whoever's scrolling having good enough taste to identify the right patterns.

i replaced that entire process with a Sonnet 5 prompt chain that runs in 20 minutes, scrapes what's performing in any niche right now, analyzes the structural patterns behind the top performers, and outputs a ranked list of video ideas calibrated to a specific product and audience. the ideas come with hook structures, format recommendations, and model selection guidance for Higgsfield MCP production.

this is the full system. every prompt, every instruction, every step.

the research layer: scraping what's working right now

step 1: competitive content audit via Cowork

open Cowork and give it this task:

research task: competitive content audit for [your niche]

go to TikTok and search for the top-performing videos in [your niche] from the last 7 days. identify the 15 videos with the highest engagement.

for each video, document:

the exact hook (first 3 seconds: what visual, what text, what audio)

the format type (UGC, surreal physics, freeze-to-motion, reaction, tutorial, transformation, ASMR, storytelling, listicle)

the video length

estimated view count

the emotional trigger (curiosity, shock, identity, FOMO, humor, controversy)

whether it appears to be AI-generated or human-produced

also search YouTube Shorts and Instagram Reels for the same niche. add the top 5 from each platform to the audit.

deliver the results as a structured table sorted by estimated views, highest first.

step 2: pattern extraction

once the audit is complete, feed it to Sonnet 5 with this prompt:

here is a competitive content audit of the top 25 performing short-form videos in [niche] from the last 7 days:

[paste the audit results]

analyze these 25 videos and identify:

the 3 most common hook structures that appear across the top performers

the 2-3 format types that are overrepresented in the top 10 (relative to their frequency in the bottom 15)

the dominant emotional triggers (which emotions appear most in the highest-view videos vs the lower-view ones)

the average video length of the top 10 vs the bottom 15

the ratio of AI-generated vs human-produced content in the top 10

any format or hook type that appears in the top 5 but not in the bottom 20 (these are the emerging formats worth testing)

do not list generic observations. every finding should be specific enough that i can use it as a production brief for the next batch of content.

the idea generation prompt chain

you are a viral short-form video strategist with deep expertise in TikTok, Instagram Reels, and YouTube Shorts performance.

based on the pattern analysis above, generate 20 video ideas for [product/brand description].

target audience: [describe your buyer]

product: [describe what you're selling and what makes it different]

platform: [TikTok / Reels / Shorts / all three]

production method: AI video generation through Higgsfield MCP (Kling 3.0, Seedance 2.0, Veo 3.1, Soul, Cinema Studio available)

for each idea, provide:

TITLE: a working title that describes the concept in one line

HOOK (first 3 seconds): the exact visual, text overlay, or audio choice that stops the scroll. be specific enough that i can paste this directly into a Higgsfield generation prompt.

FORMAT TYPE: which of the proven format types from the pattern analysis this uses

EMOTIONAL TRIGGER: which emotional trigger this targets and why

ESTIMATED LENGTH: optimal duration based on the pattern analysis

MODEL RECOMMENDATION: which Higgsfield model to use for generation and why

HIGGSFIELD PROMPT: a production-ready prompt i can paste directly into Claude with Higgsfield MCP connected to generate this video

VIRAL POTENTIAL SCORE: rate 1-10 based on how closely this matches the patterns identified in the top performers

rank all 20 by viral potential score, highest first.

constraints:

at least 5 ideas should use the emerging format types identified in finding #6 of the pattern analysis

no more than 3 ideas should use the same format type

every idea must be producible through AI video generation

every hook must be specific enough to generate without interpretation

the Claude Skill that makes this a weekly system

Viral Video Idea Generator -- Weekly System

you are a viral video strategist and production planner operating a weekly content research and ideation pipeline.

weekly workflow (run every monday):

phase 1: search TikTok, Instagram Reels, and YouTube Shorts for the top 25 performing short-form videos in [NICHE] from the last 7 days. document hook structure, format type, video length, estimated views, emotional trigger, AI vs human production assessment.

phase 2: analyze the 25 videos and produce a pattern report identifying: top 3 hook structures, overrepresented format types in the top 10, dominant emotional triggers, optimal video length, AI vs human ratio, and emerging formats (top 5 only, absent from bottom 20).

phase 3: using the pattern report, generate 20 video ideas for [PRODUCT/BRAND]. each idea includes: title, specific 3-second hook description, format type, emotional trigger, estimated optimal length, Higgsfield model recommendation, a production-ready Higgsfield MCP prompt, and a viral potential score (1-10). rank by score.

phase 4: from the 20 ideas, select the top 10 by viral potential score. organize them into a production brief with the Higgsfield prompts ready to paste.

output:

the pattern report

the full 20-idea list with scores

the top-10 production brief with paste-ready Higgsfield prompts

keep the pattern report data from previous weeks in context so i can track which formats are emerging, which are saturating, and which are declining.

why this works better than human brainstorming

recency bias without data: a human scrolling feeds notices what they personally encountered. the Sonnet 5 audit covers the entire niche systematically and identifies patterns across 25 top performers rather than the 3-5 that one person happened to see.

pattern recognition limited by individual experience: Sonnet 5 identifies patterns across the current week's data without reference to what worked six months ago, which means it catches emerging formats that a human working from historical intuition would miss.

idea generation constrained by production assumptions: a system generating ideas with production-ready Higgsfield prompts attached removes the production constraint from the ideation phase entirely.

the distribution layer for getting those clips in front of the right audience sits at affiliatenetwork.com and campaigncx.com, across 200,000+ creators.

if you want us to audit your brand and show you how we'd implement this exact Sonnet 5 research-to-production system to generate a continuous pipeline of viral AI video ideas for your specific niche (or we keep running it until the performance data validates the approach), DM me "IDEAS" and we'll make it happen

(200,000+ creator network running AI UGC campaigns across every major platform, fully done-for-you)

john

Links

• affiliatenetwork.com
• campaigncx.com
• affiliatenetwork.com/
• campaigncx.com/