Why filming first is the most expensive mistake
A typical short-form video takes 90 minutes from idea to publish (script, record, edit, captions, thumbnail, upload). A long-form video takes 8-15 hours. Whether the topic resonates is mostly decided before the first frame is recorded.
The default creator workflow inverts this — film whatever feels exciting in the moment, then hope the algorithm rewards it. The data on this is brutal: about 80% of solo-creator videos underperform that creator's own median view count. Most flops were predictable.
A validation framework moves the decision from "after I felt inspired" to "before I committed an hour of production time." It is the highest-leverage 15 minutes of your week.
The validation arbitrage
Filming an unvalidated topic: 90 minutes invested before you know if it works. Validating a topic with AI: 2 minutes invested before you know if it works. Same outcome quality on average; 45× less wasted time on flops.
Step 1 — Capture raw ideas without judgment
You need a single inbox where every spark goes — comments, conversations, books you read, podcast snippets, search queries you almost googled. The friction has to be near zero, or ideas leak away. Voice memos, Apple Notes, a Telegram bot, anything that takes under 5 seconds to log.
Do not filter at capture. Filtering happens in step 3, with data. At the capture stage your job is to be a sponge — the more ideas you log, the better the AI clustering works in step 2.
Target: 20-40 raw ideas per week. If you are below 10, your inputs are too narrow (read more, talk to more people, watch competitors). If you are above 60, you are probably capturing observations rather than topic-shaped ideas.
Step 2 — Cluster with AI to find thematic patterns
Once a week, dump your raw idea log into Claude or GPT-4 with a prompt like: "These are content ideas from a creator in [niche]. Cluster them into themes. For each cluster, suggest a single sharp topic that would land all the related ideas." You will get 4-8 themed clusters from 20-40 raw ideas.
The clustering does two things: it surfaces the topic you keep circling without realizing it (a strong signal of authentic interest), and it merges adjacent ideas into a stronger single piece. Five mediocre ideas often clustered together produce one excellent piece.
Save the clusters as your candidate topic pool. You will only film 2-4 of them this week, so the pool gives you optionality. Run this clustering loop every Friday for 15 minutes.
AI clustering prompt (paste as-is)
- Role: You are a content strategist for a solo creator in [your niche, e.g. personal finance for millennials].
- Input: Here are 30+ raw content ideas I captured this week: [paste your list].
- Task 1: Cluster these into 4-7 themes. Name each theme.
- Task 2: For each theme, write the single sharpest topic that would land all the related ideas as one piece.
- Task 3: Score each cluster 1-10 on (a) likely audience interest, (b) my authority to make it, (c) freshness — would this feel new or rehashed.
- Output format: Markdown table with columns Theme, Sharp Topic, Audience, Authority, Freshness.
Step 3 — Score against external trend signals
AI clustering tells you what is hot inside your head. To find out what is hot in the world, run each clustered topic through two trend signals: search volume and platform-native search. Use Google Trends for 12-month interest curves, plus the platform you publish on (YouTube search, TikTok search, IG search).
A topic that is trending up in Google Trends AND has strong autocomplete suggestions on your target platform is a high-confidence yes. A topic that is flat on both is a maybe — it might still work for your specific audience, but the upside is capped. A topic trending down on both is a no, regardless of how exciting it felt.
You can automate the trend lookup with a custom GPT or a script that hits Google Trends + a YouTube search scraper. Or do it manually in 30 seconds per topic. Either way, the rule is the same: never film a topic that failed the trend gate without a strong qualitative reason to override.
Step 4 — Queue, do not commit
The validated topics go into a topic queue — a simple list ordered by score, with maybe 8-12 entries. You film from the top of the queue this week, then re-rank next week as new validation data comes in.
The queue is not a publishing schedule. It is an optionality buffer. When you sit down to record on Monday, you do not need to invent a topic — you pull from the top of the queue. When a current event breaks mid-week, you slot it in at position 1 and bump everything down. The queue absorbs chaos without breaking your output rhythm.
This is the workflow an AI agent can run for you weekly: ingest your raw idea log, cluster, score against trends, update the queue, send you a Friday summary. You spend 5 minutes reviewing instead of 60 minutes thinking.
Weekly 15-minute validation ritual
- Friday 4:00 — Export the week's raw idea log
- Friday 4:05 — Run the AI clustering prompt
- Friday 4:10 — Run each cluster through Google Trends + platform search
- Friday 4:15 — Re-rank the topic queue; tag top 3 for next week's record session
- Sunday — Glance at the queue, no decisions; just trust the process
- Monday — Open the queue, pull the top topic, start recording