I built a pipeline that cross-references 10+ video reviews to surface what AI creators actually agree on
I built a pipeline that cross-references 10+ video reviews to find what AI creators actually agree on
Earlier this year I was trying to decide which AI coding tool to invest time in. Cursor? Claude Code? Copilot? Codex?
I did what most people do — watched a bunch of YouTube reviews. After 10+ videos I had a pile of conflicting opinions. One creator loved Cursor's IDE. Another said Claude Code was the future. A third argued Copilot is still king for enterprise.
I couldn't find a tool that did what I actually wanted: cross-reference all these claims and show me where the experts agree and disagree.
What I built
VideoStance is an automated pipeline that extracts claims from knowledge video transcripts, then cross-references them across multiple creators to surface:
Consensus — what most creators agree on
Controversy — where opinions diverge (both sides presented)
Unique insights — noteworthy claims only one creator made
Every claim is labeled with stance (support/oppose/neutral), attributed to the original creator, and linked back to the exact timestamp in the video.
How it works
Search and select 3+ videos on the same topic (YouTube and Bilibili)
Transcribe and segment transcripts
Extract structured claims via LLM — each claim gets a stance label, reasoning, and timestamp
Cross-reference across creators to find consensus and disagreement patterns
Generate a comparison hub page with spec tables, best-pick recommendations, and FAQ
A single hub like "Best AI for Coding" processes 18 videos and 476 claims in one run.
The hard part
The biggest challenge wasn't the tech — it was getting consistent claim extraction across different creators who use completely different language to talk about the same thing. One reviewer says "Cursor has better DX" and another says "Claude Code feels more natural." The pipeline has to normalize these into comparable claims.
I also went back and forth on how to present controversy. Most aggregation tools just show the majority view. I decided to always show both sides explicitly with a "Where Experts Disagree" section — it's more honest and actually more useful for decision-making.
Current state
The platform is live at videostance.com with hubs covering:
Best AI for Coding (18 reviews, 476 claims)
ChatGPT vs Claude vs Gemini (7 reviews, 196 claims)
GLM 5.2 vs DeepSeek V4 Pro (10 reviews, 293 claims)
Claude Desktop vs Claude Code
AI Coding Agents
All free, no accounts needed. Still very early — right now it's a solo project running on a Next.js static export deployed to Cloudflare.
Would love feedback from anyone here who's also built content pipelines or tackled the multi-source analysis problem. What would you add?
