Tell if a YouTube video is helpful or trying to sell in 10s

YouTube's 2.0 billion monthly viewers—and the advertisers, creators and platforms that serve them—often can't tell within seconds whether a video is genuinely helpful or primarily a sales pitch, which wastes attention, weakens trust, and creates compliance risk. Viewers seeking quick answers, brands needing accurate context signals, creators navigating disclosure rules, and regulators demanding transparency all face this problem. You could build a real-time, 10-second classifier that ingests multimodal inputs—speech-to-text, on-screen text and objects, thumbnails, metadata and links—and emits a helpful vs. promotional label with confidence, timestamped evidence snippets, and a browser extension plus API for platform and brand integrations. Monetization would be a mix of consumer subscriptions, API licenses to platforms and brands, and verification partnerships with creators and compliance vendors, aligning with an addressable market estimated at $12.0B and an intended ARPU of about $6 annually. The timing is favorable: transformer-based multimodal models and reliable speech-to-text make sub-10s semantic and intent decisions feasible, creator-driven monetization increases undisclosed promotions, and regulatory pressure raises demand for independent verification. Market indicators—a market score of 92/100 and revenue potential 86/100—suggest meaningful demand and willingness to pay for accurate, fast signals. To stand out in a low-competition field you must prioritize high precision across languages, clear explainability (showing the transcript snippets or visual cues behind each label), and enterprise-grade APIs so platforms and brands can trust and scale the signal. Real challenges remain—building adversarial-resistant labeled datasets, minimizing false positives that alienate creators, and navigating disclosure and IP law—so early platform partnerships and a conservative, accuracy-first roadmap are advisable.