J’ai testé l’IA de YouTube (je ne m'attendais pas à ça)

Overview

The creator tests YouTube Studio’s new integrated AI chatbot (referred to as the Studio assistant or “Lia”), which can access channel data — comments, video content/transcripts, scripts/timecodes, performance metrics and retention curves — to produce channel-specific analysis and suggestions.

• The creator provides a free downloadable sheet of 48 prompts (first link in the video description) for use with Studio AI.

Key product features demonstrated

• Channel-level analysis
  • Scans recent videos to identify top-performing topics, video formats, and audience behavior (views, engagement, CTR).
• Title and CTR analysis
  • Compares best/worst-performing titles, surfaces high-performing title templates, and suggests title angles (e.g., specificity, “hidden truth” angles, personal-experience framing).
• Audience segmentation (shorts vs long-form)
  • Calculates subscriber conversion rates per format (example: long-form videos converted ~6× better than shorts; shorts are useful for discovery/top-of-funnel).
• Comment analysis and summarization
  • Extracts recurring problems, common questions, brand adjectives, turning points viewers cite, and ranks impactful testimonials.
• Video-level diagnostics
  • Reads scripts and retention curves to identify attention peaks and drop-off points, pinpoints the best-performing moments, and explains why certain moments spike retention.
• Practical optimization advice
  • Reduce long intros; get to the point faster.
  • Use curiosity loops (start with a striking benefit or tease).
  • Integrate CTAs in the narrative and sandwich CTAs between teasers and rewards.
  • Add frame changes/visual interrupts every ~45–60 seconds.
  • Use one concrete example per idea.
  • Avoid long theoretical conclusions; end quickly after the last actionable tip.
• Idea generation and brainstorming
  • Proposes video topics, outlines, and angles from comment sentiment and performance signals; suggests controversy or debate-worthy topics to stimulate engagement.
• Usability options
  • Ability to select a single video for deep analysis (comments + retention + script) and to request specific filters (e.g., only business videos, exclude shorts).

Concrete analyses and example outputs

• Top-performing channel themes: YouTube/social media analysis and personal development/psychology (ADHD, autism).
• CTR-winning title templates example:
  • “I analyzed 1000 sales emails. Here’s the only sentence that gets clicks.”
• Audience metrics (example outputs):
  • Long videos generated far more new subscribers than shorts.
  • ~74% of shorts viewers were new; only ~6% watched both formats.
• Comment-derived insights:
  • Three recurring problem topics: diagnosis access for neurodivergent adults; authenticity vs delegation; small-creator discouragement.
  • “Moustache effect” — audience notices and jokes about the creator’s mustache.
• Brand perception (audience-associated adjectives):
  • Authentic; insightful/lucid; reassuring; inspirational; strategic.
  • Strengths: radical lucidity, vulnerable authenticity, minimalist aesthetic.
  • Weaknesses: repetition, lack of technical evidence, perceived inconsistency/weather-vane.
• Retention diagnostics:
  • Identified a retention peak at 4:37 tied to a strategic algorithm explanation.
  • Identified a ~55% drop in the first 30 seconds (intro-length issue).
  • CTAs mid-video can cause audience dips unless followed by a reward.
• Content ideas and outlines:
  • Examples: “How to create a micro-celebrity guide,” ADHD productivity tools demo, updated angles for past business videos.

Limitations, pitfalls, and cautions observed

• Limited context window
  • Primarily analyzes recent videos (practically about the last 30–365 days); may miss older evergreen hits in large catalogs.
• No cross-channel or macro YouTube trends
  • Cannot (or did not) leverage broader YouTube-wide data or competitor insights beyond the creator’s own channel.
• No image or document upload
  • Cannot analyze thumbnails or other external assets.
• No integration with business/sales data
  • Cannot connect to CRM/sales metrics, so it cannot attribute customers to videos or measure true conversions.
• Risk of hallucination
  • The assistant can fabricate answers when lacking precise data (e.g., claiming comment timestamps or alignments that may not exist); outputs must be verified.
• Privacy / partial view
  • Tool only “sees” YouTube data (comments, retention, views); off-platform feedback or sales are not visible, so purely optimizing to YouTube metrics may misalign with business outcomes.
• Context retrieval bugs
  • The tool sometimes misses older content, misclassifies shorts vs long-form, or mixes contexts.

Practical use cases and recommended workflows

• Brainstorm video ideas from comment debates, unresolved audience questions, and retention hotspots.
• Rapid title testing and generating CTR-focused alternatives based on historical CTR patterns.
• Improving retention
  • Use the AI’s retention-curve explanations to locate weak points (intros, CTAs, theoretical sections) and apply fixes (curiosity loop, visual interrupts, concrete examples).
• Comment mining
  • Filter and summarize comments to extract FAQs, turning-point phrases to reuse in marketing, and testimonial candidates.
• Pre-publish checks
  • Run scripts against retention patterns and comment sentiments to anticipate backlash, sensitivity points, or comprehension issues.
• Augment with off-platform expertise
  • Use human coaches and off-platform analytics for conversion optimization — don’t replace business-level analysis with Studio’s metric-optimized guidance.

Resources mentioned

• Free downloadable sheet: 48 prompts for YouTube Studio AI (first link in the video description).
• Creator’s coaching program (YouTube Dividend) — paid offering (second link in the description).

Main tips the creator emphasized

• Long-form video builds a loyal audience; shorts are discovery tools.
• Integrate CTAs strategically (use the sandwich method), shorten intros, and end immediately after the last actionable advice.
• Use AI for brainstorming and data summarization, but validate outputs and supplement with off-YouTube business data and human feedback.

Main speakers / sources

• Antoine BM — the video creator and tester of YouTube Studio AI (primary narrator).
• YouTube Studio AI assistant (referred to as Lia / the Studio chatbot) — the tool being tested.
• Mentioned third-party models/tools for context: Gemini (speculated), Claude (paid alternative referenced for planning quality).