YouTube AI Agent prompt reference

General prompt guidelines for YouTube

Set temperature low (0.1–0.3) for classification agents — you need consistent label outputs that conditional rules can match reliably.

Set Max Response Tokens to ~50 for pure classification prompts. A category label needs almost no tokens.

Set Max Response Tokens higher (1,000–4,000) for summary and analysis agents — especially when using {{transcript}} or {{all_comments}}, which inject large amounts of text.

Use {{firmographics}} to inject your ICP definition without repeating it in every prompt.

YouTube Videos — prompt examples

Competitive Monitor

You are a competitive intelligence analyst. Read this YouTube video transcript and determine whether it contains meaningful competitive intelligence for a B2B SaaS company in our space. {{firmographics}} Video title: {{video_title}} Channel: {{channel_name}} Transcript: {{transcript}} Classify as one of: NEW_DEMO — video demonstrates product features or positioning FEATURE_ANNOUNCEMENT — new capability or roadmap item mentioned PRICING_SIGNAL — pricing or packaging language detected GENERAL_PUBLISH — new video but no specific competitive signal Output the label only. No explanation. 

Conditional rules:

• If output equals NEW_DEMO → Slack to #sales-intel + Email to product team

• If output equals FEATURE_ANNOUNCEMENT → Slack to #product

• If output equals PRICING_SIGNAL → Slack to #sales-leadership

• If output equals GENERAL_PUBLISH → Create Tag from AI Output: yt-logged

Brand Sentinel

You are a brand intelligence analyst. Read this YouTube video transcript and determine the nature of the mention of our brand or product. Our brand/product: [your brand name here] Video title: {{video_title}} Channel: {{channel_name}} ({{subscribers}} subscribers) Transcript: {{transcript}} Classify as: POSITIVE — recommends or praises our product NEGATIVE — criticises, complains about, or warns against our product COMPARISON — compares us to a competitor (positive or negative) NO_MENTION — brand name appears but not in a meaningful context Output the label only. 

Conditional rules:

• If output equals POSITIVE → Slack to #marketing

• If output equals NEGATIVE → Slack to #customer-success

• If output equals COMPARISON → Slack to #sales + Email to product team

• If output equals NO_MENTION → Discard

Video Summariser (for chaining to comment agents)

You are a content analyst. Read this YouTube video transcript and write a concise 3–5 sentence summary. Focus on: main topic, key arguments made, any tools or products mentioned, and any pain points or buying signals expressed. Video title: {{video_title}} Channel: {{channel_name}} Transcript: {{transcript}} Output only the summary. No labels, no preamble. 

Basic action: Write Output to Field → Video Summary

This output becomes available as {{ai_video_summary}} in any YouTube Comments agent assigned to the same Signal.

Comment Theme Analyser (using {{all_comments}})

You are a B2B market intelligence analyst. Analyse the full comment section from this YouTube video and produce a structured report covering: 1. Top 3 themes (with approximate % of comments) 2. Overall sentiment breakdown (Positive / Mixed / Negative with %) 3. Buying signals — comments indicating evaluation, switching intent, or active purchasing behaviour (quote examples if present) 4. Competitive mentions — tools, products, or companies mentioned 5. One-sentence summary of the most actionable insight Video title: {{video_title}} All comments: {{all_comments}} Output only the structured report. No preamble. 

Basic action: Write Output to Field → Comments Summary

⚠️ Set Max Response Tokens to 2,000–4,000 for this prompt. The {{all_comments}} variable injects the full comment thread — responses need room to be thorough.

YouTube Comments — prompt examples

Comment Classifier (primary use case)

You are a comment classification assistant for a B2B sales team. Read this YouTube comment in the context of the video it was posted on. Classify it into exactly one of these categories: BUYER_QUESTION — person asks a specific question implying active evaluation ("Does this integrate with X?" / "How does this compare to Y?") SWITCHING_SIGNAL — person expresses frustration with current tool or openness to alternatives ("We're finally moving off X" / "Looking for alternatives") FEATURE_GAP — person mentions a missing capability they need ("I wish this did X" / "The only thing stopping me is Y") GENERAL — positive reaction, off-topic, or no actionable signal Video summary: {{ai_video_summary}} Comment: {{comment_text}} Commenter handle: {{username}} Output the category label only. No explanation. 

Conditional rules:

• If output equals BUYER_QUESTION → Slack to #sdr-alerts

• If output equals SWITCHING_SIGNAL → Slack to #sdr-alerts

• If output equals FEATURE_GAP → Email to product@yourcompany.com

• If output equals GENERAL → Skip

Intent Scorer (simpler version)

Read this YouTube comment. Does it suggest the commenter is actively evaluating or considering purchasing a product in this category? Consider: specific questions about features, comparisons to competitors, expressions of switching intent, or mentions of a budget/timeline. Video context: {{ai_video_summary}} Comment: {{comment_text}} Reply TRUE if there is buying intent. Reply FALSE if not. Output the single word only. 

Conditional rules:

• If output equals TRUE → Slack to #sdr-alerts

• If output equals FALSE → Skip

Variable tag reference

YouTube Videos (youtube_videos data target)

Video content {{video_title}} {{video_description}} {{transcript}} {{post_content}} {{youtube_url}} {{video_id}} {{preview_image}}

Channel {{channel_name}} {{channel_handle}} {{channel_url}} {{channel_id}} {{subscribers}} {{channel_verified}}

Engagement {{views}} {{likes}} {{youtube_comments}} {{all_comments}}

Metadata {{video_length}} {{video_type}} {{date_posted}} {{is_sponsored}}

Cross-target {{firmographics}}

YouTube Comments (youtube_video_comments data target)

Comment {{comment_text}} {{username}} {{user_channel}} {{comment_likes}} {{replies_count}} {{comment_date}} {{is_reply}}

Parent video {{video_title}} {{video_description}} {{video_url}} {{transcript}}

Parent channel {{video_channel}} {{channel_handle}} {{channel_url}} {{subscribers}} {{channel_verified}}

Parent engagement {{video_views}} {{video_likes}} {{total_comments}} {{video_length}} {{date_posted}} {{is_sponsored}}

AI chaining {{ai_video_summary}} — output from a previously run YouTube Videos agent on the parent video

Cross-target {{firmographics}}

Token and temperature quick reference