Research trending topics, collect source articles, and generate a structured research brief for content creation. Stop writing from thin air — write from real sources. Use this skill when the user wants to research a topic before writing, collect sources for an article, create a research-backed content brief, or says "research [topic] for me", "find sources about [keyword]", "content brief for [topic]", "what's the latest on [product]", "research before writing", "collect articles about [keyword]", "trending news about [topic]", "gather sources for my article", "brief me on [topic]", "what are people saying about [product]", "news roundup for [keyword]", "research brief", "source collection", "content research", "prep research for writing".
Research trending topics, collect source articles, and generate a structured research brief for content creation. Stop writing from thin air — write from real sources. Use this skill when the user wants to research a topic before writing, collect sources for an article, create a research-backed content brief, or says "research [topic] for me", "find sources about [keyword]", "content brief for [topic]", "what's the latest on [product]", "research before writing", "collect articles about [keyword]", "trending news about [topic]", "gather sources for my article", "brief me on [topic]", "what are people saying about [product]", "news roundup for [keyword]", "research brief", "source collection", "content research", "prep research for writing".
Research a topic by collecting 5-10 real source articles, auto-tagging them by theme,
extracting key data points, and synthesizing unique content angles. The output is a
structured research brief that any downstream content skill can consume.
The problem this solves: Most AI-written affiliate content is generic because it's
written from the model's training data — not from real, current sources. This skill
forces research-first content creation: find real articles, extract real data, then
write from those sources. The result is content with specific stats, real quotes, and
current information that readers (and Google) actually value.
Inspired by the content-pipeline approach:
Topic → Search → Select sources → Synthesize → Write with context.
Stage
This skill belongs to Stage S2: Content — but acts as the research foundation for all content skills.
When to Use
Before writing any article, blog post, or long-form content
When you need current data and stats about a topic (not just AI-generated claims)
When creating comparison content (need real feature/pricing data from sources)
When writing about a product launch, funding round, or industry trend
After trending-content-scout identifies a topic — research it deeper
When you want unique angles: N sources → N different content pieces
Input Schema
topic:string# (required) "HeyGen AI video tool", "email marketing trends 2024"source_count:number# (optional, default: 7) How many sources to collect (3-10)source_types:string[]# (optional, default: ["news", "blog"])# Options: "news" | "blog" | "linkedin" | "youtube" | "reddit" | "academic"freshness:string# (optional, default: "month") "day" | "week" | "month" | "year" | "any"product:object# (optional) Focus research on a specific productname:string
# "HeyGen"
url:
string
# "https://heygen.com"
language:
string
# (optional, default: "en") "en" | "vi" | any ISO 639-1 code
angle_count:
number
# (optional, default: 3) How many unique content angles to generate
Workflow
Step 1: Search for Sources
Execute multiple searches to find diverse, high-quality sources:
Primary search:
web_search "[topic]" → top results
Source-type-specific searches:
IF "news" in source_types:
web_search "[topic] news [current year]" → recent news articles
IF "blog" in source_types:
web_search "[topic] blog review analysis" → in-depth blog posts
IF "linkedin" in source_types:
web_search "[topic] site:linkedin.com" → LinkedIn posts/articles
IF "youtube" in source_types:
web_search "[topic] site:youtube.com" → YouTube videos with descriptions
IF "reddit" in source_types:
web_search "[topic] site:reddit.com" → Reddit discussions with real user opinions
IF "academic" in source_types:
web_search "[topic] research study data statistics" → data-heavy sources
Product-specific (if product provided):
web_search "[product.name] review [current year]"
web_search "[product.name] alternatives comparison"
web_search "[product.name] pricing features"
web_search "[product.name] news launch update"
Collect 15-20 search results, then filter down to source_count best sources.
Step 2: Fetch and Extract Source Content
For each selected source:
web_fetch [url] → extract full article text
If fetch fails (paywall, timeout) → use search snippet as summary, note limitation
Extract from each source:
Title and URL
Published date (if available)
Key data points: stats, numbers, percentages, dollar amounts
Key quotes: noteworthy statements from experts or users
Main argument/thesis: what is this source's core message?
Unique information: what does this source have that others don't?
Step 3: Auto-Tag Sources
Tag each source with 1-3 theme tags:
Tag
Trigger Keywords
AI
artificial intelligence, machine learning, GPT, neural, model
Funding
raised, funding, series A/B/C, investment, valuation, IPO
how to, guide, step-by-step, tutorial, walkthrough
Opinion
I think, in my experience, hot take, unpopular opinion
Step 4: Extract Key Data Points
From all sources combined, extract a master list of:
Stats & Numbers:
Revenue/valuation figures
User counts / growth rates
Market size data
Performance metrics
Pricing data points
Quotes & Insights:
Expert opinions
User testimonials (from Reddit, reviews)
Founder/CEO statements
Analyst predictions
Facts & Features:
Product features mentioned across multiple sources
Recent updates/launches
Integration ecosystem
Competitive positioning
Step 5: Synthesize Unique Angles
From the collected sources, generate angle_count unique content angles.
Angle generation rules:
Each angle must use a DIFFERENT primary source as its foundation
All angles use ALL sources as context (richer data)
Each angle must have a distinct hook and perspective
At least one angle should be contrarian or non-obvious
For each angle:
Angle:title:string# Specific, could be a headlineprimary_source:string# Which source drives this anglehook:string# Opening linekey_data:string[]# 2-3 data points from sources that support this angleformat_suggestion:string# "linkedin_post" | "blog_article" | "tiktok_script" | "twitter_thread"unique_value:string# What makes this angle different from generic AI-written content
Step 6: Compile Research Brief
Organize everything into a structured brief that downstream skills can consume.
Step 7: Self-Validation
Before presenting output, verify:
All sources are real URLs (not hallucinated)
Data points are attributed to specific sources
At least 3 sources were successfully fetched (not just search snippets)
Angles are genuinely different from each other (not rephrased versions)
Tags accurately reflect source content
Brief includes both positive and critical/balanced perspectives
If any check fails, fix before delivering. Do not flag checklist to user.
Output Schema
output_schema_version:"1.0.0"topic:stringsources_collected:numbersources_fetched:number# how many were fully fetched vs snippet-onlysources:-title:stringurl:stringpublished_date:string|nulltags:string[]# ["AI", "Tools", "Pricing"]key_data_points:string[]# extracted stats and numberskey_quotes:string[]# notable quotesmain_thesis:string# 1-sentence summaryunique_info:string# what's unique about this sourcefetch_status:"full"|"snippet"# transparencymaster_data:stats:string[]# all stats across all sources, deduplicatedquotes:string[]# all notable quotesfacts:string[]# key facts and featurestimeline:string[]# chronological events if applicableangles:-title:stringprimary_source:stringhook:stringkey_data:string[]format_suggestion:stringunique_value:stringrecommended_next_skill:string
Output Format
## Content Research Brief: [Topic]
📚 **[X] sources collected** | [Y] fully fetched | Freshness: [month]
🏷️ **Top tags:** AI (5), Tools (3), Pricing (2), Comparison (2)
---
### 📰 Sources
| # | Title | Tags | Date | Status |
|---|-------|------|------|--------|
| 1 | [Title](url) | AI, Tools | Mar 2024 | ✅ Full |
| 2 | [Title](url) | Pricing, Comparison | Feb 2024 | ✅ Full |
| 3 | [Title](url) | Trends, Growth | Mar 2024 | ⚠️ Snippet |
| ... | ... | ... | ... | ... |
---
### 📊 Key Data Points (from sources)**Stats:**- [Stat 1] — Source: [#1]
- [Stat 2] — Source: [#3]
- [Stat 3] — Source: [#2, #5]
**Quotes:**- "[Quote]" — [Person], [Role] (Source: [#4])
- "[Quote]" — [Person] (Source: [#2])
**Key Facts:**- [Fact 1] — mentioned in [X] sources
- [Fact 2] — mentioned in [Y] sources
---
### 🎯 Content Angles (ready to write)#### Angle 1: "[Title]"-**Primary source:** [#2] — [title]
-**Hook:** "[Opening line]"
-**Key data:** [stat 1], [stat 2], [quote]
-**Best format:** LinkedIn post
-**Unique value:** [Why this isn't generic]
→ Run: `viral-post-writer` with angle: "[this angle]"
#### Angle 2: "[Title]"-**Primary source:** [#5] — [title]
-**Hook:** "[Opening line]"
-**Key data:** [stat 3], [fact 1]
-**Best format:** Blog article
-**Unique value:** [Why this is different from Angle 1]
→ Run: `affiliate-blog-builder` with angle: "[this angle]"
#### Angle 3: "[Title]" (Contrarian)-**Primary source:** [#7] — [title]
-**Hook:** "[Opening line]"
-**Key data:** [counter-stat], [user complaint from Reddit]
-**Best format:** Twitter thread
-**Unique value:** Goes against the dominant narrative — [reasoning]
→ Run: `twitter-thread-writer` with angle: "[this angle]"
---
### 🚀 Next Steps1.**Pick an angle** and run the suggested content skill
2.**Combine angles** — use `content-pillar-atomizer` to turn one angle into 15+ pieces
3.**Add visuals** — use `infographic-generator` to create a data infographic from the key stats
Error Handling
Topic too vague: Ask user to narrow down. "'Marketing' is too broad. Can you specify? e.g., 'email marketing automation tools' or 'TikTok marketing for SaaS'."
Few sources found: If <3 sources, note: "Limited sources available for this topic. The brief may lack depth. Consider broadening the topic or checking if it's too niche."
Most sources behind paywalls: Use search snippets. Note: "[X] sources couldn't be fully fetched (paywalls). Brief uses search snippets for those. Data may be less detailed."
Sources are all from the same perspective: Note bias. "Warning: all [X] sources are positive reviews. No critical perspectives found. Consider adding 'reddit' or 'opinion' to source_types for balanced content."
Outdated sources: If freshness filter returns old results, widen the time range and note: "Most recent sources are from [date]. This topic may not have recent coverage."
Non-English topic: Research in the specified language. Note if source diversity is limited in that language.
Examples
Example 1:
User: "Research HeyGen for a LinkedIn post"
→ topic: "HeyGen AI video", source_types: ["news", "blog", "linkedin"], freshness: "month"
→ Collect 7 sources: 2 news (HeyGen raises $60M), 3 blog reviews, 2 LinkedIn posts
→ Tags: AI (7), Funding (2), Tools (5), Comparison (1)
→ Key stats: "$60M Series A", "40K+ businesses", "Avatar 3.0 launch"
→ Angles: (1) "HeyGen just raised $60M — here's what it means for AI video" (LinkedIn),
(2) "I tested HeyGen vs Synthesia for 30 days" (blog), (3) "AI video tools are killing
the $45B video production industry" (Twitter thread)
Example 2:
User: "Brief me on email marketing trends, I want to write a comparison blog post"
→ topic: "email marketing trends 2024", source_types: ["news", "blog", "reddit"]
→ Collect 8 sources covering: AI personalization, interactive emails, privacy changes, deliverability
→ Angles focused on comparison: "ConvertKit vs Mailchimp in 2024: the real differences after
Apple Mail Privacy Protection"
Example 3:
User: "Research what people are really saying about ClickUp on Reddit"
→ topic: "ClickUp", source_types: ["reddit", "blog"], freshness: "month"
→ 4 Reddit threads (raw opinions), 3 blog reviews
→ Unique angle: Reddit users love the free tier but hate the learning curve →
"ClickUp: the free tool that takes a month to learn (and why it's still worth it)"
Feedback & Issue Reporting
When this skill produces unexpected, incomplete, or incorrect output, generate a
skill_feedback block (see shared/references/feedback-protocol.md for full schema).
Skill-specific failure modes:
Most sources paywalled: <3 sources fully fetched. Report as data_quality, list which URLs failed.
All sources same perspective: No balanced/critical viewpoints found. Report as data_quality, note bias direction.
Hallucinated stats: Agent generated a stat not from any fetched source. Report as hallucination, critical severity.
Angles not unique: All 3 angles are rephrased versions of the same take. Report as wrong_output.
Auto-detect triggers:
sources_fetched < 3 (most failed)
All source tags are identical (no diversity)
Any data point in master_data.stats cannot be traced to a specific source URL
