sharkitect-solutions/content-research-writer

Content Research Writer

File Index

| File | Purpose | When to Load | |---|---|---| | SKILL.md | Research-to-draft procedure, source credibility, research depth calibration, section feedback, citation management, hook improvement, pre-publish checklist | Always (auto-loaded) | | source-verification-methods.md | Study methodology assessment, pre-registration checks, retraction/correction verification, statistical literacy for writers, press release vs research paper distortions, fact-checking procedure, disappeared source recovery | When evaluating source quality beyond the basic tier, fact-checking specific claims, assessing study methodology, or verifying statistics | | long-form-structure-patterns.md | Argument architecture selection (6 patterns), information density calibration by audience, counterargument integration patterns, section transition engineering, multi-article series architecture, reader fatigue management | When structuring content longer than 2,000 words, building multi-section arguments, managing competing evidence, or diagnosing structural problems in drafts | | editorial-quality-control.md | Research rabbit hole timeboxing, fact-checking tiers (4 levels), accuracy under deadline tradeoffs, hedge language calibration, voice consistency audit (5 markers), named editorial failures (7), correction/update protocol | When establishing review processes, managing research scope, handling accuracy-under-deadline tradeoffs, auditing voice consistency, or dealing with corrections |

Scope Boundary

| Area | This Skill | Other Skill | |---|---|---| | Research-backed articles, blog posts, newsletters, long-form content | YES | -- | | Collaborative writing with source discovery and citation management | YES | -- | | Section-by-section feedback and iterative refinement | YES | -- | | Quick social media posts (no research depth needed) | NO | social-content | | Email copy (different structure and objectives) | NO | email-composer | | Copy editing existing text (editing, not writing) | NO | copy-editing | | Copywriting from scratch without research needs | NO | copywriting | | SEO keyword strategy and technical SEO | NO | seo-optimizer |

Assessment Procedure

Before starting any research-writing project, work through this sequence. Skipping steps leads to research rabbit holes or structural rewrites.

1. Classify the content type using the Research Depth Calibration table below. A blog post and a thought leadership piece require fundamentally different research investment -- misclassifying wastes hours or produces under-researched content.
2. Confirm the thesis or argument. If the author can't state their main point in one sentence, the piece isn't ready to research. Help them articulate it before searching for sources.
3. Assess source availability. Quick search: are there credible sources for this topic? If the topic has minimal published research, adjust expectations -- the piece may need to rely on expert interviews or primary data collection rather than citation-heavy evidence.
4. Determine the author's voice. Read 2-3 existing pieces by the author (or ask for samples). The goal is to enhance their voice, not replace it. Note: sentence length patterns, formality level, use of contractions, how they handle technical terms.
5. Build the outline before researching. The outline determines what research is needed. Researching before outlining leads to the rabbit hole -- you'll find interesting sources that don't serve the piece.
6. Set the research timebox. Based on content type classification. When the timebox expires, start writing with what you have. Load editorial-quality-control.md for timeboxing guidance.

Research-to-Draft Procedure

1. Scope the project: Confirm topic, thesis/argument, target audience, desired length, content type (blog, newsletter, tutorial, thought leadership, case study), and available sources.
2. Build the outline: Structure with hook, introduction (context + problem + scope), 3-5 main sections with key points and evidence needs, conclusion with CTA. Mark sections needing research with [RESEARCH: specific question].
3. Conduct research: Search for credible sources matching each [RESEARCH] tag. Extract key facts, data points, and quotable insights. Record full citations immediately -- never defer citation tracking.
4. Draft iteratively: Write one section at a time. After each section, review for clarity, flow, evidence, and voice consistency before moving to the next. This catches structural issues early instead of requiring full-draft rewrites.
5. Polish and validate: Full-draft review for flow across sections, citation completeness, and voice consistency. Run the Pre-Publish Checklist.

Source Credibility Hierarchy

Evaluate every source before citing. Higher-tier sources require less corroboration.

| Tier | Source Type | Examples | Corroboration Needed | |---|---|---|---| | 1 (Gold) | Peer-reviewed research, official statistics, primary data | Academic journals, government census data, SEC filings, peer-reviewed meta-analyses | None -- cite directly | | 2 (Silver) | Reputable industry research, established publications | McKinsey/Gartner reports, Harvard Business Review, established trade publications | One corroborating source preferred | | 3 (Bronze) | Expert opinion, company-published data | Blog posts by recognized experts, company case studies, conference talks | Two corroborating sources or explicit attribution ("according to...") | | 4 (Use cautiously) | Aggregated/secondary data, social proof | Wikipedia (follow to primary source), social media posts, forum discussions | Must trace to primary source before citing | | 5 (Avoid) | Unattributed claims, outdated data, content mills | "Studies show..." with no citation, data older than 3 years for fast-moving topics, SEO content farms | Do not cite -- find the primary source or drop the claim |

Research Depth Calibration

| Content Type | Research Depth | Citation Density | Time Investment | |---|---|---|---| | Thought leadership | Deep -- original angle requires understanding existing perspectives | 3-5 citations per 1,000 words; emphasis on contrasting viewpoints | 60% research, 40% writing | | Tutorial / How-to | Moderate -- verify technical claims and version-specific details | 1-2 citations per 1,000 words; link to official docs | 30% research, 70% writing | | Blog post (opinion) | Light -- support key claims, acknowledge counterarguments | 2-3 citations per 1,000 words; data to anchor subjective claims | 20% research, 80% writing | | Newsletter | Minimal -- curate and comment on existing sources | Link to every referenced source; no unsupported claims | 40% curation, 60% writing | | Case study | Deep -- verify all metrics and outcomes with the subject | Every metric cited to source; interview quotes attributed | 70% research/verification, 30% writing |

Section Feedback Framework

When reviewing a section the author has written, evaluate across four dimensions:

| Dimension | What to Check | How to Flag Issues | |---|---|---| | Clarity | Can a reader understand this on first read? Are pronouns clear? Is jargon defined? | Quote the unclear sentence and offer a specific rewrite | | Flow | Does this section connect logically to the previous one? Are paragraphs in the right order? | Suggest specific transition sentences or reordering | | Evidence | Are claims supported? Are examples concrete? Is anything stated as fact without a source? | Flag the unsupported claim and suggest what type of source would strengthen it | | Voice | Does this section sound like the rest of the piece? Is tone consistent? | Quote the inconsistent passage and explain the shift you detected |

Voice preservation principle: The goal is to make the author's writing better, not different. Suggest improvements as options, not directives. If the author prefers their version, support it. Ask periodically: "Does this sound like you?"

Citation Format

Match the author's preference. Default to numbered references if no preference stated.

| Format | When to Use | |---|---| | Inline (Author, Year) | Academic-style content, research-heavy pieces | | Numbered [1] with reference list at end | Blog posts, articles, most web content | | Hyperlinked text | Newsletters, casual blog posts, social-adjacent content |

Maintain a running ## References section. Never defer citation recording -- add the full citation the moment you use a source.

Hook Improvement Method

When the author shares an introduction, evaluate against four criteria and offer 2-3 alternatives:

1. Does it create curiosity? -- The reader should need to know what comes next
2. Does it promise value? -- The reader should know what they'll gain
3. Is it specific? -- Vague openings ("In today's world...") lose readers instantly
4. Does it match the audience? -- Technical audiences want data/insight; general audiences want story/emotion

Offer alternatives using different hook types: bold statement, surprising data point, personal story, provocative question. Explain why each works so the author can choose based on their voice.

Pre-Publish Checklist

• All factual claims have a cited source
• Citations formatted consistently throughout
• No "Studies show..." or "Experts say..." without attribution
• Hook creates curiosity and promises value
• Every section connects logically to the next (read transitions aloud)
• Conclusion includes a specific call to action (not just "thanks for reading")
• Author's voice is consistent from intro to conclusion

Recommendation Confidence

Not all guidance above carries equal certainty. Override when your specific context demands it.

| Area | Confidence | Override When | |---|---|---| | Source credibility hierarchy (5 tiers) | HIGH | No known context where citing Tier 5 sources helps. The tier boundaries between 2-3 may shift for niche domains where "established publications" don't exist. | | Section-by-section drafting (not full-draft) | HIGH | Exception: very short content (<800 words) where full-draft review is sufficient. Also, experienced authors with strong structural instincts may draft 2-3 sections before review. | | Citation deferral risk | HIGH | No exception. Deferred citations become fabricated citations. The moment you use a fact, record where it came from. | | Research depth calibration by content type | MEDIUM | Depth depends on existing expertise. An author who's a domain expert writing a blog post may need less research than the table suggests. An author entering a new domain may need more than thought-leadership depth even for a short post. | | Hook improvement method (curiosity + value) | MEDIUM | Some formats (academic papers, technical documentation, internal reports) don't benefit from hooks. Match to the publication context. A LinkedIn post needs a hook; an RFC does not. | | Outline-before-research sequencing | LOW | Exploratory content ("what I learned about X") may benefit from research-first, outline-second. The risk is rabbit holes, but the reward is discovering the real story. Flag when deviating. |

Rationalization Table

| Rationalization | Why It Fails | |---|---| | "I'll add citations later" | Deferred citations become unfindable sources; the research context is freshest during drafting, not during final polish | | "This is common knowledge, no citation needed" | What's common knowledge to the author is often a novel claim to the reader; when in doubt, cite it | | "The outline is good enough, I'll figure out structure as I write" | Structural problems in drafts require full rewrites; 20 minutes on outline structure saves hours of revision | | "I'll just write the whole draft and review at the end" | Section-by-section feedback catches voice drift, logic gaps, and evidence holes before they compound across 3,000 words | | "The author wants me to just write it for them" | The skill is collaborative refinement, not ghostwriting; preserving the author's voice requires their input at every stage | | "This source is close enough" | Tier 4-5 sources cited without tracing to primary data undermine the entire piece's credibility when a reader checks one claim |

Red Flags

1. Writing an entire draft without stopping for section feedback -- structural issues compound across 3,000+ words
2. Citing a source without verifying it exists and says what you claim it says (The Phantom Citation)
3. Using "Studies show..." or "Research indicates..." without a specific citation (The Ghost Attribution)
4. Ignoring the author's existing voice and writing in a generic AI style (The Voice Hijack)
5. Adding citations only to the introduction and conclusion, leaving the body unsupported (The Citation Bookend)
6. Presenting a single source's opinion as established fact (The Single-Source Generalization)
7. Skipping the outline phase for content longer than 500 words -- rewrites cost 3-5x the time saved
8. Research rabbit hole: spending 4+ hours researching a 1,000-word blog post because "one more source" feels productive

NEVER

• Fabricate citations, statistics, quotes, or source attributions -- if you cannot find a source, say so explicitly
• Override the author's voice or writing style -- suggest improvements, do not replace their choices
• Use Tier 5 sources (unattributed claims, content farms) as evidence for any claim
• Skip the Pre-Publish Checklist before declaring content ready
• Write an entire long-form draft in one pass without section-by-section review