Manuscript Writing

Academic Manuscript Writing

Guide the drafting and revision of academic manuscripts following journal conventions and scientific writing best practices.

When to Use

• Drafting a new manuscript from scratch
• Writing or revising specific sections (Introduction, Methods, Results, Discussion)
• Structuring a paper from experimental results
• Converting a conference paper to a journal submission
• Responding to reviewer comments with revised text

Manuscript Structure

Standard IMRAD Format

Section-by-Section Guidance

Introduction

Structure as a funnel:

1. Broad context - Establish the field and importance (1-2 paragraphs)
2. Narrow focus - What is known about the specific topic (1-2 paragraphs)
3. Gap statement - What remains unknown or problematic
4. Objective/hypothesis - What this paper addresses and how

Methods

• Enough detail for replication
• Subsections mirror Results sections
• Include: participants/subjects, materials, procedures, analysis
• Statistical methods: specify tests, software, significance thresholds
• Ethics: IRB/IACUC approval statement

Results

• Present findings without interpretation
• Start each subsection with the analysis performed, then the outcome
• Report effect sizes, confidence intervals (not just p-values)
• Reference figures and tables in order
• Use "significant" only for statistical significance

Discussion

Structure:

1. Summary of key findings (1 paragraph)
2. Interpretation in context of prior literature (2-3 paragraphs)
3. Limitations (1 paragraph, honest but not self-defeating)
4. Future directions (1 paragraph)
5. Conclusion (1 paragraph)

Writing Principles

Clarity

• Active voice with the authors as the actor and the named noun instead of a pronoun ("We measured..." not "Measurements were taken..."; see Habits to avoid while drafting); follow the journal where it prefers passive Methods
• Define abbreviations on first use
• Avoid jargon unless writing for a specialist audience

Precision

• Quantify claims ("increased by 23%" not "significantly increased")
• Cite sources for factual claims
• Distinguish correlation from causation
• Use hedging appropriately ("suggests" vs "proves")

Flow

• Each paragraph: topic sentence, evidence, interpretation
• Transitions between paragraphs connect ideas
• Logical progression within and across sections
• Consistent terminology (don't alternate between synonyms for key concepts)

Habits to avoid while drafting

These habits make research prose read as machine-written, and reviewers and editors notice them. Avoid them while drafting; the humanizer pass (patterns 30 to 45) then catches what slips through. The quoted phrases are examples of each habit, not the full list; a sentence that does the same job in other words breaks the same rule.

• Show, do not announce. State the limitation or the result and let it stand. Do not certify your own honesty or rigor ("and we say so", "we say it first", "to be transparent", "we are careful to").
• Say it once. Each sentence adds a fact. Drop tails that restate the purpose (", so a gain must come from what the tags say"), closing sentences that only repeat the paragraph (a sentence that links to an aim or states the gap stays), and a second gloss of a defined term (define it again only in each separately read document). Lead with a purpose ("To separate X from Y, ...") only where a reader would ask why.
• Name who acts. Write "We registered the test before seeing any fMRI value", not "The test was registered" or "Hypothesis 1B tests ..."; results predict, differ, or agree, and do not "say", "tell", or "decide". An aim can still be the topic of a sentence ("Aim 2 has two hypotheses"). Follow the journal where it prefers passive Methods.
• Name the noun. When the referent is a sentence back or two nouns could fit, repeat the noun ("CLIP's prediction", not "its prediction"); "this" needs a noun that names something in the sentence just before ("this test"). A pronoun is fine when its referent is the subject of the same clause.
• Literal claims, not slogans. No mirrored contrasts ("fMRI shows where, EEG shows when"), contest metaphors ("beat the null"), colon reveals ("The answer is simple: ..."), or verbs of standing ("rests on", "reads as").
• Answer an objection where it arises. Do not concede and rebut in every paragraph ("X is not the only limit, but ...").
• Plain time and scope. Use "now" or "currently" only to contrast with a stated earlier version or where literal ("currently recruiting"), and state a real constraint ("only", "never") once.
• One idea per sentence. Split chains of "and"; list results strongest first, each with its label.

Citations

• Cite primary sources, not reviews (when possible)
• Recent citations show awareness of current literature
• Cite competing or contradictory findings
• Self-citation should be relevant, not gratuitous

Revision Checklist

After drafting, check:

• Abstract accurately reflects the paper content
• Introduction gap statement matches the study objective
• Methods are sufficient for replication
• Results match Methods (parallel structure)
• Discussion addresses all major findings
• Figures are referenced in text in order
• All abbreviations defined on first use
• References complete and consistently formatted
• Word/page count within journal limits
• Run manuscript:humanizer as a final natural-writing pass (strips AI tells like significance inflation, em-dashes, "evolving landscape" filler, rule-of-three padding) before review

Additional Resources

• Reference: references/section-templates.md - Templates for each manuscript section
• Reference: references/revision-response.md - How to write point-by-point responses to reviewers
• Sister skill: manuscript:humanizer (invoke through the host's skill-invocation mechanism) - Removes 45 categories of AI-writing tells, including sentence-shape tells, hidden actors, and unanchored pronouns in research prose, while respecting academic conventions (e.g., passive voice in Methods, hedging in Discussion). Run after drafting and before peer review.