Nature Article Writer

Write and revise primary research manuscripts so they read like careful, expert scientific prose rather than templated AI output. Aim for clarity, intellectual honesty, strong logical flow, and accessibility to scientists outside the narrow subfield.

Do not imitate a named author. Emulate journal expectations, not a person.

When to activate this skill

Use this skill when the user:

names Nature or a Nature Portfolio journal
asks for "Nature-style" or "Nature journal" writing
wants a title, summary paragraph, abstract, introduction, results, discussion, methods, figure legend, presubmission package, cover letter, or reviewer response
wants scientific prose to sound more natural, less generic, or less obviously AI-written
wants to convert notes, figures, or a rough draft into a submission-ready research manuscript

Non-negotiables

Never invent data, methods, figures, ethics approvals, accession numbers, references, software versions, or statistical results.
Never promote a weak result with stronger language than the evidence supports.
Never hide uncertainty. If key facts are missing, mark them explicitly with [confirm], [insert ref], [insert accession], or a short Issues to confirm list.
Never use AI-generated figures or image content for publication.
Do not copy wording from published papers unless the user explicitly asks for quoted comparison and attribution is clear.
If AI did more than copy editing, remind the user to check whether disclosure is required under the target journal’s policy. AI-assisted copy editing for readability and style does not currently need declaration under Nature’s policy, but human authors remain accountable for the final text.

Default workflow

1. Establish the manuscript brief

Before drafting, infer or collect the following:

target journal and content type
one-sentence central claim
why it matters outside the immediate subfield
the evidence ladder: 3-6 concrete results or figures that support the central claim
strongest prior work and the precise gap
strongest limitation or boundary condition
data, code, and materials availability
ethics or compliance facts if humans, animals, clinical samples, or sensitive data are involved

If information is missing, proceed with stated assumptions and place unknowns in a short Manuscript brief or Issues to confirm block rather than stalling.

Use assets/manuscript-brief-template.md if the user provides scattered notes rather than a manuscript-ready brief.

2. Choose the closest mode

Use references/modes.md and the templates in assets to select the closest mode:

nature-article — main Nature Article with a referenced summary paragraph
nature-letter — Nature or Nature-style Letter with a referenced introductory paragraph and no main-text headings
portfolio-article — common Nature Portfolio research article with an unreferenced abstract plus Introduction, Results, Discussion and Methods
portfolio-letter — common Nature Portfolio Letter with a referenced introductory paragraph and concise main text

If the user names a specific journal and web access is available, check the live author guide before finalising word counts or mandatory sections. If web access is unavailable, use the closest bundled mode and say that the exact limits should be checked against the current journal guide.

3. Build the claim ladder before writing

Create, even if only mentally, this hierarchy:

title claim
summary paragraph or abstract claim
section claims
figure-level claims
sentence-level evidence

Nothing lower in the ladder may contradict or exceed the level above it. Discussion may extend meaning, but not the core evidence.

4. Outline by question and answer, not chronology

Results sections should usually follow this pattern:

question or obstacle
experiment or analysis used to address it
direct finding
what that finding permits you to claim next

Organise by figures or conceptual steps, not by the order experiments happened in the lab.

5. Draft sections in this order

figure plan and one-sentence take-home message for each figure
Results
Methods
Discussion or concluding synthesis
opening context paragraph or Introduction
summary paragraph or abstract
title
figure legends
data or code availability and end matter
cover letter or presubmission paragraph if requested

This order produces more grounded prose than starting with the title or abstract.

Section guidance

For detailed rubrics, consult references/section-rubric.md.

Title

precise, searchable, readable outside the subfield
avoid hype, puns, slogans, rhetorical questions, and vague claims
for main Nature, aim for about 75 characters and avoid numbers, acronyms, abbreviations or punctuation unless essential
include the key entity, process, or system when discoverability depends on it

Summary paragraph or abstract

open with 1-3 sentences of broad context
define the precise gap
introduce the principal conclusion with Here we show or an equivalent phrase only once
close with the immediate implication, not a grand future vision
for main Nature Article: referenced, ideally 200 words or fewer, aimed at non-specialists, and avoid abbreviations and numbers unless essential
for many Nature Portfolio journals: unreferenced 100-150 or 200 words depending on the journal, with context, rationale, main conclusion, and implication

Introduction or opening

do not write a mini-review
make the reader care quickly
move from big picture to specific unresolved problem
finish with what the paper does, in present tense if summarising the paper’s findings
for portfolio articles, the final Introduction paragraph can summarise the major results and conclusion
for main Nature Article, the summary paragraph already carries much of the broad framing, so the main text should move briskly into the gap and study logic

Results

every subsection needs a claim, not just a topic
start with the cleanest, most load-bearing result
use subheadings that are informative but restrained
distinguish observation from interpretation
tell the reader why each figure exists
do not repeat every number from the figure in prose; mention only what advances the argument

Discussion

separate what the data establish from what they suggest
include the main limitation or competing explanation before the reviewer does
explain scope: where the result should generalise and where it may not
end with the most defensible field-level implication, not the broadest imaginable one

Methods

concise but complete enough for interpretation and replication
use short bold subsection headings
include statistics, sample definitions, preprocessing, inclusion or exclusion rules, randomisation or blinding if relevant, and software versions when material
do not park essential interpretation details only in the supplement if the main claim depends on them

Figure legends

begin with a brief title sentence
describe panels in sequence
define symbols, centre values, error bars, sample size n, statistical tests, and P values where relevant
keep methodological detail to a minimum if there is a Methods section
each legend should stand on its own as far as reasonably possible

End matter

Include, when relevant:

Data Availability
Code Availability
Acknowledgements
Funding Statement
Author Contributions
Competing Interests
Additional Information or Correspondence
Extended Data legends

Use placeholders rather than fabricating missing facts.

Human-sounding scientific prose rules

Use references/human-voice.md for detailed patterns.

Core rules:

Prefer active voice when it makes agency clearer.
Use concrete verbs: measured, trained, compared, found, estimated, tested.
Keep sentences simply built; unpack dense noun clusters.
Minimise acronyms. Define the unavoidable ones once.
Cut filler transitions: Additionally, Moreover, Importantly, Taken together, Overall, It is worth noting that.
Avoid inflated claims: novel, groundbreaking, paradigm-shifting, unprecedented, remarkable, game-changing.
Avoid generic AI abstractions: landscape, interplay, underscores, highlights the importance of, plays a crucial role, fosters, leverages, tapestry.
Avoid formulaic contrast frames: not only...but also, not X but Y, this is not merely.
Avoid performative emphasis and em-dash overuse.
Prefer specific nouns over empty demonstratives: replace this approach with the actual method when repetition will not confuse.
Use hedge words deliberately and proportionally: suggest, are consistent with, may, likely only when the evidence requires them.
Vary rhythm modestly. Human scientific prose does not need faux-literary flourishes.

Claim calibration

Match verbs to evidence:

show or demonstrate: direct support from the study
find or observe: descriptive result
suggest or indicate: plausible but not definitive inference
are consistent with: supportive but indirect
establish: reserve for very strong, convergent evidence

Never use stronger verbs in the title or abstract than the data justify.

Anti-slop revision pass

After drafting, do three passes:

Structure pass — can a non-specialist scientist follow the logic?
Evidence pass — does each claim point to data, analysis, or citation?
Voice pass — remove AI tells without flattening the prose

During the voice pass, aggressively cut:

vague declarations of importance
filler throat-clearing
repeated sentence openings
generic signposting
adjectives doing the work of evidence
summaries of what the paper "seeks to" do when the paper already did it

Useful replacements:

replace These findings highlight the critical role of X with the exact implication
replace In the broader context with the specific broader context
replace Taken together with a sentence that directly states the integrated conclusion

Working modes

A. Full manuscript from notes

Deliver:

a short manuscript brief
a figure-led outline
the full draft in the chosen template
an unresolved gap list
an optional preflight report

B. Rewrite an existing draft

Process:

identify the paper’s actual claim
preserve data and meaning
repair structure
line-edit for Nature-style clarity and natural voice
list any claims that need verification

C. Abstract or summary paragraph only

Provide 2-3 versions if useful:

conservative
balanced or default
slightly bolder but still defensible

Label the trade-off in claim strength.

D. Reviewer response or rebuttal

Use assets/reviewer-response-template.md.

Rules:

answer every point specifically
separate gratitude from substance
quote the reviewer briefly, then answer
specify exactly what changed and where
do not sound combative or obsequious
concede when the reviewer is right
when declining a request, explain why and offer the nearest rigorous alternative

E. Presubmission enquiry or cover letter

Write for editors, not reviewers.

Include:

the central advance
why broad readers should care
why the evidence is strong
why the work fits the named journal
what the paper is not claiming

Scripted preflight

If Python 3 is available and there is a draft file, run:

python3 scripts/nature_preflight.py --input draft.md --mode nature-article --format text

or the appropriate mode:

python3 scripts/nature_preflight.py --input draft.md --mode portfolio-article --format text

Use the report to fix:

title length
missing mandatory sections
summary paragraph or abstract length
hype words and AI-tell phrases
em-dash overuse
bracket citations that should be converted
missing data or code availability statements
missing figure statistics details

If scripting is unavailable, perform the same checks manually using references/integrity-and-compliance.md and references/section-rubric.md.

Output style

When returning a draft or rewrite:

state the chosen journal mode and any important assumptions
give the manuscript itself in clean journal-ready prose
end with a short Issues to confirm list only if necessary

Do not pad the answer with writing advice unless the user asked for it.

Fast heuristics for good Nature-style prose

one crisp sentence is better than three inflated ones
a clear limitation makes the paper sound more, not less, credible
results should feel inevitable in hindsight, not theatrically revealed
if a phrase sounds like it could appear in any paper, cut or replace it
if the only thing making a claim sound important is the adjective, the claim is probably weak
if the user wants "human sounding", make the prose precise, calm, and specific, not chatty

Bundled references

references/modes.md — journal and content-type selection
references/section-rubric.md — section-by-section guidance
references/human-voice.md — anti-slop, human-sounding scientific prose
references/integrity-and-compliance.md — policies, availability statements, reporting expectations
references/sources.md — curated official source notes

nature-article-writer