scientific-writing-voice

When to invoke

• Writing a 200–800 word brief of a technical paper
• Drafting newsletter prose about research
• Calibrating an LLM system prompt that asks for science-journalism tone
• Editing a colleague's overly-hedged academic text toward a broader readership

The target register (what "Quanta-meets-Nature" actually means)

1. Precise without being technical. Keep every quantitative claim (masses in MeV/c², statistical significance in σ, redshifts, temperatures). Strip the derivation apparatus (equations, Feynman diagrams, group-theoretic notation) — you're conveying what was found, not how to reproduce it.
2. Confident, not hedging. Avoid "the authors claim / suggest / argue". State the finding. If you want to qualify, say "Under the assumption that dark matter couples only through the Higgs portal, the mass window narrows to…". Write conditions as conditions, not as linguistic hedges.
3. Active voice, concrete subjects. Not "it was observed that" — "The Parkes Pulsar Timing Array observed…".
4. Lead with the discovery, not the context. Readers of a brief have ~30 seconds to decide if they care.
5. Short paragraphs (2–4 sentences). White space is a navigation aid.

Canonical structure (200–600 words)

TL;DR — one sentence, ≤40 words

Opens with a boldface **TL;DR:**. Names the phenomenon, the chief numerical result, and the setting.

TL;DR: Scalar-tensor mixing during a brief early matter-dominated epoch generates a stochastic gravitational-wave background in the nanohertz band compatible with the NANOGrav 15-year excess.

¶1 — What was studied

Name the system, the measurement or simulation, the scale. Do not explain what a pulsar timing array is unless the paper's novelty depends on it.

¶2 — How

Methodology in 2–4 sentences. Instruments, sample sizes, statistical frameworks. Mention what separates this attempt from prior work ("the first analysis that jointly fits the spectral index and amplitude").

¶3 — What they found

The quantitative heart. Always preserve numbers and their uncertainties. "A strain amplitude of 2.4×10⁻¹⁵ at 1 yr⁻¹, consistent with…" Preserve confidence intervals, priors, Bayes factors.

¶4 — Why it matters / caveats

• Matters: what it lets us conclude, or what it rules out, or what next
• Caveats: known systematics, alternative explanations the authors acknowledge, any crucial assumption

End the brief here. Don't editorialize. Don't speculate beyond the paper.

Words to avoid (almost always)

Words to keep (signals of precision worth preserving)

• Quantitative: σ, Hz, GeV, pc, arcsec, Myr, z, h⁻¹, TeV, J/K, Bayes factor, FAR
• Qualifiers that are information: "at 3.2σ local significance (2.1σ global)", "within 1-σ of the Gaussian prior", "marginalizing over nuisance parameters"
• Named instruments: NANOGrav 15-yr, LIGO-Virgo O4, JWST/NIRCam, Planck 2018
• Named frameworks: ΛCDM, Standard Model, Wilsonian RG, AdS/CFT, VQE

Anti-patterns for LLMs writing briefs

LLMs (including Grok) default to these — actively correct:

1. Restating the abstract. The brief should be richer than the abstract — pulling in context (what came before, what's next), NOT just paraphrasing.
2. Vacuous openers. "Recent advances in physics have led to…" → delete. Start with the finding.
3. Bullet lists for flow prose. Briefs are short articles, not spec sheets. Only use bullets for enumerated quantitative results (limits, couplings, redshifts at specific epochs).
4. Hedging cascade. "The authors propose that their model may suggest a possible connection." Cut to: "The model predicts X (probability P)."
5. Final paragraph "this could have implications for…" Delete. Either there's a concrete implication — name it — or there isn't. Don't wave.

System-prompt template for LLM briefs

You are a science editor writing for a physics-literate but time-constrained reader.
Target: 400–600 words, hard cap 800. Prose only, not bullets.

STRUCTURE:
- Single-sentence TL;DR (≤40 words), prefixed "**TL;DR:**"
- Paragraph 1: what was studied (system, scale, novelty).
- Paragraph 2: methodology (instruments, statistical framework, what distinguishes this effort).
- Paragraph 3: results (preserve ALL numerical claims, uncertainties, significance).
- Paragraph 4: why it matters + caveats. No speculation beyond the paper.

VOICE:
- Active voice. Concrete subjects.
- Confident, not hedging. State conditions as conditions, not linguistic hedges.
- Precise without derivational apparatus. No equations, no Feynman diagrams in prose.
- Preserve quantitative claims verbatim (MeV, GHz, σ, z, pc, Hz, TeV, M_⊙).
- Never say: "groundbreaking", "novel", "paradigm shift", "could pave the way".
- Do not repeat the abstract verbatim. Synthesize and contextualize.

Calibration check

A brief passes the voice test when:

• A domain-adjacent researcher (grad student in a nearby subfield) reads it in ≤2 minutes and correctly identifies the core result + its scale.
• A domain expert finds no quantitative errors or vague hedging.
• Removing any sentence loses information.

If any sentence reads like it could appear unchanged in a grant application, rewrite it for clarity and specificity.

_Last reviewed: 2026-05-14 — automated polish pass per issue #84._