Ethical uses
Never Do This
These tasks are widely accepted by publishers. AI assists but you retain creative and intellectual control. Disclose as a courtesy (even when not strictly required).
1. Improve Clarity and Flow
The use case is that you wrote the contant, and AI helps polishing it.
Example prompt: in this case AI rewrites your content (needs to be checked!)
Rewrite the paragraph below as a concise, publication-ready version. Constraints: • Preserve the scientific meaning exactly. • Do not change any methodology details, numerical values, units, gene/protein names, statistical terms, or experimental conditions. • Do not add new information, interpretation, or emphasis. • Reduce verbosity and improve clarity only. • Keep the output as a single paragraph. • Return only the rewritten text. Text: [paste your paragraph]
Alternative: ask for feedback instead:
Analyse the following paragraph for a scientific communication. Please identify and highlight the weak parts. Produce a markdown table: “Original sentence (excerpt)” | “Issue” | “Suggestion” Please identify issues like: complicated sentences, lack of clarity or ambiguous parts, verbose or superflous parts that can be removed to produce a shorter version
Real bioscience example:
Original (researcher’s draft):
“We ran the shotgun reads through fastp to trim adapters and low quality bases, then removed host reads by mapping to the human genome and taking out anything that hit. After that we classified the reads with Kraken2 using the standard database and looked at the results at genus level.”
Prompt:
## Prompt
You are helping a researcher polish a rough draft of a bioinformatics methods
section into clear, precise scientific prose suitable for a manuscript.
Rules:
1. Improve grammar, terminology, and structure so it reads like a standard
methods paragraph (formal tense, standard tool/field names, logical flow).
2. Do NOT invent, assume, or infer any specific fact the draft does not state
— this includes software version numbers, reference genome builds, database
names/releases, parameter values, statistical tests, sample sizes, or
thresholds.
3. For every such missing detail that a methods section would normally need,
insert a bracketed placeholder instead of a value, e.g. `[VERSION]`,
`[REFERENCE GENOME]`, `[DATABASE NAME/BUILD]`, `[ALIGNER]`. Where helpful,
give a short example inside the bracket (e.g. `[REFERENCE GENOME, e.g.
GRCh38]`) to guide the researcher on what's expected — but never supply a
default as if it were fact.
4. Keep everything the draft *did* specify (tool names, general workflow order,
the analysis performed) — do not soften or drop it.
5. Output only the polished paragraph, no commentary.
Draft:
"""
{researcher's rough draft}
"""
After AI polish:
“Raw shotgun metagenomic reads were quality-trimmed and adapter-filtered using fastp (v[VERSION]) with default parameters. Host-derived reads were removed by alignment against the [REFERENCE GENOME, e.g. GRCh38] reference genome using [ALIGNER, e.g. Bowtie2 v[VERSION]], discarding all mapped read pairs. Taxonomic classification of the remaining reads was performed with Kraken2 (v[VERSION]) against the [DATABASE NAME/BUILD, e.g. Standard-8 2024-06] database, and profiles were summarised at the genus level.” The researcher provided the data and interpretation. AI only improved the language.
Nature explicitly says “AI assisted copy editing” — improvements to readability,
style, grammar, and spelling — does not need to be declared.
2. Generate a Shortlist of Titles
AI can brainstorm options. You can check the options, select and refine. This can help removing the blank page effect
Example prompt:
I am writing a paper about [brief description of findings]. Suggest 10 alternative titles. Vary the style: some descriptive, some declarative, some with a colon structure.
AI titles are starting points, not final answers. You know your field’s
conventions better than any model.
3. Structure Brainstorming
AI helps you think about organisation without writing the content for you.
Example prompt:
I have the following results from my study on the effect of [treatment] on [pathway] in [organism]:
- [Finding one]
- [Finding two]
- [Finding three]
Suggest three different ways I could structure the Discussion section. For each option, explain the logical flow and what argument it would support best.
This is AI as a thinking partner — it proposes structures, you fill them with
your expertise and verified knowledge.
4. Draft Figure Legends
Figure legends follow predictable formats. AI can draft them from your description.
Example prompt:
Draft a figure legend for a microscopy image showing [describe exactly what the figure shows: sample, stain, magnification, key observations]. Follow the style conventions of [target journal].
You must verify every detail. AI does not “see” your actual figure, even if you paste it.
5. Create Lay Summaries
Translating your work for a general audience expands its impact. This can prime a routine task, but you must revise the whole product as AI can mis-interpret the key finding.
Example prompt:
Rewrite the following abstract as a plain-language summary suitable for a general audience with no scientific background. Keep it under 150 words. Avoid jargon. Explain any essential technical terms.
[paste abstract]
6. Paraphrase to Avoid Self-Plagiarism
When you need to describe the same method in a new paper, AI can rephrase while preserving technical accuracy.
Example prompt:
I need to describe this method again for a new manuscript. Rephrase the following methods section without changing any technical details, numerical values, or protocol steps. Vary the sentence structure and word choice.
[paste methods section]
You must verify the output preserves all technical accuracy.
This is a legitimate use — text recycling without the ethical problems.
Caution