Academic writing and research methodology
Systematic approaches for scholarly writing, research design, and academic communication.
Research design fundamentals
Research question development
Literature review strategy
Citation management
Paper structure and writing
IMRaD structure (scientific papers)
Academic writing style
Common writing problems
Peer review and revision
Responding to reviewers
Handling criticism
Grant proposals
Proposal structure (NSF/NIH style)
Budget justification
Publishing strategy
Journal selection
Cover letter template
Preprints and persistent identifiers
Preprint servers (post a working version, get cited earlier)
Preprints are author-submitted manuscripts posted before formal peer review. Most journals now permit (or actively encourage) preprinting; verify the target journal's policy before posting if you're unsure.
Why preprint: Faster timestamp on priority claims, citable months before journal acceptance, and broader feedback before final review. (The 2026 OSTP Nelson Memo governs immediate public access to the peer-reviewed version of federally-funded US research at publication, it is not a preprint mandate. Preprinting is a separate, voluntary choice that complements but doesn't satisfy Nelson Memo compliance.)
Persistent identifiers, get an ORCID and use DOIs
- ORCID iD (orcid.org), Free 16-digit identifier that unambiguously links you across publishers, funders, and institutions. Required by most major journals and funders. Connect it to your CV, manuscripts, datasets, software, and grant applications.
- DOI (doi.org), Persistent identifier for scholarly objects. Assigned automatically by journals at acceptance; you can also mint DOIs for your own datasets, code, posters, and preprints via Zenodo (zenodo.org) or your institutional repository, useful for citing non-journal outputs.
- ROR (ror.org), Persistent identifier for research organizations, used by funders and publishers to disambiguate institutional names.
Research ethics
Ethical considerations checklist
Avoiding research misconduct
AI / LLM use in academic writing
LLM-assisted writing is the defining 2024-2026 ethics issue across academic publishing. Every major journal, funder, and ethics body has issued policy in this window, and policies are still tightening. Treat anything below as the floor, not the ceiling: read your target journal's current submission guidelines AND your funder's most recent policy before submitting.
What's universally prohibited (as of 2026)
- LLM authorship. ICMJE, COPE, Nature, Science, NEJM, Cell, JAMA, Lancet, and the major university presses all explicitly prohibit listing an LLM (ChatGPT, Claude, Gemini, etc.) as a co-author. LLMs cannot meet the accountability and approval criteria authorship requires. Use the acknowledgments section or a methods/disclosure statement instead, never the author byline.
- Generating fabricated citations. LLMs are well-documented to produce plausible-looking but fabricated DOIs, page numbers, and even author/title combinations that don't exist. Every citation in a manuscript must be verified against the source, the LLM-induced fake-citation rate has been a top retraction trigger since 2023.
- Generating data, results, or images. Synthesizing experimental data, fabricating figures, or using generative AI to "fill in" results that weren't actually obtained is research misconduct under COPE's definition.
- Undisclosed substantial use. Most journals require disclosure of any LLM use beyond trivial spell-check/grammar assistance. Failing to disclose meaningful use can be grounds for retraction.
What's typically permitted (with disclosure)
- Idea brainstorming and outlining
- Language polishing and grammar correction
- Translation of your own writing
- Code generation for analyses (with explicit testing)
- Summarization of your own notes or transcripts
- Generating boilerplate sections (cover letters, IRB language) that you then fact-check and own
Disclosure language (template)
Most journals want a methods/acknowledgments statement that names the tool, version (if available), and what it was used for. Example:
During the preparation of this work the author(s) used [tool name, e.g., GPT-5.6 Sol, Claude Opus 5, Gemini 3.7 Flash] in order to [specific use, e.g., language polishing of the introduction; drafting code for the cluster analysis in section 3.2]. After using this tool, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the content of the publication.
Adjust to match the target journal's exact required wording, Elsevier, Springer Nature, Wiley, Taylor & Francis, IEEE, and ACM each publish their own preferred language.
Disclosure checklist before submission
- Read the target journal's AI/LLM policy in the current submission guidelines (policies updated multiple times per year).
- Read your funder's policy, some (Wellcome, NSF) have stricter rules than the journal.
- Add a disclosure statement using the journal's preferred language.
- Verify every citation manually, do not trust LLM output for any DOI, author list, page range, or quoted passage.
- Verify every numerical claim, date, and named fact.
- Re-read the full manuscript to ensure your voice and argument structure dominate, not the LLM's.
- If the LLM was used for code or analysis, re-run with seeds and verify all reported numbers reproduce.
Detection tools (limited reliability)
GPTZero, Turnitin AI Detection, Originality.ai, and similar tools have documented high false-positive rates against non-native-English writers and against legitimately human-written technical text, and high false- negative rates against current models with light editing. Pangram Labs (pangram.com) has published more rigorous benchmarks and tends to outperform the older detectors at roughly equivalent settings, but is still not a definitive arbiter and shares the same fundamental limits when authors edit LLM output substantially. Detection output should never be the sole basis for a misconduct finding; journals that rely on it as a gatekeeping signal are increasingly walking that back. Treat detector results as a flag for follow-up discussion with the author, not as evidence on their own.
