Chapter 7: AI-Assisted Writing and Research Communication
Using Generative AI as a Writing Partner, Not a Ghostwriter
What You Will Learn
How to use AI tools practically at each stage of the research writing process, from first outline to final manuscript.
What the major journal policies on AI use actually say, and how to write a disclosure statement.
How AI can help with professional research communication beyond manuscripts, including emails and lay summaries.
A “Try This” section at the end with exercises you can apply to your current writing project.
One thing to keep in mind throughout: AI is here to help you think and communicate more clearly, not to write your research for you.
A Familiar Scene
It is 10:30 at night. You have been staring at the same partially written discussion section for the past hour. You know what you want to say. You ran the analysis, you understand the results, and you have a mental picture of how this finding connects to the existing literature. But the words just will not come together into coherent sentences.
This experience is not specific to any one field. A clinical researcher might be trying to explain why a treatment effect showed up only in older patients, the opposite of what the literature predicted. A social scientist might be wrestling with how to frame an unexpected pattern in survey data without overstating what the numbers can actually support. A computational researcher might know exactly what their model does but struggle to write an introduction that non-specialist reviewers will follow. The domain is different, the data is different, but the feeling is the same: you understand your work, and you cannot write it right now.
Now imagine telling an AI tool exactly that. Not “write my discussion section,” but something more like: “I found a result that surprised me, here is what I expected and here is what I got, and I need to write a paragraph that honestly addresses this gap and connects it to two papers in my library.” The AI will come back with something imperfect, maybe quite imperfect. But it will be a paragraph, and having a paragraph to react to is a completely different mental state than staring at a blank page.
That is the practical core of this chapter. Not AI as a ghostwriter, but AI as a writing partner that helps you get ideas onto the page faster so you can do the real intellectual work of shaping them.
What “AI-Assisted Writing” Actually Means
Before diving into tools and techniques, it helps to be clear about the role AI should play. When you use AI in research writing, you are still the author. You are responsible for every claim, every citation, every inference. The AI’s job is to help you organize your thinking, push past stuck points, sharpen your language, and communicate more clearly with your intended audience.
Think of it the way you might think of working with a writing center consultant or a thoughtful labmate who reads your drafts. They can point out where an argument feels unclear, suggest a different way to open a section, or ask “what do you actually mean by this?” But they are not writing your paper. You are.
This distinction matters because the way you interact with AI tools should reflect it. If you hand AI a vague prompt and paste whatever it generates directly into your manuscript, you are setting yourself up for problems: factual errors, hallucinated citations, prose that sounds nothing like you, and potential disclosure violations with your journal. If instead you use AI as a thinking scaffold, you get the actual benefits without those risks.
Where AI Fits in the Writing Process
Research writing is not linear. Most researchers cycle through phases of brainstorming, outlining, drafting, getting stuck, revising, and revising again. Here is where AI genuinely helps at each stage.
Getting Started: Brainstorming and Framing
When you are in the early stages of a piece, AI can help you break out of your own perspective. You might ask a tool like ChatGPT, Gemini, or UM-GPT something like: “I am writing a research article about X. Here is what I found. What are three different ways I could frame the significance of this finding for a general biomedical audience?” The responses may not be right, but they will give you something to react to, and reacting is often easier than starting from nothing.
This works especially well for titles and abstracts, which are notoriously hard to write because they demand extreme compression. You can draft a rough abstract yourself, share it with an AI tool, and ask it to generate three variations that emphasize different aspects of the work. Then you pick the framing that best matches your intent and refine from there.
Building Structure: Outlining Before You Draft
One of the most underused applications of AI in research writing is using it to help with structure before you have written a single paragraph. You can describe your study, your key findings, and your target journal to an AI tool and ask it to sketch a potential section outline. This gives you a starting scaffold that you can accept, reject, or rearrange.
A useful prompt for this might look like: “I am writing a methods section for a clinical study that used retrospective EHR data and a propensity score matching approach. I need to cover data source, cohort definition, exposure and outcome variables, and statistical analysis. Can you suggest a logical order and sub-section structure for this?” The AI will give you something to work with. You are not copying it, you are using it to think.
Drafting: Getting Words on the Page
For drafting, the most practical approach is to use AI to generate rough first-pass text for sections where you are stuck, then revise heavily. The key is to give the AI enough context to be useful. Instead of “write my introduction,” try “I am writing an introduction for a paper on cardiovascular disease prediction in older adults. My paper addresses a gap in the literature: most existing prediction models were built on younger patient cohorts. Here are three papers I want to cite. Can you draft an opening paragraph that establishes this gap?”
Even if the draft is not usable, it will almost certainly get you moving. And in research writing, momentum matters.
A few important cautions here. First, AI tools like ChatGPT and Gemini regularly fabricate citations. They will confidently generate a reference that does not exist, or that exists but says something completely different from what was claimed. Always verify every citation independently, using a tool like Semantic Scholar, Zotero, or direct journal search. Second, never paste AI-generated text directly into a manuscript without substantial revision. You need to ensure the content is accurate, that it reflects your actual data and findings, and that it reads in your voice.
Editing and Polishing
This is where AI tools are probably most reliable and most widely used. Tools like GrammarlyGO, Wordtune, and the Overleaf AI assistant can catch grammar issues, suggest cleaner phrasing, identify sentences that are too long or too dense, and help you adjust tone between a technical audience and a general one. These are genuinely useful, low-risk applications.
For polishing, a helpful prompt pattern is to give the AI a specific task rather than a general one. Instead of “edit this paragraph,” try “can you revise this paragraph to reduce jargon while keeping the technical accuracy? The audience is a biomedical researcher who is not a specialist in this area.” Specificity gets better results.
Translating for Different Audiences
If you are writing a lay summary, a patient-facing document, a grant abstract for a non-specialist review panel, or a press release, AI can dramatically speed up the translation process. You write the technical version, then ask AI to help convert it. Always review the output carefully because conceptual precision tends to get soft in translation, and domain-specific nuance can be lost. But as a starting point, this works well.
What About Emails and Other Professional Communication?
Yes, this is worth including. Researchers write far more than manuscripts. A single grant cycle might involve dozens of emails: to program officers, collaborators, institutional review staff, and potential co-investigators. Writing a clear, appropriately toned email to a program officer you have never met, or a follow-up to a colleague about a delayed data agreement, takes real effort and often does not get the attention it deserves.
AI can help here too. If you have a difficult email to write, one where you need to raise a concern diplomatically, or request a timeline extension, or push back on a reviewer comment in a cover letter, you can describe the situation to an AI tool and ask for a draft. You will almost always need to revise it to sound like you, but you will have something to start with.
Other communication formats where AI is practically useful include: responses to reviewer comments (drafting the rebuttal framing), lay summaries for grant applications, presentation abstracts, slide narrative drafts, and short-form research summaries for institutional newsletters or social media posts.
The same principles apply across all of these: you supply the intellectual content, the AI helps with structure and language, and you review everything carefully before it goes out.
Institutional and Journal Guidelines
If You’re at U-M
The University of Michigan has published guidance through the Office of Research and the U-M Library emphasizing that AI tools may be used for drafting and editing as long as authors maintain responsibility for accuracy and originality [U-M Library, 2024]. Sensitive or proprietary data must not be uploaded to public AI platforms unless approved through appropriate institutional channels. See AI Resources at the University of Michigan for a full list of approved tools and platforms.
What Major Journals Say
Policies are evolving quickly, but as of early 2025 the major journals have converged on a few shared principles: AI tools cannot be listed as authors, use must be disclosed, and authors remain fully responsible for the accuracy and integrity of the work.
Journal |
Key Policy Points |
Policy Source |
|---|---|---|
Nature |
AI tools cannot be listed as authors; AI-generated text must be disclosed; fabricated citations are prohibited. |
|
Science |
Fully prohibits AI-generated text unless explicitly permitted; AI cannot be used to generate conclusions; no AI authorship. |
|
PLOS |
Allows AI for language editing; requires disclosure; prohibits AI-generated data or results without verification. |
|
Elsevier |
AI cannot be listed as author; use must be disclosed; authors remain responsible for accuracy. |
Because policies change frequently, always check your target journal’s author guidelines directly before submitting.
When Your Journal Has a Restrictive Policy
Some journals, particularly in clinical fields such as radiology, oncology, and surgery, have adopted policies that go further than disclosure requirements. A growing number prohibit AI-generated text from appearing in manuscripts at all [Smeds et al., 2023]. If you are writing for a journal with a policy like this, here is how to think through it practically.
The first step is to read the policy carefully. A prohibition on “AI-generated text in the manuscript” is not the same as a prohibition on using AI at any stage of your research process. Most restrictive policies are focused on what appears in the submitted document, not on how you planned your study, organized your thinking, or worked through your literature review. If the policy is specifically about manuscript text, you still have room to use AI as a planning and thinking tool throughout earlier stages of the project, and none of that would violate the policy.
If the policy restricts AI involvement more broadly, the honest path is to draft without it. AI can still play a role before writing begins, helping you refine your research question, stress-test your design, or organize your notes, without any of that becoming text in the manuscript itself.
Whatever the policy says, disclose accurately. If your journal requires an AI use statement and you did not use AI in the writing process, say so directly. If you used AI in earlier stages but not in drafting, a brief and honest description of that is the right approach. The guiding principle is transparency, not minimizing what you report.
It is also worth knowing that the major international bodies that set standards for publication ethics have taken positions that are more nuanced than some individual journal policies. The Committee on Publication Ethics (COPE) and the International Committee of Medical Journal Editors (ICMJE) both emphasize that AI tools cannot be listed as authors and that use must be disclosed, but neither calls for blanket prohibition of AI assistance in the research process [Committee on Publication Ethics, 2023] [International Committee of Medical Journal Editors, 2023]. Policies in this space are still evolving, and individual journals are making different calls about where to draw the line. Reading your journal’s policy against that broader context can help you understand what the underlying concern is and how to respond to it honestly.
If you are genuinely uncertain whether a particular use of AI is permitted under your journal’s policy, the right move is to contact the editorial office and ask. Journals are generally willing to clarify, and getting a clear answer before you submit is far better than a disclosure question arising during peer review. The Committee on Publication Ethics (COPE) has also published position statements that many journals follow, which you can find at https://publicationethics.org/.
A Note on Disclosure Language
Most journals now ask authors to include a statement describing how AI was or was not used. A simple, honest disclosure might look like: “The authors used ChatGPT (OpenAI) to assist with initial drafting of the introduction and to suggest alternative phrasings during editing. All content was reviewed and revised by the authors, who take full responsibility for the accuracy and integrity of the work.” Some journals have specific templates; check the author instructions. If you would like a more structured starting point, see the AI Usage Statement template in Chapter 29: Quick Reference Templates.
Tools Worth Knowing
For writing and drafting: ChatGPT, Claude, Gemini, UM-GPT, and Maizey are the most commonly used conversational AI tools for writing assistance. Maizey is particularly useful if you are working with UM-hosted research materials, while NotebookLM works well when you want an AI that reasons specifically from documents you have uploaded.
For editing and polishing: The same general-purpose tools above work well for editing tasks when you give them a specific prompt (for example, “revise this paragraph to reduce jargon for a non-specialist audience”). If you want a more integrated editing experience built into your writing environment, Grammarly has an AI companion called GrammarlyGO, and the Overleaf AI Copilot is available if you write in LaTeX. That said, many researchers find that sticking with a conversational AI tool they already know, and just being specific about what kind of editing they want, is the most practical approach.
For citation integrity: This is a place where AI tools have real weaknesses, so supplement with dedicated tools. Semantic Scholar provides AI-powered recommendations and summaries grounded in real literature. Scite.ai classifies whether a paper supports, contrasts, or simply mentions a given claim, which is genuinely useful for lit review work. Zotero remains the gold standard reference manager, and it integrates well with AI writing tools even though it is not itself AI-powered.
For visual communication: When you need a conceptual diagram, a schematic for a presentation slide, or a placeholder illustration for teaching materials, image generation models can produce a usable draft faster than drawing one by hand or commissioning custom artwork. Stable Diffusion is the most widely used open-source option and can run locally if your content is sensitive. For quick experimentation, cloud-based interfaces are available through Hugging Face Spaces. Keep in mind that generated images are communication aids, not data: journals and conferences have increasingly explicit policies about AI-generated figures, and disclosure is always the right practice. Chapter 20 covers how these models work and what they are actually suited for.
Try This
These exercises are designed to be used with your current writing project. You do not need to start anything new.
Exercise 1: Unstick a stuck section. Take a section of a paper or report you are currently working on that you have been avoiding. Describe the section to an AI tool (what it needs to cover, what your data shows, what the reader should understand after reading it) and ask for a rough draft. Do not use the draft. Instead, use it as a prompt: what did the AI get wrong or miss? What would you say differently? Now write the section yourself, using your reaction to the AI draft as your starting point.
Exercise 2: Framing test. Take your current abstract or a one-paragraph summary of your project and share it with an AI tool. Ask for three alternative ways to frame the significance of the work for different audiences (your disciplinary peers, a general biomedical audience, and a funding panel). Compare the framings. Which elements does each version emphasize? Does this change how you think about your own framing?
Exercise 3: Check your journal’s policy. Find the AI policy for the primary journal you are currently writing for, or planning to submit to. Write a one-sentence disclosure statement that accurately describes how you have used (or not used) AI in your writing process on this project.
Exercise 4: Difficult email. Think of a professional email you have been putting off because the tone is tricky. Describe the situation to an AI tool and ask for a draft. Revise it until it sounds like you and says what you actually need to say. Notice where the AI’s version was helpful and where it went wrong.
References
U-M Library. Genai and library instruction. 2024. Accessed 2025-02-26. URL: https://guides.lib.umich.edu/genai-library-instruction.
Matthew R Smeds, Bernardo Mendes, Leigh Ann O'Banion, and Sherene Shalhub. Exploring the pros and cons of using artificial intelligence in manuscript preparation for scientific journals. J Vasc Surg Cases Innov Tech, 9(2):101163, May 2023. doi:10.1016/j.jvscit.2023.101163.
Committee on Publication Ethics. Authorship and AI tools: COPE position statement. https://publicationethics.org/cope-position-statements/ai-author, 2023. Accessed 2025.
International Committee of Medical Journal Editors. Recommendations for the conduct, reporting, editing, and publication of scholarly work in medical journals. https://www.icmje.org/recommendations/, 2023. Accessed 2025.