A core principle for ethical AI use in research writing.

Key points:

• Enhancement: using AI to improve work you've created (editing, clarity, structure, formatting)
• Generation: using AI to create content you then minimally review and publish
• Enhancement is appropriate; generation without verification is ethically problematic
• The distinction: Do you take responsibility for the content? Have you verified accuracy?
• In enhancement, you're still the author and thinker; AI is a tool
• In generation, you're outsourcing thinking, creating ethical issues
• This principle applies to every research writing context

AI excels at tasks that are routine, lower-stakes, and don't require novel thinking.

Key points:

• Clarity improvement: AI suggests rewordings that are clearer without changing meaning
• Conciseness: removing redundant phrases and tightening prose
• Grammar and style checking: catching errors, flagging awkward constructions
• Tone adjustment: helping achieve appropriate academic tone
• Format consistency: ensuring citations, references, figure captions follow standards
• Outline generation: AI suggests structure for papers, proposals, or sections
• Routine section drafting: initial versions of methods sections with standard procedures you'll verify
• Translation: translating abstracts or key sections to other languages (with human review)
• All of these maintain researcher control and responsibility

AI creates risk in writing tasks requiring expertise and where accuracy is critical.

Key points:

• Novel findings interpretation: using AI to explain what your results mean risks inaccuracy
• Hypothesis generation for discussion: AI suggests ideas from training data, not novel thinking
• Statistical interpretation: using AI to interpret statistics without understanding them
• Mechanistic explanations: AI generates plausible-sounding mechanisms without domain understanding
• Literature synthesis: using AI-synthesized findings without verifying sources creates hallucination risk
• Novel argument development: AI can suggest arguments but cannot assess whether they're true
• Critical evaluation: using AI instead of your own judgment about what matters
• All of these require expertise and verification that defeats the efficiency purpose

Understanding emerging norms and policies about AI disclosure in research writing.

Key points:

• Many journals now require disclosure of AI use in papers
• Some journals prohibit AI-generated text without human review
• Funding agencies are developing AI use policies
• Universities are establishing guidelines for appropriate AI use
• General principle: transparency about how AI was used
• Disclosure typically appears as a statement: "An AI language model was used to improve clarity in the methods section; the authors verified all technical content"
• Nondisclosure when AI was used creates integrity concerns
• The question "Would I disclose this use to editors and my institution?" is a good integrity test
• Failing to disclose AI use that shaped major content is plagiarism of the AI's work

How to integrate AI into your writing process effectively and responsibly.

Key points:

• Start with outline and content (you do this)
• Draft initial text (you or AI, depending on task—routine procedures can be drafted by AI)
• Verify all claims and facts (you do this—non-negotiable)
• Use AI to improve clarity and structure (AI does this with your review)
• Review AI suggestions (you do this)
• Incorporate changes and revise (you do this)
• Disclose AI use appropriately (you do this)
• This workflow gains efficiency without sacrificing responsibility
• Different for different writing tasks: dissertation vs. paper vs. conference abstract
• Different for different sections: methods can use more AI; results require more human verification

—

Practical Research Use Cases

Scenario: You need to write a methods section for a paper on a standard immunoassay procedure with minor modifications.

Appropriate AI use:

• You outline key steps
• You ask AI to draft a standard immunoassay methods description
• You review the draft and verify it's technically accurate
• You add your specific modifications and parameters
• You use AI to improve clarity and check for grammatical issues
• You verify all technical details one more time
• Final product: you're responsible for all technical content; AI helped with drafting routine sections

Why this works: Methods sections often follow standard formats. AI can generate standard description. You verify it matches your actual procedure. This saves time without risking accuracy because you remain responsible.

Scenario: You have novel findings that need careful interpretation in the context of existing literature.

Appropriate AI use:

• You write initial draft explaining findings and connecting to literature
• You ask AI to improve clarity and suggest structural improvements
• You review suggestions and keep what improves readability
• You add novel insights and critical interpretation (this is yours, not AI)
• You DON'T ask AI to generate interpretations you haven't verified against literature
• Final product: your thinking with AI-enhanced clarity

Inappropriate AI use:

• You ask AI to write your discussion explaining your findings
• You review it briefly and submit with minimal revision
• You risk AI generating plausible-sounding but inaccurate interpretations
• You're outsourcing thinking to AI

Why the distinction matters: Discussion is where your expertise matters most. AI should enhance your thinking, not replace it.

Scenario: You need to write a background section synthesizing 30 papers on your topic.

Appropriate AI use:

• You read the 30 papers and understand key findings and disagreements
• You write an outline organizing papers by theme
• You use AI to suggest ways to structure themes coherently
• You draft key synthesis statements with AI helping to make them concise and clear
• You verify every citation against the original paper (hallucination check)
• You ensure synthesis accurately represents the literature
• Final product: your understanding of literature with AI-enhanced clarity

Inappropriate AI use:

• You paste 30 papers into AI and ask it to summarize them
• You copy the AI summary into your paper with light revision
• You risk missing important nuances and hallucinated citations
• Your synthesis is AI-generated, not your understanding

Why this matters: Literature synthesis demonstrates your grasp of the field. AI can assist but shouldn't replace your engagement.

Scenario: You have a complete paper and need to write a 250-word abstract.

Appropriate AI use:

• You write a draft abstract capturing key points
• You use AI to check that it's concise and well-structured
• You use AI to improve wording for clarity
• You verify all claims match the paper
• You revise based on AI suggestions
• Final product: your abstract, AI-enhanced for clarity

Why this works: Abstract writing is more routine; clarity is essential; AI excels at both.

—

Hands-On Exercise

Objective: Identify where AI could ethically enhance your writing and where it would create risk.

Steps:

1. Identify your typical writing tasks:

• Papers (methods, results, discussion, introduction, abstract)
• Proposals (aim, significance, approach, timeline)
• Conference abstracts
• Grant applications
• Reports or white papers
• Other field-specific writing

1. For each type of writing, categorize by sections:

• Routine sections (follow standard formats)
• Novel sections (require your unique expertise)
• Highly technical sections (require verification)
• Conceptual sections (require your thinking)

1. Design an appropriate AI workflow for each category:

• Routine + lower-stakes: High AI involvement acceptable (draft, then verify)
• Routine + higher-stakes: Medium AI involvement (AI for structure/clarity, you draft)
• Novel + any stakes: Low AI involvement (AI for clarity only, you create content)

1. Write a personal policy:

• When will I ask AI to draft vs. enhance?
• When must I verify claims?
• How will I disclose AI use?
• Who do I need to check my policy with (advisor, institution, journal)?

1. Create a template disclosure statement:

• Draft a disclosure statement appropriate for your field
• Example: "AI language models were used to enhance clarity in the methods and improve grammar throughout the manuscript; all scientific content was verified by the authors."

1. Reflect on integrity:

• For each type of writing, would you be comfortable disclosing your AI use to editors, advisors, or colleagues?
• If not, you might be using AI inappropriately

Time required: 45-60 minutes

—

Common Mistakes and Misconceptions

This rejects useful tools. Using AI to improve clarity of sentences you wrote is not wrong. Using it to generate content you barely review is wrong. The distinction is engagement with content and responsibility for accuracy.

Disclosures should reflect meaningful AI involvement. Running your paper through Grammarly doesn't need disclosure. Using AI to draft major sections does. Disclosures are for substantive AI contributions, not routine tool use.

If you wrote something and asked AI to make it clearer, you wrote it. If you asked AI to write something and tweaked it, AI wrote it. The distinction is who created the content. Improvement of your content is your content; generation of content is AI's contribution.

Lack of prohibition is not permission. The ethical test is: would I disclose this use? If the answer is no, it's ethically questionable. Emerging norms are shifting rapidly; conservative approach is safer.

Norms are moving toward transparency. When in doubt, disclose. Overdisclosure of minor AI use is safer than underdisclosure of major AI use. Better to be transparent and appear cautious than hide AI involvement.

—

Key Takeaways

• Enhancement (improving content you created) is appropriate; generation (having AI create content you minimally review) is ethically problematic and violates integrity when undisclosed
• AI works well for routine writing tasks (clarity, grammar, structure, formatting, outlining, initial drafting of standard procedures) and poorly for tasks requiring expertise and novel thinking
• Verification of content remains your responsibility: all factual claims, citations, and interpretations must be checked before publication
• Emerging norms require disclosure of substantive AI use in writing; transparency about how AI was used is becoming standard in journals and institutions
• Effective AI integration maintains researcher responsibility: AI assists, enhances, and amplifies your work while you remain the author and authority
• The integrity test: would you disclose this AI use? If the answer is no, you're likely using AI inappropriately

—

Reflection Questions

1. Your current writing practice: For your typical research writing, which tasks take most time and feel most routine? Which tasks require your deepest expertise? How could AI assist with the former without replacing the latter?

1. Verification workflow: For your field's standard paper sections, how would you verify content if you used AI for initial drafting? What would need human expertise versus what could AI help check?

1. Disclosure strategy: How will you decide what AI use to disclose in your research? What's your personal policy that's more conservative than you think absolutely necessary?

1. Team writing: If you work with collaborators, how would you establish norms about AI use in collaborative writing? How would you ensure everyone's comfortable with how AI is used and disclosed?