The Do’s and Don’ts of AI Use in Academic Research and Publishing

Using AI in Academic Publishing When Language Refinement Is Not Allowed: Do’s and Don’ts

Introduction

Artificial intelligence tools have become widely available to researchers, offering quick help with everything from keyword suggestions to similarity checks. At the same time, universities and publishers have reacted by clarifying what is—and is not—permitted in academic work. A growing number of policies now treat any AI-generated, paraphrased, or AI-“refined” text as unauthorised content creation. In other words, if AI touches the wording of a sentence, that sentence may be considered AI-written and therefore unacceptable.

This strict position can surprise researchers who assumed that using AI only for “polishing” or “corrections” was harmless. However, from the perspective of academic integrity, language is part of the intellectual product: word choice, structure, and nuance all contribute to the originality of a scholarly text. Allowing AI to rewrite or improve sentences blurs the line between human authorship and machine-generated content.

This article explains how to navigate this environment safely. It outlines permitted uses of AI that do not involve language creation or correction, such as discovery support and compliance checks, and it details practices that must be avoided, including AI drafting, rewriting, and translation of manuscript content. The goal is to show how AI can remain on the technical periphery of the publishing process while authors retain full responsibility for the language and ideas in their work.

The Do’s: Ethical AI Use Without Touching the Text

1. Use AI for Search and Discovery, Not for Writing

AI can be a useful navigation aid in a crowded information landscape. Within strict policies that prohibit language refinement, AI tools may still be used to:

• suggest keywords and subject headings for database searches;
• propose related topics or concepts that you can explore independently;
• generate lists of potential search queries to use in traditional academic databases.

However, once you have located relevant sources, the reading, note-taking, and summarising must be done by you. AI should not produce written notes, paraphrases, or text that you later reuse. The safest approach is to treat AI as a directional signpost at the search stage, not as a tool that produces any language content for your project.

2. Use AI for Plagiarism and Similarity Screening

Plagiarism detection and similarity analysis are widely accepted, policy-compliant applications of AI. These systems analyse your already written manuscript and report patterns of overlap with previously published works; they do not generate or modify language themselves.

Responsible practices include:

• running your completed draft through tools such as iThenticate or Turnitin (where permitted by your institution);
• examining flagged passages and manually rewriting them yourself to ensure genuine paraphrasing and proper citation;
• checking that reused material (e.g., from your own earlier publications) is clearly referenced and justified in line with journal and institutional policies.

In this workflow, AI acts purely as a diagnostic tool. Only you, the author, rewrite and correct the text in response to its report.

3. Use AI for Structural and Administrative Checks

Some AI-based systems can help verify non-linguistic aspects of your manuscript and submission. Provided they do not rewrite sentences, these uses can be compliant even under strict “no language changes” policies. For example, AI can assist with:

• checking whether all required sections are present (abstract, methods, funding, ethics, etc.);
• detecting missing elements such as data availability statements or conflict-of-interest disclosures;
• flagging inconsistencies in numbering (e.g., figure and table references) or missing captions;
• verifying that a manuscript file meets basic technical formatting parameters (file type, character count, number of figures).

These tools report structural issues; authors must then manually correct the manuscript using their own wording and formatting decisions.

4. Use AI Cautiously for Journal Discovery

AI-driven journal recommendation tools analyse topics, keywords, and references to suggest potentially suitable journals. When language generation and editing are prohibited, these tools can still be useful as long as you treat them as advisory, not authoritative.

Ethical use includes:

• using AI suggestions to create a preliminary list of candidate journals to investigate further;
• manually checking each journal’s aims, scope, indexing status, and publisher reputation;
• consulting supervisors or senior colleagues before deciding where to submit.

Journal finders should never replace human judgement. They are starting points, not submission instructions.

5. Disclose Any Non-Textual AI Assistance

Even when you do not let AI touch language, many publishers now encourage or require transparency about any AI involvement. If you have used AI tools for similarity checking, structural compliance screening, or high-level discovery, it is good practice to state this clearly.

For example, in acknowledgements or a dedicated “AI use” statement you could write:

“The authors used iThenticate for similarity screening and an automated journal discovery tool for initial exploration of potential outlets. All writing, analysis, and final journal selection decisions were made by the authors.”

Such statements reassure editors and readers that AI has not contributed to the content or language of the manuscript itself.

The Don’ts: Practices That Count as AI Content Creation

1. Do Not Use AI to Draft, Rewrite, or “Refine” Manuscript Text

Under a strict interpretation of current policies, any AI involvement in shaping the wording of scholarly text is considered content creation and is therefore not allowed. This includes:

• asking AI to write paragraphs, sections, or entire manuscripts;
• pasting your text into AI tools to “improve wording,” “refine language,” or “make it more academic,” then using the output;
• using AI to paraphrase or rephrase passages from your own or others’ work.

Even if you review and accept only some of the suggested changes, the resulting language is no longer entirely your own. From an integrity standpoint, this undermines the principle that the author is responsible for both ideas and expression.

2. Do Not Use AI for Translation of Manuscript Content

Machine translation is a form of language generation. If you draft a manuscript in one language and use AI to translate it into the submission language, significant portions of the final text will be machine-generated. Under strict policies, this is treated as AI-authored language and is therefore disallowed.

If translation support is needed, institutions increasingly recommend:

• working with human translators who understand the field;
• using professional language-editing services staffed by subject-matter experts;
• ensuring that you review and approve every change while preserving your intellectual intent.

3. Do Not Rely on AI for Content Summaries Used in Your Text

AI-generated summaries may appear as bullet points or short paragraphs. If you then integrate those AI summaries into your own writing—even after light editing—they become part of the manuscript’s content. This is not acceptable under rules that forbid AI-created language.

Instead, you should:

• read each source yourself and produce your own notes in your own words;
• draft summaries based on your understanding, without copying phrasing from AI outputs;
• treat any AI-produced summary as a non-usable prompt, not as text to be adapted for publication.

4. Do Not Allow AI to Generate References, Citations, or Data

AI is known to invent plausible-sounding but false references, and it can also produce synthetic datasets that look realistic but have no basis in actual observations. Using these outputs in a manuscript is a clear form of academic misconduct.

Avoid:

• asking AI to “give references for X” and then copying them into your bibliography;
• using AI to fill gaps in incomplete references instead of consulting the original sources;
• creating or “enhancing” data tables, graphs, or statistics using AI generation.

All references should be drawn from verified academic databases or documents you have located and read. All data should be derived from actual research methods that you can fully document and defend.

5. Do Not Hide AI Involvement—Even When It Seems Minor

Some researchers are tempted to use AI quietly for “minor tweaks” and assume disclosure is unnecessary. Under strict policies, however, the problem is not only the scale but the type of assistance. If AI has generated or rewritten sentences, that is considered content creation, and stating that no AI was used would be misleading.

To maintain transparency:

• do not claim that a manuscript is entirely human-written if AI has contributed any of the wording;
• do not use AI to secretly rework a rejected paper before resubmission;
• if you previously used AI in ways that are now prohibited, seek guidance from your institution before reusing that text.

Working Within Strict Policies: Practical Tips

For many researchers, a total ban on AI language refinement feels restrictive, especially for non-native speakers. Nonetheless, compliance is essential. Some practical strategies include:

• Invest in your own writing skills: take academic writing courses, study style guides, and read high-quality papers in your field.
• Use human support: collaborate with co-authors who are strong writers, or use professional human editing and proofreading services where allowed.
• Plan extra time for revision: without AI rewriting, drafting and refining may take longer—build this into your schedule.
• Clarify local rules: read your institution’s and target journal’s AI policies carefully and ask for clarification if anything is unclear.

Conclusion

AI will continue to influence academic publishing, but strict policies increasingly draw a firm boundary: AI may assist with technical checks and discovery, yet it must not shape the language or substantive content of research manuscripts. Under these rules, even automated “language refinement” is considered content creation and is therefore not allowed.

The safest path for researchers is to treat AI as a background tool—useful for similarity screening, structural checks, and exploratory journal discovery—while keeping all reading, thinking, writing, and rewriting firmly in human hands. With clear awareness of the do’s and don’ts outlined in this article, authors can benefit from modern tools without compromising the originality, accountability, and integrity that define genuine scholarly work.