AI Transparency, Plagiarism & Originality: A Complete Guide for Academics

TL;DR

AI use in academia is shifting from bans to regulated disclosure; researchers should plan for transparency rather than secrecy.

Briefing Cornell Notes

Briefing

AI use in academia is no longer a question of whether it will happen, but how researchers keep intellectual ownership while meeting disclosure and integrity rules. Emanuel Sele, a professor at Lancaster University, frames responsible AI writing as an integrity problem with practical guardrails: keep your ideas, analysis, and voice as yours; use AI as an assistant for mechanical tasks; and disclose any AI contribution. Publishers and universities are increasingly moving from outright bans to regulated disclosure, including requirements that AI not be listed as a co-author but must be transparently reported when it contributes to submitted work.

Sele draws a clear line between ethical AI support and academic misconduct. He describes a “traffic light” approach used by many institutions: green-light modules where AI use is expected and assessed as part of learning; amber areas with restrictions; and red zones where certain AI practices are not authorized. Across all colors, the central safety rule is disclosure: if AI contributed in any way, it should be declared. Major journals now commonly require AI disclosure at submission, and tools are beginning to embed templates and workflows to reduce friction—so authors can specify what AI was used for (e.g., brainstorming, outlining, paraphrasing, or adapting text) rather than leaving reviewers to guess.

Plagiarism risk has evolved with AI. Traditional plagiarism—copying text, ideas, or quotes without proper attribution—extends to AI-generated content that is presented as one’s own voice or one’s own ideas. Sele highlights common failure modes: over-reliance on paraphrasing without understanding, inadequate citation when information is synthesized from multiple sources, and summaries that subtly distort context or numbers. He warns that AI detectors and similarity scores are not the same thing as plagiarism judgments; similarity thresholds vary by institution and context, while AI detection can be unreliable and policy-dependent.

To prevent misconduct, Sele offers three practical strategies. First is the “explain test”: if a researcher cannot explain every claim without looking at AI output, the work likely wasn’t truly understood. Second is “source first, AI second”: AI can help discover literature—especially interdisciplinary work—but key references must be independently validated, using a triage-like approach to confirm authenticity and alignment with original papers. Third is maintaining an “AI interaction log” and version history—recording tools used, prompts, and how outputs were modified—so disclosure is accurate and defensible if challenged.

On originality and voice, Sele argues that AI-written text often shares similar sentence structures, which can make writing look interchangeable. His technique is “write first and polish later”: draft ideas in one’s own words (even as bullet points), then use AI for grammar, clarity, structure, and tone—not for generating the argument from scratch. He also stresses verification: summaries can misstate study populations or outcomes, so researchers should cross-check AI-derived claims against the source paper.

In Q&A, Sele addresses edge cases such as non-native speakers editing abstracts, using AI for literature mapping, and whether AI should do peer review. His consistent answer: editing and assistance are acceptable when they support understanding and are disclosed; peer review still requires human judgment about novelty and impact; and avoiding problems comes from rewriting in one’s own voice, verifying facts and references, and being transparent about AI use. The bottom line is that AI can boost productivity, but academic authorship—and the ability to defend claims—must remain human.

Cornell Notes

AI in academia is here to stay, and the safest path is regulated, transparent use rather than prohibition. Emanuel Sele says integrity depends on keeping one’s ideas, analysis, and voice as the author’s own, while using AI for mechanical support like grammar, structure, and reference organization. Disclosure is the key protection: if AI contributed to any part of the work, it should be declared using institution- and publisher-approved templates, and AI must not be listed as a co-author. To avoid plagiarism and “AI-like” writing, he recommends the explain test (can you defend every claim?), source-first validation (verify AI-suggested papers and numbers), and maintaining an AI interaction log plus version history. Originality comes from drafting first in one’s own words, then polishing—never outsourcing the argument.

What does “responsible AI use” mean in academic writing, beyond simply avoiding plagiarism?

Sele frames it as honest research with three ownership anchors: your ideas, your analysis, and your voice must remain yours. AI should function as an assistant that augments critical thinking, not a co-author that replaces intellectual work. He also emphasizes boundaries via a “traffic light” policy model used by many universities: green-light modules where AI use is expected and assessed, amber areas with restrictions, and red zones where certain AI practices aren’t authorized. Across all cases, disclosure is the non-negotiable rule—if AI contributed in any way, it must be declared.

How should researchers handle AI disclosure when publishers or universities require it?

Sele says disclosure should be specific and aligned with templates provided by publishers or tools. Major publishers increasingly require AI disclosure at submission, and guidelines typically prohibit listing AI as a co-author while requiring transparent reporting of AI’s role. In practice, disclosures often describe what AI was used for (e.g., brainstorming, outlining, paraphrasing, reviewing/adapting text). In Q&A, he adds that if prompt-level detail isn’t explicitly required, providing more context is safer: tool name, purpose, and—if available—what prompts were used and how outputs were incorporated.

Why can similarity scores or AI detectors be misleading when assessing plagiarism or misconduct?

Sele distinguishes plagiarism/similarity from AI detection. Similarity percentages can be normal in research because concepts and terminology overlap across sources; he cites that early-teen to mid-20% similarity can be acceptable, while very high similarity (e.g., 50%+) is more concerning. AI detectors, meanwhile, are policy- and context-dependent and can produce false flags. The practical response is not to treat detector output as a verdict, but to verify the work, ensure citations are correct, and—if challenged—use evidence like version history and an AI interaction log to show authorship and understanding.

What are the “explain test” and “source first, AI second” strategies used to prevent misconduct?

The explain test: if a researcher can’t explain every claim without consulting AI output, they likely don’t understand the material well enough to defend it. Sele recommends rewriting in one’s own words to force comprehension. Source first, AI second: AI can help discover literature (especially interdisciplinary work via semantic search), but key references must be independently validated. He uses a triage analogy—checking whether AI-suggested sources are authentic and whether AI-derived claims match what the original papers actually say.

How can researchers maintain originality and avoid writing that looks “AI-generated”?

Sele’s core method is “write first and polish later.” Draft ideas and argument structure in one’s own words (bullet points are fine), then use AI mainly for grammar, clarity, structure, and academic tone. He warns against accepting AI-generated arguments or analysis wholesale because it undermines the ability to explain claims in oral tests (mini vivas) and can produce writing that sounds like everyone else. A final check is whether the finished text sounds like the author; if it doesn’t, that’s an alarm bell.

What should a researcher log while using AI tools during writing?

Sele recommends keeping a simple, auditable record: which AI tools were used, what prompts were entered, and how outputs were modified or whether summaries were used. He also advises saving different versions of the work as it evolves. This supports accurate disclosure and provides evidence if reviewers or institutions question whether AI was used appropriately.

Review Questions

Which parts of academic authorship must remain the researcher’s own under Sele’s framework, and which parts can AI assist with?
How do the explain test and source-first validation reduce both plagiarism risk and AI-detector false-flag problems?
Why does Sele treat AI detection scores differently from similarity scores, and what practical steps follow from that distinction?

Key Points

1
AI use in academia is shifting from bans to regulated disclosure; researchers should plan for transparency rather than secrecy.
2
Keep ownership of ideas, analysis, and voice; use AI for mechanical support like grammar, structure, and reference organization.
3
Disclose any AI contribution using institution/publisher templates, and never list AI as a co-author.
4
Avoid misconduct by rewriting in your own words, passing the explain test, and verifying AI-derived facts and citations against original sources.
5
Use AI to discover literature, but validate key references independently—especially when numbers or context could be distorted in summaries.
6
Maintain an AI interaction log and version history so disclosure is accurate and authorship can be defended if challenged.
7
Similarity percentages can be normal in research; AI detector scores are unreliable and should not be treated as a plagiarism verdict.

Highlights

Disclosure is the main protection: if AI contributed to any part of the work, it should be declared using approved templates.

The explain test—being able to defend every claim without looking at AI output—functions as a practical integrity check.

“Write first and polish later” is the originality strategy: draft the argument in your own words, then use AI for refinement.

AI detectors and similarity scores measure different things; policy and verification matter more than a single percentage.

Keeping an AI interaction log and version history makes disclosure and authorship defensible under scrutiny.

Topics

Mentioned

PayPal
Paperpal
Turnitin
ZeroGPT
Grammarly
Quillbot
Typeset
Litmaps
Emanuel Sele
Nolita Kolo