How to detect AI use in academic papers
The past year has seen a significant increase in the use of AI tools for researching, drafting, and editing academic papers, and this trend shows no signs of slowing down. Organisations like COPE, DOAJ, and OASPA have declared that while AI can assist with writing, it should never be listed as an author or used to generate the bulk of a paper’s content. As such, journal editors must share strict guidelines with authors clarifying the journal’s stance and policy on AI. Additionally, the journal must provide reviewers with guidelines for identifying potential instances of AI use. This should include how AI can and cannot be used, how it should be declared in the paper, and how the journal will respond if heavy use of AI is detected.
While setting out guidelines can help curb excessive AI use in submissions, journals must still review all accepted articles for AI use. But this leaves journal editors and peer reviewers in a difficult position; you are expected to catch improper use of AI, yet no reliable detection tool currently exists. Most AI detection software works by analysing patterns in language, such as sentence structure, word choice, and predictability. However, these patterns are not unique to AI. Skilled writers, non-native English speakers, and authors who use editing software can produce text that resembles AI-generated content, leading to false positives. Not to mention, as large language models rapidly evolve, they may already be exhibiting new patterns that existing software cannot detect.
The challenge becomes even greater when AI-generated text has been heavily edited by an author. Once a researcher revises, restructures, and supplements AI-generated content with their own ideas, many of the original indicators disappear. Likewise, legitimate uses of AI, such as language editing, translation, or brainstorming, can leave traces in a manuscript without compromising its academic integrity.
For this reason, editors should avoid treating AI detection as a simple pass-or-fail exercise. Rather than relying solely on detection software, journals should evaluate manuscripts holistically, considering the quality of the research, the accuracy of references, the consistency of arguments, and compliance with the journal’s AI policy. In most cases, the goal is not to prove whether AI was used, but to determine whether it was used responsibly and transparently.
However, as AI-assisted writing becomes more common, certain patterns have emerged that may indicate when a manuscript has relied too heavily on AI-generated content. These signs should not be treated as definitive proof, but rather as red flag indicators signalling that further review or clarification may be needed.
KNE CLUE: COMMON SIGNS OF AI USE
Excessive Em Dashes
Many generative models tend to heavily use em dashes to connect sentences or interrupt clauses. While there is nothing wrong with em dashes in academic content, a paper saturated with them can signal AI involvement, especially when the tone feels unnatural.
Non-existent References
One of the most common giveaways is the inclusion of fabricated citations. AI tools can generate seemingly real references that do not exist, combining plausible author names, publication years, and journal titles into convincing yet fictional entries. These are usually generated when users ask AI to “add 10 more references” or “back this up with citations.”
Mismatched References
Sometimes, even when references exist, they are misused. This includes citations that do not support the argument, irrelevant sources, DOIs that lead nowhere, and titles that do not match the listed authors. This inconsistency often arises when AI generates references in isolation without checking accuracy or context.
Inconsistent Punctuation and Style
AI writing often exhibits a mix of British and American English, with the word “colour” appearing alongside “analyze.” There may also be a combination of curly and straight quotation marks. Additionally, switching between ‘single’ and “double” speech marks within the same paragraph can suggest text that has been stitched together from multiple prompts.
Lack of an Original Voice
Some papers have clean grammar and strong cohesion, but they strangely do not sound human. While good writers can produce clean prose and high-quality language, AI often produces a slightly flat, neutral tone throughout.
Overly Perfect Structure
Human writing often contains slight variations in rhythm and emphasis. AI-generated content, by contrast, can be unusually symmetrical, with every paragraph following the same pattern, sentence length, and organisational structure.
Contradictions in the Manuscript
When AI-generated sections are assembled from multiple prompts, inconsistencies can emerge. The introduction may describe one research aim while the discussion addresses another, or key terms may be defined differently throughout the paper.
Vague Methodological Descriptions
AI can describe research methods convincingly without providing enough detail for replication. Statements may appear technically correct yet lack specific information on sample sizes, procedures, software, analytical frameworks, or ethical approval.
Repetitive Content
Generative AI frequently restates the same idea using different wording. Editors may notice that multiple paragraphs contribute little new information and simply rephrase earlier points.
Unsupported Claims
AI systems often present statements confidently, regardless of whether evidence exists. These include strong conclusions, statistics, or factual assertions that lack supporting citations or data.
There is nothing inherently wrong with using AI tools to support research and writing. However, its misuse in an academic setting leading to concerns about overall academic rigour, accuracy, integrity, and originality. The aim of reviewing a paper to identify AI use is not to police high-quality work but rather to ensure that the research ecosystem remains honest and authentic. The most effective safeguard is not a detection tool but a combination of clear policies, vigilant editorial oversight, and a commitment to research integrity. Providing both editors and reviewers with clear guidance on identifying and responding to potential AI misuse helps ensure a more consistent and informed evaluation process. Rather than asking reviewers to detect AI itself, journals should equip them to assess whether the manuscript demonstrates the accuracy, originality, and scholarly rigour expected of published research. As the world evolves, so must our ability to keep pace and grow as critical thinkers.