6 ways to spot an AI-generated medical paper
You can spot an AI-generated medical paper by
- Meaningless figures and tables
- Garbled text in figures and tables
- Suspiciously smooth or clean data
- Fabricated citations and references
- Factual absurdities and inconsistencies
- Chunks of empty text
Paper mills have been around in science for quite a while, but have really accelerated after generative AI, especially all-purpose, free large language models (LLMs) like ChatGPT, became mainstream. Fake, AI-generated scientific manuscripts, commonly termed AI slop, are clogging up journal submission systems and frustrating editors and peer reviewers. And as some slip past editorial checks, these fake papers are increasingly polluting the scientific literature too. This confuses other researchers trying to genuinely expand their knowledge and erodes public trust in science.
Why is it difficult to spot AI-generated medical papers
The classic signs of AI writing are not so obvious in medical research papers, which makes it difficult to flag papers that are AI slop. Research articles are usually impersonal, with no scope for personal opinions or anecdotes. They follow a set prescribed structure. Terminology can’t be varied much without compromising on accuracy. And for standard procedures (like informed consent or data analysis), there are only so many ways you can phrase a sentence.
How to tell if a medical paper is AI-generated?
Here are some key giveaways that the paper you’re reading is a complete AI fabrication and unlikely to report a real study:
1. Meaningless figures and tables
Figures and tables are meant to convey data in a research paper, so a fellow researcher in the field should be able to gain actual insights from them. A heatmap that looks like a preschooler’s art project or a micrograph that looks like a teenager’s first attempt at making an omelette? Probably AI-generated.
2. Garbled text in figures and tables
Another dead giveaway for AI is garbled or random text in figures and tables. When you zoom into a forest plot or representation of a signaling pathway, you should be able to read clear and meaningful text, not someone’s Scrabble rack.
3. Suspiciously smooth or clean data
Data from real studies contains noise, gaps, and outliers, regardless of how well the study was conducted. Be suspicious of longitudinal studies with no loss to follow up, or surveys with 100% valid response rates. Unrealistic data is another indicator of an AI-generated paper.
4. Fabricated citations and references
LLMs are infamous for generating completely made-up references. If you find that a large number of papers cited in the article don’t actually exist, it’s a clear sign that an LLM was used to write the paper.
5. Factual absurdities and inconsistencies
General-purpose LLMs often don’t get terminology right or make all kinds of factual errors when they use terminology. If ultrasound is being conducted at 30 Hz, there are 1000 cycles of RT-PCR for a standard diagnostic test, or correlation coefficients are whole numbers, it’s probably AI.
6. Chunks of empty text
A classic sign of AI writing is impeccably phrased text that doesn’t actually contain any meaningful insights or observations. Researchers can be long-winded and try to cram too many topics into the same paper. But if you’re reading the same paragraph multiple times to try to figure out what it means, despite correct grammar and sentence construction, you could be looking at AI slop.
Frequently Asked Questions
1. What tools can help identify AI-generated papers?
There are quite a few AI detectors available, of which I’d recommend Paperpal’s free AI detector because it’s trained on millions of real research papers. Most mainstream AI detectors like GPTZero don’t perform so well on academic papers, because academic writing is meant to be structured and use set terminology. So mainstream AI detectors come up with a lot of false positives in academic papers: identifying text as AI when it’s not AI-generated.
2. What should I do if my research paper has a high AI score?
A high AI score doesn’t mean that your paper will be automatically rejected. You should be able to provide enough proof that your paper is not AI fabricated, by offering to share your raw data, lab notes, etc. with the journal editor and peer reviewers.
3. How can I reduce the AI score of my research paper?
You can reduce your research paper’s AI score to some extent by
- Adding your own critique or opinions where relevant
- Mentioning actual details of studies that you cite (e.g., at the 6-month follow-up, Duckylegs et al. found …)
- Varying sentence length
- Changing the passive voice to active wherever it makes sense
- Refer to specific findings from your own study, instead of broad general observations, especially in the Discussion section
