An interesting article from the Journal of the Medical Library Association
How would a scientific paper be titled if the title were written by ChatGPT instead of the researcher?
This is the simple but intriguing question explored in an interesting article published in the Journal of the Medical Library Association (JMLA).
The study compares the titles of 50 medical research articles written by their original authors with titles generated by ChatGPT-4.0 using the same abstracts. The results reveal some interesting differences in how humans and ChatGPT approach the task of writing a scientific title.
The article:
Sebo P, Nie B, Wang T. How ChatGPT writes scientific titles in medical research: structural and content differences compared to human authors. J Med Libr Assoc. 2026;114(3):222–237. doi:10.5195/jmla.2026.2266.
Read the full article on PubMed Central
What did the researchers do?
The researchers selected 50 research articles published in 2000 from 10 high-impact general internal medicine journals. Choosing articles from 2000 meant that the original titles had been written well before the arrival of generative AI.
They then took the structured abstract from each article and submitted it to ChatGPT-4.0 using a standardised prompt, asking the AI to generate a title.
The researchers could then compare two titles describing exactly the same research:
- the original title written by the human authors
- the new title generated by ChatGPT
They examined the titles for both structural features—such as length and punctuation—and content features, including whether the title mentioned the study design, population, outcomes, clinical or public health context, and time period.
ChatGPT’s titles were longer
One of the clearest differences was title length.
The median length of the original, human-written titles was 12.5 words. For the ChatGPT-generated titles, it was 16 words.
That may not sound like a dramatic difference, but the researchers found that the AI-generated titles were statistically significantly longer.
And there was another striking difference: punctuation.
A colon appeared in all 50 ChatGPT-generated titles, compared with only 30% of the human-written titles.
So, if you have ever noticed that ChatGPT seems rather fond of titles containing a colon, this study provides some data to back up that impression!
ChatGPT also made the titles more explicit
The differences were not only about length and punctuation.
ChatGPT-generated titles were more likely to explicitly mention the study design, provide details about the population, and emphasise the outcomes of the research.
The AI also more frequently framed the findings in clinical or public-health terms and incorporated temporal information.
In other words, ChatGPT tended to produce titles that were more descriptive and structured, making more of the information contained in the abstract visible in the title itself.
This is particularly interesting because a title has to perform several jobs at once. It should tell the reader what the article is about, attract attention, and distinguish the paper from other research. It is also one of the first pieces of information encountered when we search the scientific literature
So, does ChatGPT write better titles?
Not necessarily—and that was not what the study set out to establish.
The researchers were looking at differences, rather than deciding whether human or AI-generated titles were better.
A longer, more explicit title may provide more information to the reader. On the other hand, a concise title may be easier to read and remember.
What makes a “good” scientific title is not simply a matter of adding more information.
What this study does show is that ChatGPT has a recognisable way of approaching the task. Given the same abstract, it tends to produce titles that are longer, more explicit and more structured than those originally written by human researchers.
Why did we find this article interesting?
We found this study interesting because it takes something that researchers do routinely—writing a title—and examines how an AI approaches the same task.
As AI becomes increasingly involved in scientific writing, it is useful to look not only at whether AI can produce text, but also at how its approach differs from that of human authors.
And there is another reason this caught our attention.
Scientific titles are not merely labels. They are an important part of how research is communicated and discovered. The authors themselves point out that titles play a role in research visibility, interpretation and discoverability.
Could the increasing use of AI-generated titles eventually influence the way scientific articles are titled—and therefore the way researchers encounter and discover them?
That remains to be seen.
For now, this paper provides a fascinating snapshot of the way ChatGPT approaches one small but important part of scientific communication.
An interesting read for anyone involved in medical research, scientific writing—or simply curious about how AI is changing the way we communicate science.
Read the article in PubMed Central
ChatGPT #ArtificialIntelligence #ScientificWriting #MedicalWriting #MedicalResearch #ScientificPublishing #AcademicWriting #Research