Writing with ChatGPT:

An Illustration of its Capacity,

Limitations & Implications
for Academic Writers THE WRITER’S CRAFT

LORELEI LINGARD

In the writer’s craft section we offer simple tips to improve your writing in one of
three areas: Energy, Clarity and Persuasiveness. Each entry focuses on a key writing
feature or strategy, illustrates how it commonly goes wrong, teaches the grammatical
underpinnings necessary to understand it and offers suggestions to wield it effectively.
We encourage readers to share comments on or suggestions for this section on Twitter,
using the hashtag: #how’syourwriting?

ChatGPT and other artificial intelligence (AI) tools are raising alarm bells across academia.
Much of the alarm centers on how ChatGPT will affect the educational mission. How will it
affect student learning? Will it lead to rampant student cheating? Will it mean the death of
traditional knowledge assessments [1]? Recently, the alarm has reached our scholarly mission
as well. Is it a new technological resource, or a threat to scientific integrity? What uses are
appropriate, and how should they be acknowledged?
These are not abstract questions. ChatGPT has already been credited with authorship in
preprints and peer-reviewed published articles since January 2023 [2]. Concerns have been
raised about its uncredited or fraudulent use [3], and major journals are now declaring their
positions on the issue. For instance, the Springer Nature journals have declared that ChatGPT
cannot be a co-author because it cannot take responsibility for the work, and they require that
researchers document any use of ChatGPT in their Methods or Acknowledgements sections
[4]. Academic Medicine guides authors to disclose the use of AI tools in scholarship, describe
transparently the nature of that use, and be aware of limitations that affect accuracy and
CORRESPONDING AUTHOR:
integrity [5]. A recent systematic review in the domain of healthcare education, research and
practice acknowledged ChatGPT’s promise but concluded that it should be embraced with Lorelei Lingard

“extreme caution” considering concerns with “ethical, copyright, transparency, and legal issues, Western University, CA
the risk of bias, plagiarism, lack of originality, inaccurate content with risk of hallucination, [email protected]
limited knowledge, incorrect citations, cybersecurity issues, and risk of infodemics.” [6].
We already use technology to assist our research and writing. Imagine how you’d function
without SPSS or NVivo to manage your data analysis, Reference Manager to organize your TO CITE THIS ARTICLE:
citations, or Grammarly editing software to correct your spelling and grammar. This Writer’s Lingard L. Writing with ChatGPT:
Craft aims to familiarize writers with ChatGPT so that they might use it effectively and An Illustration of its Capacity,
Limitations & Implications for
appropriately. Drawing on chats I had with ChatGPT4 in March and April 2023 to illustrate its
Academic Writers. Perspectives
capacity and its limitations, I extract a series of Cautions and Insights and guide writers in on Medical Education. 2023;
how to use incremental prompting to ‘train’ the software, how to use it for brainstorming and 12(1): 261–270. DOI: https://doi.
generating content like outlines and summaries, and how to employ it as an editor. org/10.5334/pme.1072
Lingard Perspectives on Medical Education DOI: 10.5334/pme.1072 262

WHAT IS CHATGPT, EXACTLY?

ChatGPT is an AI large language model. Current versions

(3.5 and 4.0) have been pre-trained on massive amounts
of data pre-2021, allowing it to learn language patterns
and associations which it uses to generate human-like
conversational text when prompted. When you input a
prompt into ChatGPT, it analyzes the input and rapidly
generates a response drawn from information acquired
via machine learning in engagement with the internet. It
excels at processing, distilling and presenting information That’s somewhat better: it’s talking about style and tone,
verbally in human-like text [7]; it has the potential, and the example of pronouns is more specific. But it is still
therefore, to be an important asset for academic writers, more generic than I would like. I ask it to:
particularly if we can outsource some of the labor of
writing [8] given time constraints and other resource
limitations.
ChatGPT generates responses that are grammatically
correct and semantically meaningful. They are not,
however, always accurate. This is because AI large
language models don’t have ‘knowledge’ in the usual
sense of the word; they don’t store or retrieve data; they
don’t crawl the web like a search engine for information.
They are “just good at predicting the next word(s) in a
sequence” [9] based on what they have learned. Thus,
ChatGPT has “uneven factual accuracy” [10]; moreover,
it does not try to ensure that the content of its text is
true, robust, verifiably, valid, generalizable, etc. [11]. In
machine learning terms, it can “hallucinate”, confidently
presenting legitimate-sounding material that it is not
real [12].

TRAINING CHATGPT THROUGH This response employs more formal linguistic features to
INCREMENTAL PROMPTING talk about voice in academic writing, and offers illustrations
of deixis, modality, and coherence. However, these aren’t
ChatGPT’s default is to offer generic and descriptive the primary features that scholars are concerned with
responses. Incremental prompting is the process by which when they talk about academic voice. So my next prompt
you gradually focus its attention and train it to give you asks about a prominent linguistic scholar of academic voice:
responses tailored to your interests and your level of
understanding. General prompts are fine as a starting
point, such as my question below:

This is accurate, but still generic and selective:

It goes on to explain active and passive voice, but that’s not Hyland’s theory has three main components, only one
really what I’m interested in. I try to get a more specific of which is explicitly represented in its response (“stance”).
answer by telling it about myself: My next incremental prompt tries to focus it further:
Lingard Perspectives on Medical Education DOI: 10.5334/pme.1072 263

USING CHATGPT FOR BRAINSTORMING

ChatGPT is better at some things than others. The main

distinction is between content generation and structural
support. In terms of content generation, as the previous
ChatGPT is now able, in a few seconds, to offer a 360-word conversation about academic writing voice illustrated, you
description of Hyland’s theory with helpful, illustrative need to know the content well to judge the quality of what
examples of its three components. But now, I wonder, what ChatGPT is giving you. One form of content generation that
other linguistic theories of voice did it not tell me about, due the software has been identified to be good at is the
to my focusing in on Hyland in my prompting? So I ask: creation of outlines [13]. I wanted it to provide an outline
for this Writer’s Craft for me, so I began by asking what it
knew about the Writer’s Craft genre:

Now ChatGPT describes four important theories of voice, which

I could probe further with incremental prompting. I could
also ask it to compare these theories, or to describe tensions Stop right there: the Writer’s Craft is not a series of books,
among them, or to suggest how more recent theorists have and I have never written books with these titles. As this
built on them. Once you get to this level of specificity, ChatGPT illustrates, at each stage of the incremental prompting, you
can be very helpful in pointing out connections among ideas should be alert for false responses. ChatGPT is a text
for you to pursue in your own reading and writing. generator, not a brain: it is putting together words that are
As this conversation illustrates, incremental prompting likely to be found together around the topic you’ve asked
is a key to maximizing the value of ChatGPT’s responses. about. That doesn’t mean these words ‘belong’ together or
But effective incremental prompting requires domain- that they are ‘true’. In fact, ChatGPT seems to enjoy making
specific knowledge: you need to know the domain quite sh*t up. You absolutely cannot trust the references it gives
well already, so that you can judge the extent to which you. And not only because it doesn’t have access to
ChatGPT’s response is accurate, selective or comprehensive, material post-2021; the Writer’s Craft series extends back
and use follow up prompts to improve it. If you ask it about to 2015, but it doesn’t know them and apparently doesn’t
domains you don’t know well, it will be unclear what try to find them. Instead, it “hallucinates”, making up a
information ChatGPT has selected from, and why it has book series. Happily, you can stop it in its tracks when it
selected some things and not others. does this, by hitting the “Stop Generating” button, which I
One final point about ‘training’ ChatGPT: what it learns did. Then I corrected it:
does not appear to consistently transfer beyond the current
chat. When I opened a separate chat and asked again “What
do you understand about Ken Hyland’s theory of academic
voice in writing”, the first response included described all
three features of this model: that seemed to represent
learning transferred from the previous chat. But this is not
always the case, as I will discuss in the next section.

Caution: ChatGPT’s default responses are both generic

and (invisibly) selective.

Insight: Incremental prompting can increase specificity,

guide selections & reveal exclusions.
Lingard Perspectives on Medical Education DOI: 10.5334/pme.1072 264

This answer too, is mostly repetition of what I told it in the Some of these are pretty good: I especially like 1 and 6, and
prompt, so I try to focus it down and get specifics by asking if I were going to write a new title it would likely combine
for the titles of the Writer’s Craft series: pieces of these favorites.
Back to my attempt to get ChatGPT to create an outline
for this Writer’s Craft on ChatGPT.
I tried to get it to familiarize itself with actual Writer’s
Craft pieces, but it resisted.

The problem is, these aren’t real titles: they’re made up.
This is one of ChatGPT’s main weaknesses: when you prompt At first I’m puzzled that it resists actually reading the entire
it to provide specific references, it often invents them [14]. series, as it surely does have the ability and the time (!).
Because these inventions are based on word associations, Perhaps if I gave it all the references it would respond
they will have realistic combinations of keywords and better. Here though, it reverts again to generalities: “The
coauthors. But they are as likely as not to be bogus. series consists of approximately 20 articles, each of which
Techniques for addressing this problem are emerging, such likely focuses on a different aspect of academic writing…”;
as extensions that improve ChatGPT’s access to the Web or The series likely provides practical advice and strategies for
restrict it to Google Scholar references; however, reports of researchers to use in their academic writing…”, and so on.
their effectiveness vary [15]. So far, I can’t have confidence that ChatGPT has actually
read any of the Writer’s Crafts, even the specific one that I
Caution: ChatGPT will lie to you, a phenomenon known provided as an example. But then I remember that ChatGPT
as an AI hallucination. is neither reading nor analyzing – it’s just recognizing
language patterns. Thus, I prompt again:
Insight: Double check any content it generates,
particularly references.

I will say, though, that ChatGPT is rather creative when it

comes to titles. I wish, for instance, that I had written this
fake Writer’s Craft: “Lingard L. Convince me: the art of
argumentation. Perspectives on Medical Education. 2013;
2(2): 75–78.” This suggests that one way to use the
software effectively would be to ask it for title suggestions.
As an experiment, I give it the introductory paragraphs
from a recent Writer’s Craft (that I had entitled “Writing for
the Reader: Using Reader Expectation Principles to Maximize
Clarity) and I ask it for possible titles: That’s getting closer. There is sufficient detail that I can tell
ChatGPT has picked up the main structure of this Writer’s
Craft. Notice though, how important my own existing
knowledge is: I know this article well (I wrote it!) and so I
have been able to readily discern when ChatGPT is spouting
generic stuff and when it is actually talking about the
article(s) I’ve directed it to.

Caution: Don’t rely on ChatGPT to read articles for you.

Insight: Use ChatGPT in relation to familiar domains
rather than unfamiliar ones.
Lingard Perspectives on Medical Education DOI: 10.5334/pme.1072 265

Now that I’ve (finally) gotten it to recognize the features of What if you wanted help weaving in some additional ideas,
a specific Writer’s Craft, I can get to my original aim: getting to deepen the piece of writing? Just ask, remembering to
ChatGPT to brainstorm an outline for this paper: be as specific as you can be:

These are all relevant points, and they provide me with

search terms I could input into Google Scholar to round out
my understanding of each
As my chat above illustrates, ChatGPT can be used to
create solid outlines. You need to train it on the genre you’re
going to write in and you need to judge its knowledge
about the subject areas you’ll cover, but once you’ve taken
those steps you can quickly request a series of outlines
with different orders, sections, emphases. This can help
you imagine different ways of approaching the manuscript:
This is useful. Admittedly, it’s not precisely the format from choose the best one, and start drafting.
the Writer’s Craft I trained it on: it has reverted to a somewhat ChatGPT is fast once you get it pointed in the right
more generic structure. And it is not (you will have noticed) direction (this whole chat took less than 10 minutes), but
how I actually decided to organize this piece. It does, that training effort doesn’t transfer to new chats. The
however, help me conceptualize some of the necessary system saves all your chatlogs: you can see them on the
sections. And it was useful to return to during the drafting sidebar and go back and access them, but they are discrete
process, to see what I was emphasizing and overlooking. entities. “Contextual memory only applies to your current
Once you get ChatGPT to this point through incremental conversation. ChatGPT’s stateless architecture treats
prompting, you’re on the cusp of all kinds of brainstorming conversations as independent instances; it can’t reference
bounty. Let’s say you want some specifics to help you flush information from previous ones. Starting new chats always
out each section: resets the model’s state” [16]. Not knowing this, a few days
later I started a new chat and asked it again “Tell me what
you know about the Writer’s Craft series written
(predominantly) by Lorelei Lingard to help researchers
improve their academic writing”, only to be told again
about 5 books I had never written. When I went back into
saved chat logs and picked up my prompting where I’d left
off, the result was better but not consistently so, which
may be due either to limits on ChatGPT’s contextual
memory or to its tendency to ”break character” due to
“dropping instructions it deems irrelevant” [16].

Caution: ChatGPT doesn’t transfer the training you’ve

Insight: If you dread the blank page, a ChatGPT outline
done across chats.
could jumpstart your drafting process.
Insight: Try returning to saved chat logs; you may be
Caution: Its ideas are generic; use them as a starting able to build on the training you’ve done through previous
place, not a replacement for your own. prompting.
Lingard Perspectives on Medical Education DOI: 10.5334/pme.1072 266

GENERATING COUNTERARGUMENTS, Insight: ChatGPT can be a good brainstorming resource.

SUMMARIES, AND ABSTRACTS Caution: But don’t accept its suggestions blindly. You are
smarter than it is (at least in its current form).
Outlines are not the only useful way to use ChatGPT for
content generation. You can also ask it to review a section AI tools like ChatGPT can also help you to make your writing
of your argument and suggest counter arguments. Keep in more accessible and inclusive of a wider audience. For
mind that the version of ChatGPT you use matters. The free instance, I gave it the opening paragraphs of a recent grant
ChatGPT 3.5 has a limit of about 500 words on what it can application and asked it to generate a 100-word lay
read and respond to, so if you input your whole results or summary, to which it responded:
discussion section you’ll get this error message:

ChatGPTPlus (the paid version) is supposed to handle up

to 25,000 words at a time, but I still received the error
message when I tried to input more than a few paragraphs
for it to read and respond to. Therefore, I think it’s better
to give it a rough summary of a section (say, while you’re This was still a bit formal, suggesting that the meaning of
still drafting it) and ask it for counterarguments. In this ‘lay summary’ was perhaps not self-evident in my initial
prompt I summarize the gist of the results of a paper I’m prompt, so I prompted it further:
currently working on, tell ChatGPT what I want to argue
based on those results, and ask it to suggest
counterarguments:

In about 3 seconds (it still leaves me breathless how quickly This is an acceptable first draft that I can now
it works), ChatGPT offers this: rework; for instance, I would probably reinstate a
few of the keywords (like “scientific integrity”) that
were removed. But having a workable draft to start
from has probably saved me at least 30 minutes of
work.
Abstracts are another piece of writing labor that
ChatGPT can help with. It can’t help you with your first
abstract draft, because it can’t read your entire paper due
to limits on the words you can enter in a prompt. But it can
help you take your abstract’s first draft, which is invariably
too long, and reduce it to the required word limit. This is
labor most of us would be happy to outsource: I have yet
to meet a writer who cherishes the task (and time!) of
whittling words from their abstract. But even with this
This is enough to get my wheels turning: I could readily task which seems perfectly suited to AI, you can’t entirely
start writing a ‘counterargument’ section of my discussion. trust it.
If I don’t understand fully some of these ideas or I want As illustration, I asked it to remove 27 words from a
more specificity, further prompting would focus these structured abstract which I provided, adding that I wanted
responses. it to rewrite as little as possible as I liked the content as it
Lingard Perspectives on Medical Education DOI: 10.5334/pme.1072 267

was. It produced an unstructured version about 100 words the key ideas. I’ll trim the extra 8 words when I rework
long. I responded: this version; that’s certainly faster than doing the whole
reformat myself.

Now it returned a structured abstract, but still much shorter EDITING WITH CHATGPT TO IMPROVE
than I had requested. I prompted again (I confess, a bit CLARITY AND COHERENCE
irritated), this time not asking it to do the mathematical
task of subtracting the number of words, but setting a word Another way to use the tool is to strengthen the clarity
limit (300) for its response: and coherence of sections of your draft, particularly those
dense spots where you think you might lose the reader.
To strengthen internal coherence, you could input a single
paragraph and ask it to rewrite so that the ideas develop
more convincingly, including suggesting where you should
add token sentences to illustrate your points. I inputted a
paragraph I’d drafted and asked it for three possible topic
It seems like it understands, but the next version wasn’t sentences, to which it responded:
300 words either (yes, by this time I was copying and
pasting all the attempts into a Word document to check
the word count):

Seeing the different emphases in each topic sentence

helped me to identify the issues that were vying for
attention in the paragraph, and make it more coherent.
Finally, success: the last version was 310 words (we’ll give
ChatGPT the point, because 10 of those were the structured Insight: Asking ChatGPT for topic sentences can help
headings). reveal issues with paragraph coherence.
Caution: Don’t use those sentences verbatim. They are a
Insight: ChatGPT apparently can’t count.
signal, not a solution.
Caution: Even with concrete tasks, be alert for failure.
Theoretically, you can also strengthen external coherence
ChatGPT can count, of course: if you ask it to solve math with ChatGPT, by inputting a series of paragraphs and
problems, it can do so. But exact word counts clearly asking it to suggest new topic and transition sentences. I
aren’t its forte. Nevertheless, ChatGPT is still useful for asked ChatGPT for help with the opening paragraphs I had
reformatting existing abstracts for new purposes. I have a drafted for this Writer’s Craft:
structured conference abstract of 250 words, which I want
to submit to another conference that requires unstructured,
100-word abstracts:

Here’s what it suggested:

It still hasn’t counted accurately (this is 108 words), but it

has reformatted to an unstructured abstract and retained
Lingard Perspectives on Medical Education DOI: 10.5334/pme.1072 268

I noticed two things immediately: first, ChatGPT changed all grammatical/rhetorical/linguistic concepts behind that
the sentences, not only the topic and transition sentences feature of your writing. You don’t know what it “knows”
as requested. And second, it also changed the writing style: until you ask it. I wanted to see if it could help one of my
for instance, it uses many passive voice constructions (e.g., students identify and improve their tendency to write left-
“has sparked”, “has ignited”, “has been acknowledged”), branching sentences (those that introduce a lot of detail
and changed out my simple subjects for more elaborate early, leaving the main idea until late and thus potentially
constructions (e.g., my “ChatGPT” has been changed to creating confusion for the reader who needs the main idea
“the rising influence of ChatGPT within academia”). Now, I to organize all the other details). I started by asking it:
will be the first to admit that I can be a bit precious about
my writing, but this is not what I asked it to do. Thus, I
clarify:

As you can see, it had them backwards. I corrected it:

Ugh. It has overdone – and mixed! – the metaphors. And

the tone has swung from stuffy academic to effusive
adolescent. My next prompt reveals my irritation:

This is closer to my writing style, and the topic and

transition sentences are effective at connecting and
developing the opening argument. If you’re struggling
with internal or external coherence in a piece of writing,
this could be a helpful resource. But if you’re not struggling
with coherence (as I wasn’t particularly in this piece), its
suggestions are unlikely to excite you – and some of them I don’t like all of the sentences it has created, but they are
may frustrate you. more right-branching. This would be a useful coaching
resource, once a writer’s habits are identified and we have
Caution: ChatGPT will edit your writing style as well as ensured that ChatGPT has accurate knowledge of the
your content. grammatical features we’re interested in. For instance,
Insight: As part of your rewrite of ChatGPT-generated many writers struggle to expand their repertoire of strong
material, make the style your own. verbs. We could give ChatGPT a few paragraphs of their
writing and ask it to rewrite with stronger, more dynamic
Whatever you want ChatGPT to help improve in your verbs. Ask it for a few different versions and suddenly you
writing, you need to ensure that it understands the have a nice catalogue of new verbs to choose from.
Lingard Perspectives on Medical Education DOI: 10.5334/pme.1072 269

More generally, ChatGPT could also serve as a free can be used for labor, such as reformatting abstracts or
language editor for scholars writing in English as an reducing the length of sections, but it can’t replace the
additional language (EAL). Many EAL writers now incur the thinking a writer does to determine why some paragraphs
costs (both time/effort and financial) of language editing: it or ideas deserve more words and others can be cut
could alleviate some of those costs, particularly during the back. It can be inaccurate: in fact, rather stubbornly so,
drafting and revision stages, and free writers to focus on persisting with inaccuracies even after they are pointed
the ideas and worry less about the grammar. out, while at the same time presenting its next attempt
as corrected. I know it isn’t sentient and doesn’t have
motivations or emotions, but I can’t help but think in some
A NOTE ON ETHICS of our exchanges that it was being sullen, intractable, even
deliberately insincere.
Much of the alarm about ChatGPT has to do with the ethics Still, writers can harness its power to make our processes
of its use: is it ‘fair’ to have it write for you? As you will have more efficient and our products more robust. Do check
noticed, I don’t advise having it write for you. Most of my your target journal, as policies about writing with AI tools
examples involve putting my own writing into ChatGPT are emerging and evolving. Within journal parameters,
and asking it to make suggestions (here’s my introduction, however, leverage ChatGPT to your advantage. Identify
please suggest some good titles), to do some tiresome the moments in your writing process where you get stuck:
labor (here’s my abstract, please cut it in half), to illustrate can ChatGPT help you there by generating an outline or
grammatical changes (here’s my left branching sentence brainstorming the next points in the storyline? Use it to
pattern, please suggest right branching alternatives). I help address your grammar challenges (e.g., if you default
would argue that these are ethical and appropriate uses of to passive voice, ask it to change sentences to active so
ChatGPT. I’m not asking it to do all the intellectual, creative you can compare); use it to strengthen coherence of
work, I’m outsourcing some of the labor [14]. Where I have a complex section of your argument; get it to increase
asked ChatGPT to create something for me (an outline, clarity by converting your right-branching sentences to
a list of possible counterarguments, a passage improved left-branching. Distinguish the laborious from the creative
with stronger topic and transition sentences), I treat it as a writing tasks: use ChatGPT to support the former, and
starting point for my next round of revisions. This isn’t only keep the latter for yourself. And always view what it has
to avoid presenting ChatGPT’s writing as my own, although generated as a first draft which you will refine and rework,
that’s of course important. It is also because I don’t want infusing it with your own particular emphases, your unique
to outsource the writing craft, which (on some days, at voice and style.
least) gives me joy. And I certainly don’t want to ‘sound’
like ChatGPT – I want my writing to sound like me. Based
on my experiences so far, it will take less time (and be more COMPETING INTERESTS
satisfying) to work on my voice than to work on getting
ChatGPT to mimic me. The author has no competing interests to declare.

IN SUMMARY AUTHOR AFFILIATION

Lorelei Lingard orcid.org/0000-0002-4150-3355
Rather than being alarmed or anxious, writers need to Western University, CA
understand ChatGPT’s strengths and weaknesses. It
is better at structure than it is at content. It is a good
brainstorming tool (think titles, outlines, counter- REFERENCES
arguments), but you must double check everything it tells
you, especially if you’re outside your domain of expertise. 1. Rudolph J, Tan S, Tan S. ChatGPT: Bullshit spewer or the end
It can provide summaries of complex ideas, and connect of traditional assessments in higher education? Jl Applied
them with other ideas, but only if you have put a lot of Learning & Teaching. 6(1). DOI: https://doi.org/10.37074/
thought into the incremental prompting needed to shift jalt.2023.6.1.9
it from its generic default and train it to focus on what 2. Stokel-Walker C. ChatGPT listed as author on
you care about. Its access to information is limited to what research papers: Many scientists disapprove. Nature.
it was originally trained on, therefore your own training 2023; 613: 620–621. DOI: https://doi.org/10.1038/d41586-
phase is essential to identify gaps and inaccuracies. It 023-00107-z
Lingard Perspectives on Medical Education DOI: 10.5334/pme.1072 270

3. Else H. Abstracts written by ChatGPT fool scientists. Nature. 10. Vincent J. AI-generated answers temporarily banned
2023; 613(7944): 423–423. DOI: https://doi.org/10.1038/ on coding Q&A site; 2022, December 5. Stack Overflow.
d41586-023-00056-7 https://www.theverge.com/2022/12/5/23493932/
4. Author guidelines, Springer Nature Press. Guidance on chatgpt-ai-generated-answers-temporarily-banned-stack-
the use of Large Language Models (LLM) e.g. ChatGPT; overflow-llms-dangers.
2023. https://www.springer.com/journal/10584/ 11. Hutson M. Robo-writers: The rise and risks of language-
updates/24013930. generating AI. Nature. 2021; 591(7848): 22–25. DOI: https://
5. DeVilbiss MB, Roberts LW. Artificial Intelligence Tools doi.org/10.1038/d41586-021-00530-0
in Scholarly Publishing: Guidance for Academic Medicine 12. Welborn A. ChatGPT and fake citations. Duke University
Authors. Acad Med; April 28, 2023. DOI: https://doi. Libraries. https://blogs.library.duke.edu/blog/2023/03/09/
org/10.1097/ACM.0000000000005261 chatgpt-and-fake-citations/. Published March 9, 2023.
6. Sallam. ChatGPT Utility in Healthcare Education, Accessed June 1, 2023.
Research, and Practice: Systematic Review on the 13. Bilal M. Become an efficient academic writing with AI apps;
Promising Perspectives and Valid Concerns. Healthcare 2023b. Published tutorial. Accessed on April 20 at: https://
(Basel). 2023; 11(6): 887. DOI: https://doi.org/10.3390/ efficientacademicwriter.carrd.co/.
healthcare11060887 14. King MR, chatGPT. A Conversation on Artificial Intelligence,
7. Pavlik. Collaborating With ChatGPT: Considering the Chatbots, and Plagiarism in Higher Education. Cel Mol Bioeng.
Implications of Generative Artificial Intelligence for 2023; 16: 1–2. DOI: https://doi.org/10.1007/s12195-022-
Journalism and Media Education. Journalism & Mass 00754-8
Communication Educator. 2023; 78(1): 84–93. DOI: https:// 15. Bilal M. ChatGPT generates fake citations to research papers
doi.org/10.1177/10776958221149577 that don’t even exist. But you can make ChatGPT give you
8. Bilal M. Five ChatGPT and Bing Prompts to Polish Your references to (real) published papers; 2023c. Twitter thread,
Writing; 2023a; Newsletter, March 10. April 23rd. Accessed on June 1, 2023.
9. Cooper K. OpenAI GPT-3: Everything you need to know; 2021, 16. Luansing J. Does ChatGPT Learn From User Conversations?
November 1. Springboard. https://www.springboard.com/ 2023. https://www.makeuseof.com/does-chatgpt-learn-
blog/data-science/machine-learning-gpt-3-open-ai/ from-user-conversations/ Accessed June 1, 2023.

TO CITE THIS ARTICLE:

Lingard L. Writing with ChatGPT: An Illustration of its Capacity, Limitations & Implications for Academic Writers. Perspectives on Medical
Education. 2023; 12(1): 261–270. DOI: https://doi.org/10.5334/pme.1072

Submitted: 06 June 2023 Accepted: 06 June 2023 Published: 29 June 2023

COPYRIGHT:
© 2023 The Author(s). This is an open-access article distributed under the terms of the Creative Commons Attribution 4.0 International
License (CC-BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source
are credited. See http://creativecommons.org/licenses/by/4.0/.
Perspectives on Medical Education is a peer-reviewed open access journal published by Ubiquity Press.