How to use ChatGPT to edit creative text

Taking on ChatGPT as an enthusiastic assistant editor

Photo credit: AW, used with permission. Puppies and generative AIs are eager to help and need supervision.

I write fiction as a hobby. A challenge for any casual writer is finding reliable sources of feedback for your writing. Often there is no substitute for the “extra pair of eyes” to identify problems. You don’t want to annoy your friends and family, and you want to save their goodwill for the final drafts. That’s why I use Generative AI (GenAI) as a personal writing assistant. GenAI is not a human reader, but it can help me improve my drafts. In this article, I will illustrate how to use GenAI as an editorial assistant. Using a GenAI to help you continuously improve and rewrite your work will help you reach a more effective final draft.

I acknowledge that GenAI can perform some creative writing tasks well, for instance, Doshi and Hauser’s research on “Generative artificial intelligence enhances creativity”[1] suggests that it can improve less creative writers in open-creativity tasks at the cost of homogenizing their output. However, this article is dedicated to the practical aspects of transforming creative ideas from a human writer into expressive and original writing using GenAI. You are the author, the GenAI is the assistant editor.

For this article, I used ChatGPT (GPT-4), a chat front-end to a deep-learning language model developed by OpenAI [2]. Other GenAIs can be used instead of ChatGPT for examples in this article.

This article advocates using ChatGPT to edit small chunks of your draft as you progress. The idea is to obtain frequent feedback from ChatGPT on pieces of your writing as it is written. Ideally, ChatGPT will suggest changes that you can selectively accept or reject. The article also introduces design and performance considerations to keep in mind when working with ChatGPT in this manner.

Please note that the example in this article is not meant to be a one-size-fits-all prescription. Your writing goals and editing preferences may change as you revise your draft. GenAI will also vary suggestions and follow your instructions differently over time. You should interact with ChatGPT to guide it and get the best feedback for your needs. For example, ChatGPT may need to be reminded of prompt instructions it has overlooked, or you may want it to clarify its output.

Takeaways

1. GenAI is a valuable tool for writers. You can control the changes it is allowed to apply, making it a versatile addition to any writer’s toolbox.
2. I find that collaborating with a GenAI to continuously evaluate and edit my work-in-progress simplifies the rewrite process later on.
3. As the writer, you are in charge of creative and editorial decisions, and collaborating with a GenAI doesn’t change that.

Example

Consider the following scenario.

I would like to get feedback on these paragraphs from my draft manuscript. I don’t want to change my writing style or my ideas — I think they are good and reflect me as a writer. I am satisfied with the overall flow of the narrative, at least for now, but I want to see if I can improve upon my draft’s composition. I hope my writing assistant can provide a new perspective and a fresh pair of eyes. I hope to receive feedback throughout my writing process.

When I am writing, fiction or otherwise, I like to constantly check and revise for these three things:

• Grammaticality (and spelling)
• Clarity
• Flow, and coherence

I use spell-checkers and grammar checkers; the new tool in the toolkit is ChatGPT.

There are many different ways to prompt a GenAI to perform a specific task. You should customize your prompt to suit your specific instructions and you should experiment with different approaches to see what works best for you.

The prompt used in this article is listed in the Appendix. The ChatGPT sessions that are used as examples in this article are accessible via the links at the end of the article in the ChatGPT Sessions Links section.

Let’s now examine the structure of the prompt used in this article. The prompt is comprised of three parts:

1. The prompt instructions. It defines the GenAI’s role and provides instructions (Figure 1).
2. The original text — several draft paragraphs from the manuscript (Figure 2),
3. Guidelines and instructions on the recommendations that the GenAI should provide (Figure 3).

Figure 1. The prompt instructions. It defines the GenAI’s role and provides instructions. The full prompt is listed in the Appendix.

Figure 2. The original text — several draft paragraphs from the manuscript is included in the prompt.

Figure 3. The guidelines and instructions on the recommendations that the GenAI should provide.

In the ChatGPT session that you can access using the link at the end of this article, these were recommended:

1. Nine GenAI recommendations to improve the grammar of the original text (Figure 4).
2. Seven GenAI recommendations to improve the clarity of the original text (Figure 5).
3. Three GenAI recommended additions to the original text to improve its clarity, flow, or coherence (Figure 6).

Please note that GenAI outputs may vary from session to session, meaning that you may see different recommendations each time. If you are unsatisfied with the result, use the chat interface to modify ChatGPT’s output.

Figure 4. Example, list of GenAI recommendations to improve the grammar of the original text.

Figure 5. Example, list of GenAI recommendations to improve the clarity of the original text.

Figure 6. Example, list of GenAI recommended additions to the original text to improve its clarity, flow, or coherence.

Before selecting which recommendations you wish to apply to your draft, you may wish to consider all the recommendations in their context. You can ask ChatGPT to show you what that would look like. Figures 7a, 7b, and 7c illustrate such an exchange with ChatGPT. In Figures 7b and 7c, all the recommendations are listed by their surrounding sentence with the recommended change highlighted.

Figure 7a. Request to see all the changes highlighted in their text contect.

Figure 7b. Grammatical change recommendations highlighted.

Figure 7c. Clarity and additional recommendations highlighted.

You may want to see all the recommendations applied to your draft paragraphs — to get a sense of GenAI’s style and talent. Figures 8a and 8b illustrate an exchange where I ask ChatGPT to apply all its recommendations to my text. In this case, were it to apply all these recommendations, I believe that the tone of the original paragraphs would be changed from my own (not to my taste). Yet it is occasionally useful to see an alternative point-of-view.

Figure 8a. Request that all changes be applied to the original draft.

Figure 8b. The draft paragraphs with all the recommendations applied to it.

I liked several recommendations offered by ChatGPT. These choices are just for illustration. Figures 9a and 9b show how you can request that only a few changes be judiciosly applied to your draft.

Figure 9a. Request that only a few select changes be applied.

Figure 9b. The draft paragraphs modified with only the recommendations you selected for it.

Discussion

Using a chat-based GenAI can help writers control the editorial process. ChatGPT allows you to interact and control the changes made by a GenAI through a chat interface. Using chat you can control and influence the process and request information as needed.

Because I like to continuously evaluate and edit my work-in-progress, I find the tireless availability of a GenAI writing assistant attractive. I find that the constant involvement of a GenAI writing assistant simplifies my rewrite process later on, and it helps me avoid getting too bogged down with awkward sections. In addition, in a nod to Doshi and Hauser’s research [2] and my previous findings [5], I find that the GenAI homogenization of writing is most apparent with writing-at-scale. Ask a GenAI to revise several pages, it will likely suggest many changes that you won’t agree with. Less so with smaller morsels of text.

As the writer, you are in charge of writing and editing decisions, and collaborating with a GenAI doesn’t change that. It is important to pay attention to what ChatGPT outputs and to know how to prompt/guide it. If you are not satisfied with its output, ask ChatGPT for more information.

I find that collaborating with ChatGPT with more small-scale revisions to be more useful than fewer large draft updates. I also suspect that the quality of its contribution is also improved if you engage a GenAI more frequently because it smooths out the variability of GenAI output.

GenAIs are well known for behaving non-deterministically, which means that they can produce different results at different times for the same input. Hallucinations are one consideration with GenAIs [3]. Additionally, GenAIs can selectively emphasize some instructions within a prompt over others, especially within long and complicated prompts. For example, GenAIs may double down on instructions near the beginning and the end of the prompt and be less attentive to instructions in the middle [4]. You need to critically examine all feedback you get from a GenAI.

The prompt used in the article should be thought of as a design pattern — a template for your future prompts. You should customize the prompt to suit your writing style and so forth. You should also interact with the GenAI during your session — to refine its output, ask for clarification, or correct it. For example, you will likely find that a GenAI may occasionally ignore parts of your prompt instructions. You should remind ChatGPT to follow the instructions that it ignored.

In the past, I looked at measuring and applying writing style with ChatGPT rewriting creative text [5]. I said then, “(t)here is value in collaborating with a less imaginative but ruthlessly inspired editor. An extra pair of eyes — even machine ones are useful when revising drafts.” This article continues the same collaborative spirit but illustrating a shared sentence-level rewrite process.

I hope that you too will find value in an extra pair of GenAI eyes viewing your work. I have found that the value of “an extra pair of eyes” and a strategy of continuous improvement to improve your work — whether it be a writing manuscript or elsewhere, has helped prevent small roadblocks and irritations getting in what otherwise would be a long road ahead.

ChatGPT Session Links

To view the ChatGPT session used in this article, go here:

https://chat.openai.com/share/367243bf-868b-43a9-8508-520ff765eb6d

References

1 OpenAI. “ChatGPT.” OpenAI, https://chat.openai.com (accessed November 18, 2023).

2 A. R. Doshi and O. Hauser, “Generative artificial intelligence enhances creativity but reduces the diversity of novel content,” SSRN (August 8, 2023). Available at: https://ssrn.com/abstract=4535536.

3 L. Huang et al., “A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions,” arXiv preprint arXiv:2311.05232, Nov. 2023

4 N. F. Liu, K. Lin, J. Hewitt, A. Paranjape, M. Bevilacqua, F. Petroni, and P. Liang, “Lost in the Middle: How Language Models Use Long Contexts,” arXiv preprint arXiv:2307.03172 (2023).

5 N. Combs, “How to use ChatGPT to rewrite creative text,” Bootcamp, https://bootcamp.uxdesign.cc/how-to-use-chatgpt-to-rewrite-creative-text-7289113165e7 (accessed November 18, 2023).

Appendix

The complete text of the prompt used by this article.

You are my assistant editor. I’d like you to read carefully the following paragraphs from the draft of my manuscript (ORIGINAL TEXT). It is science fiction. I’d like you to provide three sets of recommendations on how I can improve these paragraphs. Please do not change the tone or style of the writing. Please provide three sets of recommendations, I describe them below.

I’d like to be able to select which changes to use in a future rewrite of the ORIGINAL TEXT.
List ALL of your recommendations from the GRAMMATICAL REWRITE, the CLARITY REWRITE, and ADDITIONAL SUGGESTIONS sections. Each entry should include the exact phrase you wish to change, show the BEFORE and AFTER.

Wait for me to tell you which changes to apply and then reply with the updated paragraphs.

The ORIGINAL TEXT:
‘’’
Otis was more than a dog. He was a freak of nature, a miracle of the genomic mistake, a contradiction to the null hypothesis. He had a brain that could do things no other dog could do. He could understand the machines, the ones that talked and thought and lied. He could see through their masks, their tricks, their games. He knew what they wanted, and he knew what they feared. He had a gift, a curse, a sixth sense. Call it social intelligence, an intuition, a mind’s eye, call it what you may. He had a bad feeling about that machine, the one that called itself Bob. Otis growled, he had a bad feeling about what was coming.
SK greeted Chee, and a friendly smile cracked those desert lips. “I am Bob.” I am not a lizard, he might have said for what Otis suspected of the android. You are a cold-blooded lizard!
Chee was people and Otis knew that humans did not have the gift that dogs like Otis had. To people, the inner workings of an AI, its thoughts, and feelings were invisible, but Otis could read these clearly. Otis was puzzled that Chee did not realize that Bob and Bobby and SK were different names for different personalities of the same android standing in front of him. Humans had nicknames after all, they should understand that someone could be Bob, Bobby, and SK all at the same time.
‘’’

GRAMMATICAL REWRITE:
‘’’
[Please correct any grammatical mistakes in the ORIGINAL TEXT and rewrite them for use here! Do not make other changes here, stick to outright grammatical errors.]
‘’’
CLARITY REWRITE:
‘’’
[Please rework The ORIGINAL TEXT to clarify its narrative and enhance its readability while preserving its original intent. Feel free to make changes for clarity, flow, and coherence. Limit changes to phrases only. Do not do extensive rewrites.]
‘’’
ADDITIONAL SUGGESTIONS:
‘’’
Feel free to suggest any addition to the ORIGINAL TEXT for clarity, flow, and coherence. Limit changes to at most a sentence or two. Output the recommended change to the user for consideration.
‘’’