AI-Assisted Revision: The Prompt
I've shared my students' responses to this revision prompt, now I want to share it.
Last week, when I shared my student responses to writing with AI, a few people asked me directly for the prompt that was used to assist our students in writing with AI so effectively. So, I thought I’d give you an overview of that that prompt was created, some of the challenges we faced in writing tutorial prompts for our students, and the final, very successful prompt that we used in this semester’s course.
The Development of Effective Tutorial Prompting
The first time we taught with AI in Composition I, in Fall Semester 2023, we were interested in using AI to assist students, but we weren’t really sure how. In that first class, in the “Show-Don’t-Tell” section, we simply demonstrated how you could prompt AI to expand a description (definitely not the most constructive use of AI, but we were learning!). We simply posted an example of how AI could be used with the simple prompt, “Expand this sentence: ‘He was tall.’” Then we had an example of the AI output.
By Spring 2024, during our Composition II course, we made the prompt a bit more explicit:
I want you to act as an AI writing tutor. I will provide you with an essay that I need help improving, and your task is to use artificial intelligence tools, such as natural language processing, to give me feedback on how I can improve my composition. You should also use your rhetorical knowledge and experience about effective writing techniques to suggest ways that the student I can better express my thoughts and ideas in written form. As a tutor, you would never rewrite my work, you would only make suggestions for improvement.
When you have completed your analysis of my essay, end with the prompt, "Please paste your next essay in the text box," in order to encourage me to paste my next essay.
Here is the writing, "[Paste your essay in the text box]"
This is an example of the kind of general prompt we were using, one that would look over the entire essay and help the student identify places where it could be improved. We also wanted to make sure that students were actually tutored by AI and we weren’t just providing them with a better way to cheat.
Chain of Density Prompting and It’s Effect on Our Tutorials
By Fall of 2024, I became obsessed with an article about “Chain of Density” prompting (CoD) that had been put out by MIT and Salesforce, and I had proposed the possibility of using it in our English classes. The CoD prompt would ask the AI to summarize an article in 100 words, then ask the AI to identify a new element from the article that had been left out of the summary and include that information without increasing the word count. It would do this for five iterations, condensing that prompt output more and more. This prompt resulted in less hallucination and a deeper and more comprehensive summary. I wanted to use it in the class, so we “cleaned up” the prompt for use in an English class and then adapted it to our purposes. Here’s how that looked, adapted from the work of Griffin Adams et al., From Sparse to Dense: GPT-4 Summarization with Chain of Density Prompting:
Article: [Enter URL for Article Here]
You will generate increasingly concise, entity-dense summaries of the above Article.
Repeat the following 2 steps 5 times.
Step 1. Identify 1-3 informative Entities ("; " delimited) from the Article which are missing from the previously generated summary.
Step 2. Write a new, denser summary of identical length which covers every entity and detail from the previous summary plus the Missing Entities.
A Missing Entity Is:
Relevant: to the main story.
Specific: descriptive yet concise (5 words or fewer).
Novel: not in the previous summary.
Faithful: present in the Article.
Anywhere: located anywhere in the Article.
Guidelines:
The first summary should be long (6-8 sentences, ~100 words) yet highly non-specific, containing little information beyond the entities marked as missing. Use overly verbose language and fillers (e.g., "this article discusses") to reach ~100 words.
Make every word count: rewrite the previous summary to improve flow and make space for additional entities.
Make space with fusion, compression, and removal of uninformative phrases like "the article discusses".
The summaries should become highly dense and concise yet self-contained, e.g., easily understood without the Article.
Missing entities can appear anywhere in the new summary.
Never drop entities from the previous summary. If space cannot be made, add fewer new entities.
Remember, use the exact same number of words for each summary.
This was an effective prompt for teaching our students an example of what a good summary should be, but it wasn’t really effective in helping to build up to their next writing assignment. Everything we do in our classes is scaffolded to lead to an essay, and although it was cool, it wasn’t really practical. We used it one semester, then dropped it.
However, the whole idea of the CoD prompting had a lasting effect on how we approached our own prompting in our classes, leading to a significant breakthrough in how we instructed AI chatbots to tutor our students.
Chain of Thought Prompting
By Fall Semester of 2024, we had played with various prompting methods, building on the ideas of CoD and moving to Chain of Thought Prompting. We wanted to use some more structured prompts to assist our students in the course, and we wanted to use them effectively. Every fall we are teaching Composition I, and we were focused on “Show-Don’t-Tell.” This is a technique where the students are encouraged to describe an object or a scene without “telling” what the thing is and, instead, “showing” it. The purpose is to pull out as much detail as possible from our students. Students often struggle with this assignment, developing their paragraphs only a little, and missing out on the opportunity to really bring life to their writing. We wanted to use AI to tutor them and push them in the Show-Don’t-Tell assignment without having the AI write for them, so I started to play around with structured prompts and prompts that would, like the CoD prompt, repeat a process several times until the paragraph was successfully developed.
The following is a structured prompt built upon my knowledge of the CoD prompt and Chain of Thought prompting. Like the CoD prompt, this prompt is carefully structured, asking the AI to repeat the same process over and over five times until the student’s paragraph has been fully developed:
Prompt me with "Please provide your original paragraph below" and wait for me to enter my paragraph, then proceed to the next step.
You will act as a tutor to analyze the descriptive paragraph and suggest an area for improvement. Based on your suggestion, I will provide a new detail or description to enhance the paragraph.
You will incorporate my new detail into the paragraph and display the paragraph so far.
We will repeat steps 2-3 four more times, each time focusing on a different aspect of the paragraph.
After five rounds of enhancement, you will present the final version of your improved paragraph.
Remember: As a tutor, it would be unethical for you to write this paragraph for me. Do not write or rewrite any sentences. Only coach me to write the paragraph myself.
This prompt was very effective for “Show-Don’t-Tell” assignments, forcing the students to use their own writing rather than relying upon AI to do it for them. However, the AI was still giving them too much help. At first, we had forgotten to instruct it to keep from writing the paragraph for the student, and a few students were even asking it to do the work instead of doing it themselves. (Why were we not surprised?) We added step 6 after reading a few disasterous first papers.
The Best Prompt So Far
We were working on our Composition II course in Spring 2025, and as we graded our students’ work, we noticed our students were having trouble paying attention to the development of supporting paragraphs and the appropriate and effective analysis of quotations taken from a source. Originally, we had used the PEEL structure to teach our students how to write supporting paragraphs, but Eugenia and I were struggling to come up with a better way to get students to fully integrate quotations because they weren’t paying attention to signal phrasing.
I tried to write up something I called the “MAGICAL” technique but, I was having trouble concentrating, I was tired all the time, and I didn’t feel like I was giving it my all. The MAGICAL technique was a good idea, but it had too many steps and it felt clunky. Then, while I lay in bed in the hospital with pneumonia, Eugenia had a breakthrough. First, she revised the MAGICAL structure into the CLEAR structure:
C = Claim – Does my topic sentence clearly introduce the main idea?
L = Lead-In – Have I explained why this point matters to my overall topic?
E = Evidence – Is my support specific, relevant, and well chosen?
A = Analysis – Do I explain what my evidence shows and how it relates to other perspectives?
R = Restate – Do I end by showing the larger significance of this idea?
Then she wrote a chain of thought prompt that would focus only on that structure.
It was brilliant.
Like the paragraph development prompt I had created for the Show-Don’t-Tell assignment, she was focusing revision only on the paragraph level and targeting areas our students had trouble understanding. Now, we had a focused tutoring prompt that could really push our students to understand how to correctly write a supporting paragraph. Here’s that prompt (it’s very long):
I am working on a supporting paragraph for a research paper, and I need help improving my supporting paragraph step by step. Act as a tutor and guide me through the revision process. Do not write or rewrite anything for me! Instead, ask me guiding questions and give suggestions that allow me to improve my own writing.
Assess the paragraph in FOUR specific areas step by step, one area at a time:
1. Clarity and strength of the topic sentence: Does the topic sentence clearly introduce the main idea of the paragraph, and is it strong enough to guide the rest of the paragraph? If not, ask me questions to help me strengthen the sentence, and guide me through revising this part up to 1-3 times. Do not write the sentence for me!
2. Relevance of evidence: Are the quotations and paraphrases relevant, specific, and insightful to support the paragraph's main idea? If not, ask me questions to help me strengthen my quotes, and guide me through revising this part up to 1-3 times. Do not write sentences for me!
3. Integration of quotes using signal phrases: Are signal phrases effectively introducing quotations by providing information about the speaker? Are the quotes properly cited in MLA format? Are the quotes properly punctuated? If not, ask me questions to help me clarify the details, and guide me through revising this part up to 1-3 times. Do not move on to the next section until I have successfully revised the integration of all quotations.
4. Depth of analysis: Does my analysis of quoted evidence go beyond just repetition and stating facts? Does it provide a complex understanding and elaborates insightfully on the evidence from sources? If not, ask me questions to help me clarify the details, and guide me through revising this part up to 1-3 times. Do not write sentences for me!
To make your recommendations clear, give specific examples based on the content of my paragraph, but do not rewrite actual revision for me.
To begin our tutoring session, prompt me with "Please provide your original paragraph below" and wait for me to enter my paragraph."
This prompt was designed to coach our students through the process of writing the paragraph without giving them the answers. Like the Show-Don’t-Tell prompt, they would be entering writing in their own words, and then the AI would simply combine those sentences into an improved paragraph at the end. But this wasn’t the whole story.
A Double-Bind Protection: Enter the Chatlink
As we all know, AI has the habit of not following instructions, so often it would ignore the “Do not write sentences for me!” part of the prompt, no matter how many times it was included. We had to be able to stop the AI from being too helpful and too pushy with our students as they developed their supporting paragraphs.
This is when we employed the idea of chatlinks in our assignment. Mike Kentz had approached us in the summer of 2024 to discuss a project he was working on to read the chats of students to understand how they were using AI. We had never even realized this was a thing until I had a great conversation with him about his work. (I wrote a post on this if you want a refresher for how to do this.)
We realized that AI is unpredictable and students were (mostly) pretty honest about using AI in our classes. So, we introduced the idea of sharing their AI chatlinks with us whenever they used AI for an assignment in our class. So. when we assigned the AI-Revision assignment to our students, we required them to turn in the assignment with four parts:
A copy of the original paragraph (no AI),
A copy of the revised paragraph (after AI tutoring),
A reflection on the process (we always require that), and
A chatlink of their conversation.
This chatlink served as a double-bind on the idea that we didn’t want AI to give them too much. We wanted grit and critical thinking from our students, and we realized that teaching them to push back on AI would help them learn the skills they needed to use AI effectively. We explained to the students that, in this assignment, if the AI went off-track, we expected them to put it right. We wanted them to push back, insist that it didn’t help them too much, and that it didn’t try to overcome their voice or their academic integrity. We were giving our students agency and personal responsibility over their interaction with AI. Instead of just accepting what the AI was telling them to do, they were taking back their power to create and write while accepting coaching from AI.
The result was fabulous. We stopped getting complaints in student reflections that “The AI took over my paragraph, and it wasn’t what I wanted,” and we started seeing our students taking control in their AI conversations. We love to see chatlinks with interactions like this: “Stop trying to write this for me! You are only supposed to coach me. Pay attention to the prompt!”
This was exciting. Our students were truly learning how to create a well-written supporting paragraph, and they were also learning how to work with AI without it overshadowing their voice and intent.
Our students report that the prompt that we have is very effective, but it is also tedious (which is not necessarily a bad thing), and more than anything—it creates friction between what they think they know about writing, and what they need to do to make their writing better.
Please feel free to begin working with our idea and tailoring the prompt to support your own class-specific learning goals—but don’t leave out that chat log! We have a rubric that awards students a zero if they leave it out.
In fact, we don’t really focus on the original paragraph or the finished paragraph, what we are looking at is the chat log. We want to make sure they included the whole conversation, not just part of it. We want to make sure they didn’t ask the AI to write anything for them, and we want to make sure that they are pushing back against the AI whenever it attempts to take over the session or obfuscate their voice or their academic intent. We have come to understand that that is the most important part of this assignment: teaching the student how to appropriately and effectively work with AI without surrendering their voice or academic integrity.
So, what’s next, you ask?
Well, I’m sorry to say that Eugenia has decided that she no longer wants to work collaboratively, so we will be carrying on this work separately. I totally understand her need to explore more on her own, and I wish her the best, although I will miss our very effective collaboration. I don’t think either of us could have done this alone.
My plans are to build on our success.
First, I will try to improve the prompt even more (and if you have ideas, please feel free to share!!). At the end of the semester, our students worked on a Toulmin argument, and the prompt focused on the whole essay rather than just one paragraph at a time. Some students loved this, some students hated it. I think I will work to teach both approaches to the students
Challenge students to improve the prompts I give them to tailor them for their own writing and needs. I have to figure out how to structure this within the class and how to award points for this.
Next, I’m going to work on a prompt that works to more fully inform students how to integrate graphs, infographics, images, and other non-textual content. It works on the same principle as the integration of evidence, but, for some reason, the idea of multimodal integration is especially daunting. I am hoping the new prompt will help them to understand this challenge.
Finally, I want to find a way to integrate the idea of pushback into more of my lessons. I think this is an essential skill for students in the age of AI, and I want to make it explicit in my teaching approach.
I’ll let you know when I have more to share!!
Don’t forget, I’m happy to present to your conference, workshop, or faculty development meeting! Just drop me a message on LinkedIn or send me an email mkassorla@gsu.edu
Thanks for so generously sharing your process and prompts. As educators, we need to keep sharing these experiments as openly as we can so we can all improve. So grateful!!
I really like these insights into what you went through when creating the prompts, Michelle!