Tingyu Zhang
Dr. Pullman
February 17, 2025
Week5: Read Is prompt engineering a ‘fad’ hindering AI progress?
Blog about prompting in whatever way you want.
According to AI and Writing by Sidney I. Dobrin, prompt engineering means the process of writing prompts and subsequently adjusting them.  The author is a professor and chair in the English department at the University of Florida in Gainesville. He has written and edited more than 23 books about writing, ecology, and technology.

He mentions there are 5 key points in writing an effective prompt which are being clear, specific, contextual, accurate, and ethical (Dobrin 66). Being clear means the human should type what is intended in the prompts. Otherwise, it may increase the likelihood of hallucinations (Dobrin 66). Being specific refers to having sufficient details to result in a strong output (Dobrin 66). Being contextual is defined as providing as much as background information as possible such as the intended audience and the scenario that the writing is for(a piece of journalism, a brief professional email, etc.), and the intended tone and style(Dobrin 66). Being accurate is due to the fact that Gen AI programs cannot currently discern between correct and incorrect information(Dobrin 66). The inaccurate information used in a prompt also becomes part of the Gen AI’s LLMs, which may poison future outputs from that GenAI program as well (Dobrin 66). Lastly, in terms of being ethical, he argues “ In general, prompts should avoid directing GenAI toward offensive content, intentionally misleading content, and content that might lead to harm. It’s important to keep in mind as well that GenAI programs often exhibit biases and prompts should be written with an awareness of these biases” (Dobrin 66).

Regarding how to refine prompts, Dobrin claims that some strategies can be adopted to revise the initial prompts which do not lead to the outputs the person intends (Dobrin 68). The first one is to clarify details of the task including the purpose, the number of words, the type of sources the individual wishes to use, and their target audience(Dobrin 68). Secondly, people can identify any restrictions. For instance, “avoid the use of first-person language”(Dobrin 68). Thirdly, structure and order the prompt by using a fill-in-blank approach (Dobrin 68). For example, “provide a discussion of 3 key aspects of opportunity cost: 1. ____ 2.____3.____” (Dobrin 68). Fourthly, provide input/output examples to illustrate the style or format the user wishes to receive as an output (Dobrin 68). For example, we might write a prompt with the prefix, “In a voice appropriate to a business workplace, write a report…” or “Writing as if you were a biologist, write a…”(Dobrin 68).

Below is an example of using the regular ones and the Reason Features of ChatGPT, Deepseek, and Perplexity in revising the writing to become more cohesive. My goal is to test which one is the quickest in response and the most excellent in revising their output considering the audience factor.
The first prompt is   “Please revise any mistakes in this writing to be more cohesive and highlight the corrections.”
The second prompt is   “Please revise any mistakes in this writing be more cohesive and highlight the corrections. The reader is a Rhetoric and Composition professor teaching a graduate-level  Writing Research and Methodology course.”
The title of my paper is Integrating Artificial Intelligence into Freshman Composition Pedagogy: Challenges and Solutions for ENGL 1101 and 1102.

I write the introduction part as the situation and overview of the topic. Below is my original text:
“Freshman composition courses, ENGL 1101 and 1102, are required at many U.S. universities, to fulfill core curriculum requirements. These courses help students develop essential writing skills crucial for academic success and career readiness. Through tasks like literacy narratives, visual analyses, argumentative essays, and professional autobiographies, students hone transferable skills such as critical thinking, creativity, and problem-solving. Mastering tasks such as analyzing prompts, generating original ideas, and presenting logical persuasive arguments prepares students to excel in competitive job markets and tackle complex workplace challenges. 
Nowadays, emerging technologies, particularly artificial intelligence (AI), have been increasingly integrated into freshman composition pedagogy. AI, which refers to machines or algorithms capable of performing tasks or solving problems, has seen significant advancements, especially in natural language processing (Hutson et al. 2024). Tools like ChatGPT, Gemini, and CoPilot can understand and generate human language. A report by the Capgemini Research Institute reveals that 70% of American teachers believe curricula and assessments must adapt to AI’s growing influence (Heaton 2024). ”

The result of my testing is below:
Sometimes, the regular CharGPT didn’t follow all of my directions such as not highlighting the words it corrected or even not bolding these words. The regular ChatGPT even hallucinates some of what I’ve written. For instance, ChatGPT suggests I should write”such as” instead of “like” (in formal lists) → “Such as” is preferred in academic writing for listing examples. In reality, I didn’t have “like” in the original sentence of “Freshman composition courses, ENGL 1101 and 1102, are required at many U.S. universities, to fulfill core curriculum requirements. ”

Compared to the regular ChatGPT, Perplexity, and Deepseek, their Reasoning features took longer to generate the outputs but they closely followed my directions by bolding the corrections but the Gen AI could not highlight them. ChatGPT with the Reasoning feature finished the task just in 6 seconds more quickly than Perplexity and Deepseek with the same feature. However, Perplexity and Deepseek with the Reasoning feature can give a simpler and more condensed vision than ChatGPT with the same feature. For instance, in terms of advice on Flow and Cohesion, the Perplexity with Reason feature suggests I remove redundant phrases (e.g., “to fulfill core curriculum requirements” → “core requirements”). Deepseek recommends I replace repetitive phrasing (“required at many U.S. universities, to fulfill core curriculum requirements”) with “core requirements… designed to equip students” to eliminate redundancy and emphasize purpose. Perplexity and Deepseek with the Reason features give similar advice in this sentence.

Considering the prompt with requirement of the audience awareness, ChatGPT with the Reason Feature provided the answer within 16 seconds. Deepseek with the same feature came the second within 25 seconds; Perplexity with the same one did it within 30 seconds. However, ChatGPT with the Reason Feature is the only one without any detailed analysis of the correction and it doesn’t take into account the audience factor as the second version does not change too much.

Overall, Deepseek with the Reason Feature and Perplexity with Feature consider the audience well in their answers by adding more rhetoric and pedagogy terms to the revised prompts. But I cannot tell which one, Deepseek or Perplexity with the Reason Features, is better in giving answers considering the audience.  Although ChatGPT does the quickest job of these 3 Gen AI tools in giving writing revision suggestions,  the regular Chat GPT and ChatGPT with Reason Feature both lack reliability in giving revision suggestions on the original writing.

Here are some screenshots in the testing:

Regular ChatGPT：
Below is the answer to the first prompt: “Please revise any mistakes in this writing to be more cohesive and highlight the corrections.”

Regular Deepseek：
Below is the answer of the first prompt: “Please revise any mistakes in this writing be more cohesive and highlight the corrections.”

Regular Perplexity：
Below is the answer of the first prompt: “Please revise any mistakes in this writing to be more cohesive and highlight the corrections.”

ChatGPT with Reason Feature：
Below is the answer of the second prompt: “Please revise any mistakes in this writing to be more cohesive and highlight the corrections. The reader is a Rhetoric and Composition professor teaching a graduate-level  Writing Research and Methodology course.”

Deepseek with Reason Feature：
Below is the answer of the second prompt: “Please revise any mistakes in this writing to be more cohesive and highlight the corrections. The reader is a Rhetoric and Composition professor teaching a graduate-level  Writing Research and Methodology course.”

Perplexity with Reason Feature：
Below is the answer of the second prompt: “Please revise any mistakes in this writing to be more cohesive and highlight the corrections. The reader is a Rhetoric and Composition professor teaching a graduate-level  Writing Research and Methodology course.”

Works Cited

Dobrin, Sidney I. AI and writing. Broadview Press, 2023.