Prompt Instructions to Improve the Style
|
|
|
When writing a prompt for an AI tool to generate a technical document, one common strategy is called CRAFT [1-4]: context, role, action, format, and tone. First, context is the background about the document that the AI tool needs: "Our team has written an initial report (which is attached) for the design of an exhibit at a children's science museum." Second, role is the author persona that the AI tool is to adopt ("Assume that you are a member of the team"). Third, action is the task that the AI tool is to perform: "Write a summary of this report to the same audience that the report targets." Fourth, format is simply the requested format of the writing: "The summary should be a single paragraph that is no more than 1000 characters (without spaces)."
For this CRAFT strategy, tone is better thought of as the writing style. In the technical summary example, the audience for the summary is the same as for the report. To improve the style of what ChatGPT or Claude would produce, this web page presents three instructions for the style portion of the prompt. These instructions address three common weaknesses of AI-generated summaries of trusted reports [5]:
|
Instruction 1: Keep sentence length to less than 20 words.
Using this prompt and the next one listed below, we were able to increase the Flesch reading ease for a set of ChatGPT summaries from 10.4 to 43.1. Likewise, the instruction reduced the average Flesch reading level for those same summaries from 17.3 (prompt with no style instructions) to 11.8.
Instruction 2: Avoid long and pretentious nouns such as “identification” and “portability.” Instead, prefer shorter words such as “identify” and “portable.”
Using this prompt and the one listed above, we were able to increase the Flesch reading ease for a set of ChatGPT summaries from 10.4 to 43.1. Likewise, the instruction reduced the average Flesch reading level for those same summaries from 17.3 (prompt with no style instructions) to 11.8.
Instruction 3: Vary the types of sentence openers (such as adverb, infinitive phrase, and dependent clause) to make connections between sentences.
Using this prompt, we were able to increase the Flesch reading ease for a set of ChatGPT summaries from 10.4 to 43.1. Likewise, the instruction reduced the average Flesch reading level for those same summaries from 17.3 (prompt with no style instructions) to 11.8.
Using this prompt and the next one listed below, we were able to increase the Flesch reading ease for a set of ChatGPT summaries from 10.4 to 43.1. Likewise, the instruction reduced the average Flesch reading level for those same summaries from 17.3 (prompt with no style instructions) to 11.8.
Instruction 2: Avoid long and pretentious nouns such as “identification” and “portability.” Instead, prefer shorter words such as “identify” and “portable.”
Using this prompt and the one listed above, we were able to increase the Flesch reading ease for a set of ChatGPT summaries from 10.4 to 43.1. Likewise, the instruction reduced the average Flesch reading level for those same summaries from 17.3 (prompt with no style instructions) to 11.8.
Instruction 3: Vary the types of sentence openers (such as adverb, infinitive phrase, and dependent clause) to make connections between sentences.
Using this prompt, we were able to increase the Flesch reading ease for a set of ChatGPT summaries from 10.4 to 43.1. Likewise, the instruction reduced the average Flesch reading level for those same summaries from 17.3 (prompt with no style instructions) to 11.8.
References
- Michael Alley, Hala Hobrom, and Suryansh Sijwali, “Three Stylistic Weaknesses of AI Writing in Engineering Documents and Prompts to Address Those Weaknesses,” extended abstract, 2026 ASEE St. Lawrence Regional Conference (Ithaca, NY: ASEE St. Lawrence Regional Conference, 25 April 2026).
* AI was not used to generate text for this webpage.
