Since the advent of ChatGPT, people have debated its abilities, the quality of its writing, and whether it threatens jobs, especially in writing. This essay addresses these concerns by reporting the results of a quasi-experimental study using ChatGPT-4o.   

Background

In traditional experimental research, researchers use two or more distinct groups to compare the effects of one treatment against those of another. In this quasi-experimental study, I compared ChatGPT-4o (the machine’s) with human writing. To that end, I purposively selected a short news release from PR Newswire. This release was retrieved on March 26, 2025.

The chosen release was segmented into individual sentences. These sentences were then given to ChatGPT-4o, which was instructed to reassemble them into a coherent and well-structured news release using only the provided information. To enable comparisons, three different prompts were used when instructing ChatGPT-4o to reconstruct full releases.

I analyzed ChatGPT-4o’s outputs using move analysis, which John Swales originally developed (Kanoksilapatham, 2008). The three moves analyzed were the headline, lead, and body—this is how information is organized in the inverted pyramid format. The background information was analyzed as part of the body, since not all releases provide it to readers. The sub-moves within the lead and body were also analyzed.

Main Findings  

Headline Writing. The headline is the sentence that precedes the lead and tells the readers what the news release is about (Stovall, 2006). Based on analyses of ChatGPT-4o’s outputs, it produced the same headline across the three versions generated from the three prompts. It adhered to the conventions identified by Michael Swan (2016) regarding the use of the present simple tense, the omission of articles, and the use of the verb to be. However, it failed to employ the headline vocabulary identified by Swan.

Lead Writing. The lead, or intro, is the first paragraph of the news release. Across the three outputs, one was efficient: ChatGPT-4o generated a straight summary news lead in fewer words. A straight summary news lead is usually used to make announcements (Wilcox & Reber, 2014). The original news release used in the experiment was about a company announcing the opening of a new branch.

Interestingly, in two outputs, ChatGPT-4o included a parenthesized description of the company. Although newspaper editors generally prefer a different style of writing, this is acceptable in public relations writing. In PR writing, depicting the organization in a refined, favorable light (in a few words) without exaggeration is considered appropriate, primarily when the release is published in a medium shared with competitors rather than an organization-owned outlet.  In such situations, I always encourage my students to use a modified straight summary news lead instead of a straight summary news lead. To clarify, when an organization is planning an environmental initiative, the lead might read: “Driven by its strong commitment to environmental issues, [organization name] plans to organize an environmental initiative…” Journalists, of course, would not include the initial phrase in which the organization was portrayed in a favorable light, because they represent their newspapers, not the organization planning the initiative.  

In one of the leads, ChatGPT-4o used the present perfect tense reflecting trends in online media. That’s to say, with the advent of the internet, it has been observed that online mass media tend to use the present perfect tense in leads rather than the past simple tense. Previously, the print media used the past simple tense, while radio and television employed the present perfect tense (Newsom & Carrell 2001). Finally, all generated leads consisted of two sentences, whereas ideally, a lead should be a single-sentence paragraph.

Body Writing. The body is the part of a press release that follows the lead and provides details. The three news releases generated for analysis varied in length, structure, and information organization.

In terms of efficiency, one output adhered to Stovall’s (2006) guidelines on sentence length, paragraph size, and readability, making it the most precise and digestible. This output presented a clear progression from a straight summary lead to the expansion of the “what” question, followed by executive statements, and concluded with background information, consistent with inverted-pyramid conventions. Ideally, background information in news releases appears at the end (Stovall, 2006; Wilcox & Reber, 2014). The inverted pyramid structure has two main parts (moves)—the lead and the body. Background information does not constitute a separate part (move) because not all releases include background information—it is an optional sub-move.

Highlights

To generate a proper headline, users should instruct ChatGPT to write it in the fewest possible words, using headline-specific vocabulary, observing headline grammatical rules, and in the appropriate tense. The grammatical and technical rules outlined by Foster (2008) and Swan (2016) for writing headlines can be incorporated into ChatGPT’s training. To produce a proper lead, ChatGPT should be prompted, once the required information is provided, to generate the specified type of lead, in the fewest possible words, in one sentence, and in the appropriate tense. For the body, once ChatGPT has been given the necessary information, prompt it to produce the required number of paragraphs, organized from most to least important. Clear instructions should also include identifying the people involved and how to order information about them; determining the number of sentences per paragraph and the number of words per sentence; structuring the second paragraph; and finally, positioning the background information. All these instructions can be summarized in a cover-all prompt, as the model below shows

The comprehensive roadmap proposed here is expected to prove highly valuable to educators and practitioners, as it includes (a) a structured prompt model for generating news releases using ChatGPT and (b) an illustrative example for manually writing news releases.

Conclusion

This essay argues that ChatGPT-4o can be a useful tool for creating news releases, but its effectiveness relies on tailored training and precise prompting. Importantly, using this tool effectively requires a trainer who reviews outputs with a writer’s lens rather than a reader’s to ensure quality. In other words, using ChatGPT effectively for news release writing requires familiarity with the genre’s professional conventions and the tool itself. Moreover, one of the key challenges ChatGPT faces is accurately depicting the context. Human writers compose news releases based on their own observations (observed context), whereas ChatGPT relies solely on the prompt provided (reported context) and its pre-training. Therefore, the prompt should clearly describe the context so that ChatGPT can accurately depict it.

References

Foster, J. (2008). Effective writing skills for public relations (4th ed.). London and Philadelphia: Kogan Page Limited.

Kanoksilapatham, B. (2008). Introduction to move analysis. In D. C. Biber (Ed.), Discourse on the Move (pp. 23-41). Amsterdam: John Benjamins Publishing Company. https://doi.org/10.1075/scl.28.04kan

Newsom, D., & Carrell, B. (2001). Public relations writing: Form and style. Stamford: Wadsworth Publishing Company.

Stovall, J. G. (2006). Writing for mass media. Boston, New York, and San Francisco: Pearson.

Swan, M. (2016). Practical English usage. London: Oxford University Press.

Wilcox, D. L., & Reber, B. H. (2014). Public relations writing and media techniques. Edinburgh Gate: Pearson.