Journalism and Mass Communication Educator

Designing AI Tools to Coach Journalistic Writing

Journal: Journalism and Mass Communication Educator

Manuscript ID JMCE-24-0010

Manuscript Type: Essays

Keywords: Journalism, training, teaching, writing, news

Fo

Country Under Study: United States

Generative AI is among the fastest-adopted tools in the history of

rP
technology. Industry analysts predict adoption will climb to 77.8 million
users in the two years following the November 2022 release of ChatGPT,
more than doubling the adoption rate of both tablets and smartphones
(Lebow, 2023). Therefore, AI’s potential impact on journalism practice
ee

warrants significant research.

This scholarly essay chronicles the development of (name withheld for
review), a new AI chatbot that coaches beginning journalists to ideate,
Abstract:
develop questions, and improve their writing. The tool is named for
rR

(name withheld for review), the acclaimed journalist who advocated for
truth during a challenging period in American history. Rather than give
“answers” or promote cheating, the tool encourages students to develop
their critical thinking skills, credit sources, and abide by journalistic
ev

ethics. (name withheld for review) was created by the author in

partnership with (name withheld for review), a developer affiliated with
MIT’s MediaLab.
iew

http://mc.manuscriptcentral.com/jmce
Page 1 of 14 Journalism and Mass Communication Educator

1
2
3
4
5 Designing AI Tools to Coach Journalistic Writing
6
7
8 Qualitative declines in journalistic writing persist as digital publishers increasingly value
9
10
11 expediency and “clickbait” over thoroughly researched reporting and fact-checking (Lacy &
12
13
14 Rosenstiel, 2015; Wellbrook, 2016). Media organizations have always competed and coveted the
15
16 bragging rights of being the first to break news. However, the proliferation of disruptive digital
17
18
technologies, continues to adversely affect the predictable business models that once made
19
Fo
20
21 journalism a profitable and sustainable industry. Media company cutbacks, consolidations, and
22
23 closures have decimated newsrooms, costing many reporters, researchers, and copyeditors their
rP

24
25 jobs (Reinhardt & Zion, 2020; Kaplan, 2006). Indeed, the Pew Research Center reported that
26
ee

27
28 U.S. newsroom employment, which included newspapers, radio, broadcast television, cable, and
29
rR

30 other information services, declined 25% from 2008 to 2018. Among newsrooms, newspapers
31
32 had experienced the most devastating cuts – a 45% decrease during the decade, from about
ev

33
34
35
71,000 to 39,000 (Grieco, 2019).
36
iew

37 Since Open AI’s ChatGPT debuted in November 2022, dozens of human writers have
38
39
40
41 opined about generative AI’s potential to destroy journalism. The pervasive narrative is that it
42
43
44 threatens editorial writing jobs, promotes plagiarism, and frequently provides inaccurate
45
46
47
48
information (Foreani, 2023; Herrman, 2023). Previous research has addressed whether AI can be
49
50
51 trusted (Marconi, 2020), how it can be useful in conducting research (Parratt-Ferni, Mayoral-Si,
52
53
54 et al., 2021), and the development of an ontology to better understand and ease tensions between
55
56
57
58
Page 1 of 14
59
60 http://mc.manuscriptcentral.com/jmce
 Journalism and Mass Communication Educator Page 2 of 14

1
2
3
4 AI and human journalists (Zang & Perez, 2021). Slower to emerge is research that explores AI’s
5
6
7 potential to assist in writing instruction and enhance human storytelling—and specifically
8
9
10
11 journalism practice.
12
13
14 Generative AI is among the fastest-adopted tools in the history of technology. Industry
15
16
17 analysts predict adoption will climb to 77.8 million users in the two years following the
18
19
Fo
20
21 November 2022 release of ChatGPT, more than doubling the adoption rate of both tablets and
22
23
rP

24 smartphones (Lebow, 2023). Therefore, AI’s potential impact on journalism practice warrants
25
26
ee

27 significant research.
28
29
rR

30
31 This essay chronicles the development of (name withheld for review), a new AI chatbot
32
ev

33
34 that coaches beginning journalists to ideate, develop questions, and improve their writing. We
35
36
iew

37
named the tool for (name withheld for review), the acclaimed journalist who advocated for truth
38
39
40
41 during a challenging period in American history. Rather than give “answers” or promote
42
43
44 cheating, the tool encourages students to develop their critical thinking skills, credit sources, and
45
46
47
48 abide by journalistic ethics. (name withheld for review) was created by the author in partnership
49
50
51 with (name withheld for review), a developer affiliated with MIT’s MediaLab.
52
53
54
Writing Instruction Challenges
55
56
57
58
Page 2 of 14
59
60 http://mc.manuscriptcentral.com/jmce
Page 3 of 14 Journalism and Mass Communication Educator

1
2
3
4 Training journalists to write often begins before they become professionals. According to
5
6
7 Ellen Austin, a former Dow Jones Journalism Educator of the Year, “50% of journalists started at
8
9
10
11 their high school news publications, and 75% of minority journalists started there” (Author
12
13
14 interview, 2012). In post-secondary education, journalistic writing is a critical component of 119
15
16
17 accredited college-level journalism school programs (ACEJMC, 2023). Given that the proverbial
18
19
Fo
20
21 AI genie is out of the bottle, media researchers are obligated to not only examine its threats but
22
23
rP

24 also how this new technology can enhance journalism practice. Specifically, how might it benefit
25
26
ee

27 beginning and established journalists and improve their writing?

28
29
rR

30
31 Efforts to teach effective writing skills begin even earlier, in grade school. When
32
ev

33
34 surveyed, K-12 teachers agree that writing instruction decreases after Grade 3, even though
35
36
iew

37
students ’ability to communicate ideas effectively is essential in later grades (Applebee &
38
39
40
41 Langer, 2011; Graham, 2013).
42
43
44 Writing instruction also suffers because teachers aren’t adequately trained to prepare their
45
46
47
48 students. An analysis of 2,400 syllabi from teacher preparation programs revealed that training in
49
50
51 instructional approaches to motivate students to write and teach the mechanics of the craft is
52
53
54
lacking (Goldstein, 2017). In another study, writing educators described their own grade school
55
56
57
58
Page 3 of 14
59
60 http://mc.manuscriptcentral.com/jmce
 Journalism and Mass Communication Educator Page 4 of 14

1
2
3
4 writing instructional experiences as lacking (Street, 2003). Under-resourced and poorly prepared
5
6
7 teachers produce ill-prepared students who emerge as uninspired writers. Additionally, student
8
9
10
writing suffers because grading assignments and offering substantive feedback is labor-intensive,
11
12
13
14 especially given ballooning class sizes (Worley & Dyed, 2003; Elbow, 1997).
15
16
17 Writing can improve with sufficient practice and coaching. However, students from
18
19
Fo
20
21
underserved populations often lack access to quality instruction and one-on-one mentoring.
22
23
rP

24 Typically, tutoring is reserved for students from affluent families with the means to absorb the
25
26
ee

27 costs. As an additional benefit, free and nominally-priced generative AI tools, such as (Name
28
29
rR

30
31
withheld for review) and Kahn Academy’s Kahnamigo, can potentially democratize and scale the
32
ev

33
34 tutoring experience.
35
36
iew

37 Our Design Process

38
39 Our introduction to designing a generative AI-powered writing coach happened by
40
41
42 chance. I attended a conference in San Francisco where an MIT researcher demoed new tools
43
44 created by educators using his research team’s authoring platform. In casual conversation, he
45
46 informed me that a group of charter school English teachers had used his tools to develop a
47
48
49
chatbot that coached students to improve their five-paragraph essays. Within seconds, I asked if
50
51 it could be programmed to coach journalistic writing. Within a week of working with the
52
53 platform, with support from several seasoned journalism instructors, we created (name withheld
54
55
56
57
58
Page 4 of 14
59
60 http://mc.manuscriptcentral.com/jmce
Page 5 of 14 Journalism and Mass Communication Educator

1
2
3 for review), a generative AI writing coach designed to help beginning journalists improve their
4
5
6 writing.
7
8 Rather than indulge human nature’s propensity for laziness, from inception, we wanted to
9
10
11
12 create a tool that challenged students to think critically. This would mean having (name withheld
13
14
15 for review) engage with users in ways that were neither overly nor insufficiently challenging.
16
17
18
19
Bandura (1982; 2008), with his theory of self-efficacy, was among the first cognitive
Fo
20
21
22 psychologists to suggest that humans are the agents of their self-development, who can adapt and
23
rP

24
25 self-regulate to achieve desired outcomes (Zimmerman & Schunk, 2003). Bandura theorized that
26
ee

27
28
29 it is harder to feel assured of our ability to succeed when we feel weary or ineffective. This is
rR

30
31
32 especially true if we perceive negative emotional and physiological states as indicative of our
ev

33
34
35 incompetence, vulnerability, or inability to achieve desired goals. Earlier, Bandura (1969) put
36
iew

37
38
39 forth his social learning theory, which asserted that observation and modeling play a primary role
40
41
42 in how and why individuals learn. His theories may now require updating to account for
43
44
45 generative AI’s ability to mimic and seemingly replicate human “modeling” instantly. Our own
46
47
48
49 project development learning curve was simplified by not needing to know how to write
50
51
52 computer code. Our tech partner created an authoring mechanism that allowed us to use plain
53
54
55
56
57
58
Page 5 of 14
59
60 http://mc.manuscriptcentral.com/jmce
 Journalism and Mass Communication Educator Page 6 of 14

1
2
3
4 English to define perimeters for (name withheld for review)’s features and functions. There was
5
6
7 much to consider before giving our chatbot its first instructions.
8
9
10
11
12 Ethical Considerations
13
14
15
16 The speed at which we have adopted AI concerns many scholars and ethicists who fear
17
18
19 we have yet to fully consider its more significant implications (Powers & Ganascia, 2020). As
Fo
20
21
22 we were developing (name withheld for review), one of our journalism educator consultants
23
rP

24
25
26 raised concerns about the long-term academic impact of developing and releasing an AI writing
ee

27
28
29 coach. She wondered if offering machine assistance to students in the inquiry and writing stages
rR

30
31
32 would strengthen or weaken young students from becoming independent questioners and writers.
ev

33
34
35
36 Another concern was whether interacting with an AI chatbot might hinder students ’
iew

37
38
39 social-emotional experience of collaborating with peers. Team projects can involve intellectual
40
41
42
43 struggles and compromises. Would students essentially be outsourcing ideation to a machine and
44
45
46 having students cull from a list? Additional concerns centered on plagiarism. Would students be
47
48
49 tempted to plagiarize from the bot’s “example articles?” How would a teacher with a hundred
50
51
52
53
54
55
56
57
58
Page 6 of 14
59
60 http://mc.manuscriptcentral.com/jmce
Page 7 of 14 Journalism and Mass Communication Educator

1
2
3
4 students even know? These questions warranted careful consideration before releasing (name
5
6
7 withheld for review) to train journalistic writers.
8
9
10
User Experience
11
12
13
14 So, what is the user experience of engaging with (name withheld for review)? Like many
15
16
17 other generative AI tools, it involves a real-time one-on-one text-based conversation with the
18
19
Fo
20
21
user, accessible through a web browser. We directed the tool to begin sessions by introducing
22
23
rP

24 itself and informing users about the origin of its name and the chatbot’s creators. Next, we settled
25
26
ee

27 on having it offer users three areas of support that often inhibit beginning writers who are
28
29
rR

30
31 unfamiliar with journalistic conventions and practices: 1) story ideation, 2) developing interview
32
ev

33
34 questions, and finally, 3) writing analysis and critique.
35
36
iew

37
Like similar AI chatbots, users immediately notice that (name withheld for review)’s
38
39
40
41 mimics human conversational social norms, including politeness. Having AI adhere to cultural
42
43
44 standards and social norms has long been considered an essential design element in AI to ease
45
46
47
48 human anxiety about conversing with a machine (Rubino, 2023).
49
50
51 Facing a blank page can paralyze new writers, so users are encouraged to state their topic.
52
53
54 Typically, beginning journalists need support to narrow their topic into a tangible and actionable
55
56
57
58
Page 7 of 14
59
60 http://mc.manuscriptcentral.com/jmce
 Journalism and Mass Communication Educator Page 8 of 14

1
2
3
4 story angle. As an example, homelessness is a common and valid topic for exploration. However,
5
6
7 in raw form, it remains vague and potentially overwhelming until the writer chooses a narrower
8
9
10
facet of homelessness to pursue. (Name withheld for review) politely advises the user to refine
11
12
13
14 their topic to interest an intended audience. They are advised to localize the topic, include
15
16
17 underrepresented voices, seek accessible interview subjects, and consider novel story angles.
18
19
Fo
20
21
After narrowing their topic, (name withheld for review) supports users in drafting and
22
23
rP

24 organizing interview questions. Rather than provide questions, it challenges users to generate
25
26
ee

27 their own. Finally, after writing their own article, users can upload their stories for analysis.
28
29
rR

30
31
Among the many features, it encourages users to make vague words more specific, eliminate
32
ev

33
34 unnecessary repetition, break long paragraphs into shorter ones, repair or revise incomplete
35
36
iew

37 sentences, use correct grammar, and cite sources for claims or direct quotes. (Name withheld for
38
39
40
41 review) will also assess how well the user writing adheres to AP style guidelines.
42
43
44 Adding Safeguards
45
46 To mitigate our initial ethics-related concerns, we intentionally established limits and
47
48
49
50 guardrails. We held fast to our commitment that (name withheld for review) would be a tool that
51
52
53 prompted users to think and write for themselves. First, rather than attempt to address all
54
55
56
57
58
Page 8 of 14
59
60 http://mc.manuscriptcentral.com/jmce
Page 9 of 14 Journalism and Mass Communication Educator

1
2
3
4 journalistic genres with a single tool, we decided to limit (name withheld for review)’s purview
5
6
7 to news writing. Feature stories and opinion-oriented articles (including editorials, columns, and
8
9
10
commentaries) have distinctly different writing conventions we chose to address by developing
11
12
13
14 additional AI-powered writing tools.
15
16
17 Mindful that many of our users would be minors, we optimized (name withheld for
18
19
Fo
20
21
review) to not engage in discussions that pertain to illegal activities or hate speech. If users share
22
23
rP

24 about suicide ideation, self-harm, bullying, or sexual abuse, it advises them to consult a trusted
25
26
ee

27 professional. It also counsels users to observe precautions when considering stories that involve
28
29
rR

30
31
minors and other vulnerable populations. Public disclosures can potentially endanger individuals
32
ev

33
34 who are incarcerated, elderly, undocumented immigrants, or have mental health challenges.
35
36
iew

37 (Name withheld for review) urges users to exercise compassion when interviewing individuals
38
39
40
41 who are grieving or experiencing a tragedy. When analyzing a user’s writing, (name withheld for
42
43
44 review) also checks for indications of personal bias. We worked to anticipate ways users, and
45
46
47
especially mischievous teens, might attempt to “game the system” by having several educators
48
49
50
51 and students attempt to circumvent our built-in safeguards.
52
53
54
55
56
57
58
Page 9 of 14
59
60 http://mc.manuscriptcentral.com/jmce
 Journalism and Mass Communication Educator Page 10 of 14

1
2
3
4 User engagement was another consideration. Mindful of Bandura’s (1982) self-efficacy
5
6
7 theories, we realized that if interacting with (name withheld for review) was overly challenging
8
9
10
11 or too restrictive users might opt to use other AI tools that would not encourage them to think for
12
13
14 themselves. Striking the right challenge level required that we have a small group of trusted
15
16
17 educators and students test the tool and provide feedback. Initially, we designed (name withheld
18
19
Fo
20
21 for review) to ensure users generate their own questions. However, after further deliberation, we
22
23
rP

24 decided to limit the suggested questions to two.

25
26
ee

27 Before inviting users to upload a draft for analysis, (name withheld for review) reminds
28
29
rR

30
31 them that copying and pasting another writer’s work can often be detected and that doing so can
32
ev

33
34 lead to severe consequences. While (name withheld for review) cannot discern who authored a
35
36
iew

37
38
particular manuscript, it also will not revise the submission––insisting that users do so
39
40
41 themselves. These measures are in place to discourage cheating.
42
43
44 Public Release, Reception, and Future Study
45
46
47
48 Cognizant that advances in AI technology occur rapidly, within two weeks of conception,
49
50
51 we developed and released our tool to the public. We formally unveiled (name withheld for
52
53
54 review) at a specially added session in the fall (conference name withheld for review). Reception
55
56
57
58
Page 10 of 14
59
60 http://mc.manuscriptcentral.com/jmce
Page 11 of 14 Journalism and Mass Communication Educator

1
2
3
4 within the journalism education community was immediate and overwhelmingly positive, with
5
6
7 the platform recording thousands of global users within the first 72 hours of being online. (Name
8
9
10
withheld for review) also drew media coverage.
11
12
13
14 Our next step is to investigate the tools ’potential to enhance writing instruction through a
15
16
17 semester-long study that samples two comparison groups (one treatment group and a second
18
19
Fo
20
21 control group) of high school English language arts students with no prior journalism experience
22
23
rP

24 to examine the tool’s ability to improve journalistic news writing skills.

25
26
ee

27
28
Murrow’s Analysis of this Essay
29
rR

30
31 It seemed appropriate to ask (name withheld for review) to analyze this manuscript and
32
ev

33
34 give feedback. Here is a summary of the response:
35
36
iew

37 This is a well-written and informative piece about the development and implications of
38
39
AI in journalistic writing. Here are my suggestions:
40
41
42 1. Specificity: There are places where you could add more specific details. For
43
44 instance, when discussing the decline in journalistic writing and the rise of clickbait,
45
46 you could consider including a specific example or case study.
47
48
49 2. Quotes: You've included a few quotes from sources, but you could incorporate a
50
51 couple more, especially in the section discussing the ethical considerations of AI.
52
53 Doing so would lend more authority and depth to your arguments.
54
55
56
57
58
Page 11 of 14
59
60 http://mc.manuscriptcentral.com/jmce
 Journalism and Mass Communication Educator Page 12 of 14

1
2
3 3. Repetition: There are a few instances of repetition that could be revised for
4
5
6 brevity. For example, you discuss the potential negative impact of AI on journalism
7
8 twice, once in the introduction and again in the section on ethical considerations.
9
10 4. Grammar: There are a few minor grammatical errors that could be corrected. For
11
12
13
instance, "Bandura theorized that it is harder to feel assured of our ability to succeed
14
15 when we feel weary or ineffective." could be rephrased as "Bandura theorized that
16
17 feeling weary or ineffective can make it harder to feel assured of our ability to
18
19
succeed."
Fo
20
21
22 5. Citation: Ensure that all your claims, statistics, and assertions are properly cited,
23
rP

24 especially when they are not included in direct quotes from sources.
25
26 6. Bias: The article is mostly objective, but there are a few instances where your
ee

27
28
29 personal opinion shines through. For example, "the speed at which we have adopted
rR

30
31 AI concerns many scholars and ethicists who fear we have yet to fully consider its
32
ev

33 more significant implications". Be sure to provide a balanced perspective and

34
35
36
represent different viewpoints, not just those that align with your own. Remember,
iew

37
38 these are just suggestions. It's your story, and you should write it in a way that you feel
39
40 best conveys your message. Keep up the good work!
41
42
43
44
45 References:
46
47 Author, 2012 - Withheld for review.
48
49 Accrediting Council on Education in Journalism (2013) Accessed November5, 2023 from h
50
51
http://www.acejmc.org/accreditation-reviews/accredited-programs/
52
53 Applebee, A. N., & Langer, J. A. (2011). A snapshot of writing instruction in middle schools and
54
high schools. English Journal, 100(6), 14–27. Cleary, L. M. (1991). Affect and cognition
55
56 in the writing processes of eleventh graders. Written Communication, 8, 473–507
57
58
Page 12 of 14
59
60 http://mc.manuscriptcentral.com/jmce
Page 13 of 14 Journalism and Mass Communication Educator

1
2
3
4
5 Bandura, A. (1969). Social-learning theory of identificatory processes. Handbook of
6 socialization theory and research, 213, 262.
7
8
Bandura, A. (1982). Self-efficacy mechanism in human agency. American psychologist, 37(2),
9
10 122.
11
12 Bandura, A. (2008). An agentic perspective on positive psychology. In S. J. Lopez (Ed.), Praeger
13 perspectives. Positive psychology: Exploring the best in people (Vol. 1., pp. 167–196).
14 Westport, CT: Praeger Publishers/Greenwood Publishing Group.
15 Davis, F. D. (1989). Technology acceptance model: TAM. Al-Suqri, MN, Al-Aufi, AS:
16
17
Information Seeking Behavior and Technology Adoption, 205-219.
18
19 Elbow, P. (1997). Grading student writing: Making it simpler, fairer, clearer. New directions for
Fo
20 teaching and learning, 1997(69), 127-140.
21
22 Floreani, S. (2023, August 5). Is artificial intelligence a threat to journalism or will the
23
rP

technology destroy itself?. In The Guardian. Retrieved November 5, 2023 from:

24
25
https://www.theguardian.com/commentisfree/2023/aug/05/is-mutant-news-headed-our-
26 way-or-will-ai-chatbots-eat-their-own-tails
ee

27
28 Goldstein, D. (2017). Why kids can’t write. The New York Times, 2.
29
rR

30
31 Graham, S., Gillespie, A., & McKeown, D. (2013). Writing: Importance, development, and
32
instruction. Reading and Writing: An Interdisciplinary Journal, 26(1), 1-15.
ev

33
34 Grieco, E. (2018). Newsroom employment dropped nearly a quarter in less than 10 years, with
35 greatest decline at newspapers. Pew Research Center, 30.
36
iew

37
38 Herrman, J. (2023, June 1). Will Google’s AI plans destroy the media? Intelligencer. New York
39 Magazine. Retrieved on November 5, 2023 from
40 https://nymag.com/intelligencer/2023/06/will-ai-powered-google-eat-the-publishing-
41 industry.html
42
43 Kaplan, J. (2006). Damaging Ripple Effects of Newsroom Cutbacks. Nieman Reports, 60(1), 61.
44
45
46 Lacy, S., & Rosenstiel, T. (2015). Defining and measuring quality journalism. New Brunswick,
47 NJ: Rutgers School of Communication and Information.
48
49 Lebow, S. (August 11, 2023) Generative AI adoption climbed faster than smartphones, tablets.
50 Insider Intelligence. Accessed October 31, 2023 from
51
https://www.insiderintelligence.com/content/generative-ai-adoption-climbed-faster-than-
52
53 smartphones-tablets#
54
55 Marconi, F. (2020). Newsmakers: Artificial intelligence and the future of journalism. Columbia
56 University Press.
57
58
Page 13 of 14
59
60 http://mc.manuscriptcentral.com/jmce
 Journalism and Mass Communication Educator Page 14 of 14

1
2
3
4
5
Parratt-Ferní, S., Mayoral-Sí, J., & Mera-Ferní, M. (2021). The application of artificial
6 intelligence to journalism: an analysis of academic production. Profesional de la
7 información, 30(3).
8
9 Powers, T. M., & Ganascia, J. G. (2020). The Ethics of the Ethics of AI. The Oxford handbook of
10 ethics of AI, 25-51.
11
12
13
Reinardy, S., & Zion, L. (2020). Cutting deeper: US newspapers wipeout jobs and alter career
14 identities. Journalism Practice, 14(10), 1211-1224.
15
16 Ribino, P. (2023). The role of politeness in human–machine interactions: a systematic literature
17 review and future perspectives. Artificial Intelligence Review, 1-38.
18
19
Street, C. (2003). Pre-service teachers' attitudes about writing and learning to teach writing:
Fo
20
21 Implications for teacher educators. Teacher Education Quarterly, 30(3), 33-50.
22
23
rP

Wellbrock, C. M. (2016). The internet’s impact on journalistic quality. Media Convergence

24
25 Handbook-Vol. 1: Journalism, Broadcasting, and Social Media Aspects of Convergence,
26 83-97.
ee

27
28 Worley, R. B., & Dyrud, M. A. (2003). In this issue: grading and assessment of student
29 writing.(Focus on teaching). Business Communication Quarterly, 66(1), 72-97.
rR

30
31
32
Zhang, W., & Pérez Tornero, J. M. (2021). Introduction to AI journalism: Framework and
ontology of the trans-domain field for integrating AI into journalism. Journal of applied
ev

33
34 journalism & media studies.
35
36
iew

37 Zimmerman, B. J., & Schunk, D. H. (2003). Albert Bandura: The scholar and his contributions to
38
educational psychology. In B. J. Zimmerman & D. H. Schunk (Eds.), Educational
39
40 psychology: A century of contributions (pp. 431–457). Mahwah, NJ: Lawrence Erlbaum
41 Associates Publishers.
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
Page 14 of 14
59
60 http://mc.manuscriptcentral.com/jmce