Journalists create journalistic stories, with the purpose of providing accurate and newsworthy content for the audience (e.g., Attfield & Dowell, 2003; Gutierrez Lopez et al., 2022). However, generative artificial intelligence (AI) is changing journalists’ information environment, potentially affecting their interactions with information (e.g., Sundin, 2025). Moreover, generative AI is already used across news production, from gathering information to distributing news (Cools & Diakopoulos, 2026; Wu, 2026), and journalists are in a process of making sense of what generative AI means to their work and profession (Firdaus et al., 2025). Considering that journalists as news creators are key to the news production process, it is important to examine how journalists use generative AI and what are their information interactions like in this context.

The purpose of this study is to examine uses of generative AI in journalistic story creation process. In this study, a journalistic story creation process refers to a work task that is examined through its performance (Byström & Hansen, 2005) from ideation and planning to the completed journalistic story (Gutierrez Lopez et al., 2022). Therefore, the approach of this study is taskbased. Journalistic stories, the products of those work tasks, may be of various types (e.g., news reports or opinionated texts) and represent various topical areas (e.g., lifestyle or business), and they often follow the style of the publishing media outlet (Jaakkola, 2018; Mast, 2020). Journalism is an information-intensive profession, where not only the accuracy of information but also its creative use are central to the work (Gutierrez Lopez et al., 2022; Tuazon et al., 2020). Therefore, the domain offers a rich context for studying uses of generative AI.

Generative AI refers to models (such as large language models) that generate and synthesise information according to human instructions (Ai et al., 2025). Generative AI tools are tools based on such models. Moreover, generative AI tools may refer to tools that are available to the public such as ChatGPT, Gemini, or Claude, or media organisations’ in-house tools (see Thurman et al., 2025; Wu, 2026). In this study, uses of generative AI encompass uses of any tools that journalists perceive as generative AI based. Earlier research has shown that journalists associate both concerns and possibilities with the use of generative AI (Minotakis, 2025; Møller et al., 2025). However, more research is needed on its real-life uses, which is the focus of this study.

Furthermore, this study examines how journalists’ real-life uses of generative AI are related to their information interactions, i.e., interactions between human and information (Fidel, 2012, p. 17). Although recent studies have examined the uses of generative AI in journalistic work (D’haeseleer et al., 2025; Guenther et al., 2025; Liu, 2025; Sarısakaloğlu, 2025; Wu, 2026), their focus was not on information interactions per se, but rather on the values and perceptions journalists attach to the uses of generative AI tools. Notably, there is a lack of research that focuses on journalists’ information interactions while using generative AI in journalistic work. Studying information interactions is important to understand how they could be supported, especially in a situation where technology is changing the information environment and potentially affecting people’s accustomed ways of working.

Utilising a qualitative reconstruction interview method, fourteen journalists were interviewed about their journalistic story creation processes, uses of generative AI, and information interactions during the processes. Two research questions were formulated to guide the study:

There are several models depicting journalistic story creation process. Attfield and Dowell (2003) presented a research and writing process with three stages. Stage 1 Initiation involved, for example, identifying a story angle. Stage 2 Preparation included discovering and gathering information, ensuring the story angle’s originality, and increasing personal understanding. Stage 3 Production included managing the information and, if necessary, gathering additional information. Furthermore, the model depicted product constraints (e.g., word count) and resource constraints (e.g., archives or subject knowledge) in the process.

Some models concern journalists’ information seeking. Based on a systematic review, Hertzum (2022) presented a model with four information seeking stages in news story creation: (1) identifying sources (e.g., with source selection criteria such as quality or credibility); (2) interacting with sources (e.g., building trust with human sources); (3) interpreting information, including providing balancing information; and (4) managing sources (e.g., maintaining relationships with human sources). Gilbert et al. (2022) studied environmental journalists’ information seeking with a focus on their library use.

Tuazon et al.’s (2020) model focused on creativity in journalistic story creation process. The model had four phases: (1) Cognizance concerned formulating story ideas; (2) Cultivation involved gathering information, cultivating trust with human sources, and verifying and analysing information; (3) Captivation was about writing engaging stories for readers (e.g., using certain structure or word choices); and, (4) Introspection refers to reflecting on completed journalistic stories, receiving feedback, and learning from it.

Gutierrez Lopez et al. (2022) examined creativity and verification in journalistic story creation process. Their analysis showed that creativity was present throughout the process, from developing story ideas to constructing and contextualising the stories. Verification was present, for example, in finding credible sources, tracing original sources, basing the stories on facts, cross-checking sources, confirming information through experts, as well as informing readers by attributing sources and disclosing any uncertainties in the stories.

However, the studies mentioned above were conducted before the recent developments of generative AI. Therefore, this study focuses on generative AI use in journalistic story creation process. Furthermore, in this study, a journalistic story creation process is defined similar to that of Gutierrez Lopez et al. (2022) encompassing information activities from story ideation and planning to the completed journalistic story.

Recent studies examined the uses of generative AI in journalistic work across news production. Wu (2026) found that journalists used generative AI in news gathering (e.g., for finding information, generating interview questions, and transcribing), in writing and presenting (e.g., for drafting, summarisation, and coding), in news editing (e.g., for grammar), and in news promotion (e.g., for creating social media posts). Cools and Diakopoulos (2026) reported that journalists avoided using generative AI for news verification, with some exceptions, such as using it as a second pair of eyes in time-limited situations. Guenther et al. (2025) identified journalistic uses of AI and generative AI for routine tasks, but also for more creative purposes such as brainstorming or overcoming writer’s block. However, human supervision was seen as imperative in news production.

This study contributes to the existing research by examining journalistic uses of generative AI with a specific focus on related information interactions.

Information interactions with generative AI have been addressed from various viewpoints. Krakowska and Zych (2025) studied strategies used by university students when using generative AI in solving predefined tasks. The authors identified information seeking and prompt formulation strategies such as iteration (e.g., with revised or follow-up prompts). Response expansion was used for detailed information and response condensation for simplified information. A single-response strategy was used for efficiency and a step-by-step strategy for accuracy and detail. Moreover, the authors identified strategies of integrating specific sources and utilising predefined prompts in forming queries.

Mayerhofer et al. (2025) conducted a user study, with participants from a variety of educational and professional backgrounds, to examine interactions with different search modalities when solving questions about medical treatments. In the study, the participants had different strategies of utilising the search modalities. They either used traditional Web search or generative AI chat, or both. The authors also identified several tactic categories, including reflection (e.g., assessing information credibility) and meta-level tactics (e.g., choosing between search modalities).

Zhao et al. (2025) studied users’ tactics in dealing with limitations of generative AI tools. One of the findings was that when participants used generative AI tools, their understanding of how the tools worked improved. This was reflected in the tactics they used. Lee et al.’s (2025) survey study focused on knowledge workers’ critical thinking while using generative AI. As an example of the study’s qualitative results, respondents perceived that, while searching information required less effort, more effort was needed to verify AI-generated responses.

The research data for this study were collected utilising a qualitative reconstruction interview method. Fourteen journalists were interviewed about their journalistic story creation processes, uses of generative AI, and information interactions during the processes. A task-based information interaction evaluation model (Järvelin et al., 2015) informed the research design. The model enabled studying information interactions within the entire journalistic story creation process. It also provided a framework for the analysis.

In Finland, AI tools are used in several media organisations. When this study was conducted, AI tools were being developed in media organisations and adopted in journalistic work. Altogether, the journalists who participated in this study estimated their experience of using generative AI tools in journalistic tasks to be between a few months and three years. This means that in the interviews, their information interactions concerned these early uses of generative AI in journalistic tasks.

By applying the task-based information interaction evaluation model (Järvelin et al., 2015), four categories of uses of generative AI were identified in this study: (1) uses in task planning and reflective assessment, (2) uses in searching and selecting information, (3) uses in working with information, and (4) uses in synthesising and reporting. The results are presented below and then summarised in Table 2.

This study examined the uses of generative AI and related information interactions in journalistic story creation process, from a task-based perspective. The research questions were: RQ1: What types of uses of generative AI are included in journalistic story creation process? and, RQ2: What kind of information interactions do journalists exhibit when they use generative AI during the journalistic story creation process?

By applying Järvelin et al.’s (2015) task-based information interaction evaluation model, four categories of uses of generative AI were identified. They included qualitatively different use types and information interactions (see also Late & Kumpulainen, 2022). First, uses in task planning and reflective assessment concerned testing-and-learning use and support in story ideation. Testing-and-learning use included interactions such as trial and error, and it dealt with learning about and reflecting on generative AI use in story creation processes. This type of use contributes to one’s metacognitive understanding of conditions where generative AI tools could be useful, including how one personally prefers using (or not using) them (see Flavel, 1979; Pintrich, 2002). Support in story ideation concerned finding inspiration and new perspectives. However, generative AI was not expected to provide usable ideas as is. Rather it was used to stimulate one’s own ideation process. Similar observations were made in other studies (Vainikka et al., 2025; Zhao et al., 2025). Vainikka et al. stated about screenwriters that ‘Some informants described having had conversations with ChatGPT, using it as a partner to throw ideas around but at the same time keeping control and producing the final idea themselves’ (p. 62).

Second, in searching and selecting information, generative AI was used in an exploratory, topic and document content-oriented manner. Exploratory and topic-oriented uses are well-known approaches to information searching (e.g., Marchionini, 2006; Rose & Levinson, 2004). However, generative AI was perceived as enabling (for example) more extensive or more efficient information searching, providing information in a form that better meets the task performer’s current information needs compared to traditional search tools. Document content-oriented use (for example, by skimming and reading the documents while selecting them (Late & Kumpulainen, 2022)) is also a known approach in information interaction. However, this study suggests that generative AI makes these types of interactions more dynamic. It facilitates interaction with document contents that themselves are static, for example, by generating translations or summaries of the contents according to human instructions.

Third, working with information involved various uses of generative AI. For example, it was used to support creativity (e.g., creating interview questions) and to save time in more mechanistic parts of the work process (e.g., transcribing interviews). In analysing data, generative AI enabled asking questions about the data and thus supported its interpretation. It was also used to assist with procedural questions regarding coding and other tool usage. In comparison, other research indicates that support from generative AI may facilitate taking on data journalism tasks that require technical skills (De Cooker et al., 2026).

Fourth, in synthesising and reporting, generative AI was used to support drafting and editing. In drafting, participants’ information interactions concerned reviewing, revising, and rewriting the AI generated story drafts or excerpts. This resonates with the findings in Lee et al. (2025) about task stewardship. According to them, with the use of generative AI, the focus in knowledge workers’ task performance shifted toward working with and revising generative AI responses, and ensuring the quality of their work products. In editing, participants used generative AI to give suggestions or feedback regarding the text (e.g., grammar) but wanted to be in control of the process. This suggests that generative AI tools should not be overly supportive, doing more than what is asked of them for the specific editing task.

Verification of generative AI responses was a theme that recurred in different uses of generative AI. This is not surprising as verifying information is at the core of writing journalistic stories (Gutierrez Lopez et al., 2022). Participants were aware of the unreliable nature of generative AI and exhibited verification interactions, for example, in searching and selecting information, analysing data, and reviewing and revising the AI generated drafts or excerpts. Furthermore, other research reports that generative AI itself could be used for verification in journalism ‘to rapidly cross-reference statements’ (Cools & Diakopoulos, 2026, p. 887), but such use was not observed in this study.

The results of this study indicate that journalists have different preferences for using generative AI in the story creation process. Deciding whether to use generative AI tools can be seen as part of the task planning activity that, as Järvelin et al. (2015) noted, affects information interactions within the other activities in the task-based information interaction model. Moreover, the results suggest that journalists do not treat all story types the same regarding the use of generative AI. For example, generative AI was not considered useful in producing initial drafts for in-depth journalistic stories. This suggests that uses of generative AI should be supported in ways that make them applicable to different ways of working and different story creation contexts.

This study has some limitations. It focused on cognitive and behavioural activities in accordance with Järvelin et al. (2015), but did not examine the affective dimension (Kulhthau, 2004) in journalists’ information interactions. This could be a subject for future studies. Other research has shown that people seek, for example, emotional support from generative AI (Zhao et al., 2025). Second, the participants were not always entirely sure whether the tools they were using were based on generative AI (e.g., in-house tools may have integrated also other tool features than generative AI). However, the results reflect participants’ own understanding of using generative AI tools. Third, this study did not compare the uses of different types of generative AI tools, for example, in-house tools and tools available to the public. This could be addressed in future studies.

In this study, uses of generative AI were examined in the context of creating journalistic stories. The results could be transferable to other work or learning tasks where the goal is to create a written product, although this requires more research. Furthermore, this study contributes to understanding information interactions while using generative AI in real-life tasks.

Generative AI is changing journalists’ information environment. In this study, uses of generative AI and related information interactions were qualitatively examined in journalistic story creation process. A task-based information interaction evaluation model (Järvelin et al., 2015) was applied in the study. The study’s main contributions are as follows. First, the results of this study showed that generative AI was used in journalistic story creation process in various ways that are qualitatively different, from ideation and planning to the completed journalistic story. Second, this study illustrated how journalists applied some well-known approaches to information searching and selecting (namely exploratory, topical, and document content-oriented approaches) when using generative AI in their work tasks (see, for example, Late and Kumpulainen, 2022; Marchionini, 2006; Rose and Levinson, 2004). These findings increase understanding of generative AI use and information interactions, which is key to designing generative AI tools that support real-life journalistic story creation processes.

Third, this study pointed to information interactions with generative AI that require more indepth investigations in the future. Specifically, this study suggests that generative AI may facilitate journalists’ information searching and information access. However, more investigations are needed on how people search, access, or work with different types of information items (such as documents and data) in generative AI environments. Especially in a highly exploratory oriented use of generative AI, these could be intertwined. This study also suggests that journalists take various steps to verify the AI-generated content. Future studies could examine journalists’ information verification strategies when using generative AI in different information activities and work tasks.