The broad domains in which AI is explored in relation to information behaviour can be inferred from the thematic foci of the journals in which the relevant studies are published, although in this review the full texts were examined from this perspective. Collectively, the articles chart a landscape in which AI-enabled information practices are reshaping how people seek, evaluate, and use information conceptually (Hirvonen et al., 2024), through the use of AI-driven information systems (Collins et al., 2021; Rzepka & Berger, 2018), and in the context of AI use in libraries (Aboelmaged et al., 2025; Seikaly, 2022; Sousa, in press). Three specific domains of information behaviour have attracted particular attention: AI in health information seeking (Lund et al., in press; Shishegar et al., 2025); AI acceptance by consumers and its role in purchase decision-making (Jain et al., 2024; Mariani et al., 2022; Yatawara et al., in press); and AI use in research in general (Sousa, in press), and in particular for writing research reviews (Wagner et al., 2022). The review by Wagner et al. focuses on how researchers conducting research reviews can apply AI-based tools such as machine learning (tools, methods, and techniques for learning and improving task performance with experience) and natural language processing (computational tools, methods, and techniques) for analysing, interpreting, and increasingly generating natural language (p. 210). Taken together, these strands show that AI is attracting attention across a variety of research and practice areas.

Almost all authors note not only the increasing deployment of AI across different domains of practice, but also its integration with a range of digital technologies developed for different purposes. They refer to multiple AI technologies, including large language models, machine learning and deep learning, generative chatbots, and others. Collins et al. (2021) note the diversity of opinions on what AI constitutes and the absence of a common definition (p. 2). Nevertheless, attention from industry, policymakers, economists, and users to AI’s potential to increase process effectiveness and reduce costs has stimulated researchers’ interest in the development, deployment, and use of these tools (Collins et al., 2021, p. 11). The versatility of AI tools and their enhanced communicative capabilities have underpinned their market success and the widespread availability of AI-driven systems, particularly generative AI such as chatbots, AI-driven search agents, and recommendation systems (Lund et al., in press).

In this context, Hirvonen et al. (2024) deploy the concept of "information ecology," defining it as an "ecological understanding of the social integration of AI systems and the ways they are embedded in and enmeshed with various platforms, applications, and contexts rather than operated as closed and separate systems" (p. 1153). The omnipresence of AI across specialised and social media platforms, accessible via almost any digital tool and characterised by the exceptional ease of use of generative AI, has been noted by almost all review authors. This pervasive presence is the main factor driving changes in information behaviour across health, e-commerce, education, research, entertainment, and virtually every other sphere of human activity. Lund et al. (in press), for example, observe that AI tools are becoming integrated into everyday life and significantly influence health information seeking.

One emergent theme, which was anticipated given the focus of this review, concerns AI-enabled information and communication behaviour and examines how AI alters the processes of seeking, evaluating, and using information, ideally rendering them more productive, accurate, and effective overall. Hirvonen et al. (2024), whose article is most relevant to our research question, emphasise the affordances of AI systems embedded in different digital services and resources. They observe that AI systems facilitate multimodal search and active scanning, enable humanlike interaction in information retrieval, and support the creation and modification of information to address specific needs.

Lund et al. (in press) relate AI-supported information behaviour to twentieth-century models, finding that AI technologies offer convenience, speed, and enhanced Web browsing, acting as "judgment mechanisms streamlining the whole search process" (p. 5). AI thus provides shortcuts that eliminate routine tasks and delivers linear information and answers, thereby challenging models of non-linear information behaviour such as Bates’s berrypicking. Students of health sciences and medicine found AI systems valuable as medical glossaries and bilingual aids, appreciating AI-supported mentors for their immediate and democratised access to expert guidance; their teachers valued the increased efficiency, reduced workload, and scalable feedback in curriculum design (Shishegar et al., 2025, p. 16).

Library users have appreciated library chatbots for their conversational design, usability, accessibility, and effectiveness (Aboelmaged et al., 2025). In some libraries these were introduced during the COVID-19 pandemic to offer virtual assistance to distant users, not only because users appreciated the service, but also for organisational advantages such as overcoming time and space constraints, ensuring safety for staff and users, and addressing other operational challenges. Sousa (in press) builds on Aboelmaged et al. by linking AI use in academic libraries to user satisfaction in tasks including academic writing, peer review, decision-making, bibliometric data analysis, and summary writing, among other increasingly complex academic activities (p. 9). Wagner et al. (2022) detail the benefits of AI for research literature reviewers: aiding the discovery of research gaps, reducing literature search time, enhancing screening accuracy for inclusion and quality assessment, and suggesting avenues for data analysis and synthesis.

Mariani et al. (2021) established that most technology acceptance factors, such as performance expectancy, effort expectancy, social influence, and facilitating conditions, remain valid for predicting consumer acceptance of AI technologies, and can be integrated with elements like enjoyment, social presence, social cognition, and privacy (p. 764). Jain et al. (2023) determined that consumer AI adoption depends more heavily on consumers’ psychological traits, the consumption context and its objectives, and antecedents, that is, events, conditions, and causes of decisions influencing outcomes. Trust in different AI tools varies, for instance, between entertainment and information about health products (Jain et al., 2023, p. 687).

Research also shows that people attribute human-like features to AI systems and even adjust their speech when communicating with chatbots, enjoying the conversation and deriving pleasure from it, or displaying positive attitudes about a system’s responsiveness (Rzepka & Berger, 2018). Yatawara et al. (in press) support the findings of Jain et al. and Rzepka and Berger by relating user characteristics and level of knowledge to high quality experience and interaction with AI, leading to consumer satisfaction, personalisation, cognitive and emotional involvement, and loyalty across business spheres such as banking, hospitality, and e-commerce. Seikaly (2022), on the other hand, notes differences in negative and positive emotional responses to AI as they emerge across health, entertainment, and government information services research.

Limitations are understood here as the inherent boundaries, constraints, and shortcomings of information technologies that prevent them from functioning perfectly. The focus is not on technical specifications, AI types, or operational principles, but rather on user attitudes towards these limitations as identified in the selected articles. In these terms, the limitations are dynamic, arising during a phase of rapid development, and conditional, established only in relation to intended goals or functions within specific contexts and dependent on antecedents (Jain et al., 2023, p. 687). They may also be perceived differently by different users (Yatawara et al., in press, p. 6).

Inherent limitations of AI tools do not emerge as a prominent topic in the selected articles, primarily because these relate to user perceptions or authors’ assessments of AI-induced changes in behaviour rather than the tools themselves. Nevertheless, several reviews highlight user-reported limitations, including deficiencies in accessibility and usability in some tool designs (Aboelmaged et al., 2025), the sharing of data perceived as private during interactions with AI tools (Yatawara et al., in press, p. 8), and diminished psychological ownership and autonomy (Jain et al., 2023). Institutions such as libraries must prioritise user privacy and data security when integrating AI systems, and ensure compliance with relevant laws and regulations (Aboelmaged et al., 2025), entailing additional effort and expenditure.

Academic research has identified several notable limitations of AI tools in academic writing. Wagner et al. (2022) highlight concerns regarding the transparency of search methodologies and inherent biases within training data, which compromise the validity of claims and reproducibility of procedures (pp. 219-220). Additional constraints include limited interoperability with other information technology systems and usability issues in certain tools. Sousa (in press) further emphasises that AI systems generate erroneous information, including fabricated references and misinterpretations, which, although typically identified by users, undermines confidence in these technologies (p. 13).

Researchers also note that certain AI tools, particularly chatbots, characteristically produce a single response derived from what their underlying models determine to be most pertinent to a given query (Hirvonen et al., 2024; Lund et al., in press), which may result in the omission of requisite information. Conversely, AI-powered chatbots demonstrate poor capability in recognising user emotional states and generating contextually appropriate responses, adversely affecting user experience (Aboelmaged et al., 2025).

Though only a few AI limitations were found in our material, every selected article listed a variety of risks and threats, which we understand as a chance of loss or some impending peril.

Across all articles and contexts, ethics, credibility, and governance emerge as persistent concerns, especially as AI-generated content becomes more common in health information (Lund et al., in press; Shisheghar et al., 2025), scholarly communication (Sousa, in press; Wagner et al., 2022), libraries (Aboelmaged et al., 2025), information systems (Rzepka & Berger, 2018) and consumer contexts (Jain et al., 2024; Yatawara et al., in press).

According to Hirvonen et al. (2025) the affordances of AI introduced in different institutions, activity spheres and systems, “both increase and decrease levels of access and diversity of information” (p. 1159). Though they add a variety of ways in which information is consumed, they homogenize the content and support certain ways of knowing and understanding. Providing a shortcut to the required information and eliminating not only the routine search task, but also intellectual effort needed for its performance (Lund et al., in press), AI threatens deskilling of the information user, and of professionals working in health institutions (Shishehgar et al., 2025, p. 15) and libraries (Sousa, in press).

Information systems researchers are mainly concerned with the lack of understanding of the effectiveness of AI tools or how they outweigh the resources and time required when using other tools, how these effects and impact on their users can be measured or even quantified (Collins et al., 2021, pp. 12-13). According to Rzepka and Berger (2018) and Sheikaly (2022), AI tools trigger contradictory responses from users, ranging from enthusiasm and benign anthropomorphism to fear and animosity. Since the time of these reviews, the proliferation of generative AI and increased human communication with these tools has the potential to disturb users (e.g., consumers) and lead to unethical behaviour and reduced guilt (Jain et al., 2023, p. 686). The capabilities of widely accessible tools have forced academic institutions to confront new forms of plagiarism and misconduct, demonstrating that the negative impact on human behaviour is not confined to consumer contexts. Transparency and accountability are also adversely affected in academic and library settings (Sousa, in press, p. 14). Conversely, the academic community’s beneficial practices of sharing data, models, and artefacts raise concerns about potential violations of research participants’ privacy and authors’ intellectual property (Wagner et al., 2022, p. 221).

Another ethical concern emerging from consumer behaviour research is the potential for AI to violate human privacy by tracking behaviour on behalf of business actors, and there are concerns regarding lack of fairness, loss of dignity and autonomy, increased social isolation, and even dehumanisation (Mariani et al., 2021, p. 770). The authors note that these concerns may vary across different contexts, with additional emphasis on confidentiality in health-related settings, for example (Shishehgar et al., 2025), yet they constitute a barrier to the adoption of AI technologies (Mariani et al., 2021, p. 770; Sheikaly, 2022). The implementation of reliable authentication and encryption methods, as well as regular audits to identify vulnerabilities, adds additional economic costs to the growing expenses of implementing and operating these tools (Aboelmaged et al., 2025; Shishehgar et al., 2025).

AI literacy is frequently invoked across the reviewed literature as the primary response to the problems caused by AI tools and systems. Yet this framing warrants scrutiny: the reviewed articles offer divergent and sometimes incompatible accounts of what AI literacy means, who bears responsibility for developing it, and whether existing institutional infrastructures are capable of delivering it at scale. Lund et al. (in press) frame it primarily as an individual skill (prompt evaluation), Hirvonen et al. (2024) as a civic and political competence, and consumer behaviour reviews treat it as market-acquired expertise (Jain et al., 2024). These differing conceptions are not reconciled across the reviews, pointing to a conceptual gap rather than a ready-made solution. It seems that the ease of use of generative AI tools does not require high competence from the users. However, the way the AI tools operate, the benefits they generate and their limitations, cause great concern among the researchers and providers. The acceptance of, and trust in, the technology has acquired new dimensions due to the features of human-AI communication. Thus, AI literacy differs from IT, information or media literacy to some extent and requires increased technical knowledge on the part of the users of AI (Yatawara et al., 2025, p. 6).

AI tools such as ChatGPT offer a single response based on what the model deems most relevant to the user’s query, necessitating a transition from information literacy skills, such as assessing the quality of various sources, to AI literacy skills involving the ability to prompt an AI model appropriately and assess the quality of its response (Lund et al., in press, p. 4-5).

In the consumer behaviour reviews, AI literacy is addressed as consumer expertise, an antecedent for AI use (Jain et al., p. 688); as perceived competence resulting from such use (p. 689); and as a mediating factor through consumer education (p. 690), with technology literacy in relation to consumer co-creation and AI acceptance identified as a future research direction (p. 691). Technology providers can also promote acceptance by offering explanations of how AI systems operate, since this understanding appears to foster user trust (Rzepka & Berger, 2018, p. 10).

Hirvonen et al. (2025) perceive AI literacy as a civic competence providing "capabilities and resources to, first, make visible the seemingly invisible ways in which such tools shape actions and thinking; and second, to develop ways to take action toward changing practices" (p. 1161), positioning it as an obligatory element of information ecology and a wider framework for understanding AI in context.

In professional settings, specialised competence is required. Library staff will need regular training to enhance familiarity with chatbot capabilities to ensure better integration and user support (Aboelmaged et al., 2025, p. 343), while health professionals share common concerns about preparation for using AI technologies within curricula (Shishehgar et al., 2025). Information professionals may find themselves entrusted with creating resources to cultivate AI literacy by merging expertise in AI tools with conventional information literacy skills, including critical evaluation and the collection of information resources. This is particularly crucial in consumer health information seeking, given its potential impact on decisions that can be a matter of life or death (Lund et al., in press, p. 5). Sousa (in press) sees a clear relationship between the AI competence of library and information specialists and the AI literacy of the students they serve.

Developing workflows where AI outputs are systematically reviewed by humans, such as an AI-generated literature summary verified and supplemented by a librarian, or setting thresholds beyond which human intervention is mandatory for critical decisions about collection development, user privacy, or publishing ethics, allows the academic community to harness the best of both, improving productivity without compromising standards (Sousa, in press, p. 12).

Producers and operators of AI need to address corporate digital responsibility, setting shared organisational values for protecting customer rights and treating them fairly and ethically (Mariani et al., 2021, p. 770). Trust in the accuracy and relevance of AI production can be addressed through standardisation of research practice, ensuring more reliable extraction of information whilst not precluding individual expression (Wagner et al., pp. 220-221). Sousa (in press) identifies strong ethical and responsible governance of AI in academic settings as a unanimous concern in the literature he reviews, demanding explicit guidelines for AI use, user control of data, and new norms in research ethics (p. 13). Hirvonen et al. (2025) emphasise that responsible governance, transparency, and ethical guidelines must be understood not only as a systemic phenomenon but as a responsibility at each level of community and society, a concern Collins et al. (2021) echo in their future research agenda (p. 13).

The sharpest cross-domain movement is toward multidisciplinary integration, as researchers borrow theories and methods from information science, marketing, psychology, health, and information systems to understand AI’s pervasive influence (Mariani et al., 2021).