The perception and use of generative AI for science-related information search: Insights from a cross-national study

Abstract

Publicly accessible large language models like ChatGPT are emerging as novel information intermediaries, enabling easy access to a wide range of science-related information. This study presents survey data from seven countries (N = 4320) obtained in July and August 2023, focusing on the perception and use of GenAI for science-related information search. Despite the novelty of ChatGPT, a sizable proportion of respondents already reported using it to access science-related information. In addition, the study explores how these users perceive ChatGPT compared with traditional types of information intermediaries (e.g. Google Search), their knowledge of, and trust in GenAI, compared with nonusers as well as compared with those who use ChatGPT for other purposes. Overall, this study provides insights into the perception and use of GenAI at an early stage of adoption, advancing our understanding of how this emerging technology shapes public understanding of science issues as an information intermediary.

Keywords

artificial intelligence cross-national research information search survey

1. Introduction

Online audiences are increasingly turning to web-based tools and platforms to inform themselves about health, economics, and other science-related topics (Segev and Sharon, 2017). The advent of generative artificial intelligence (GenAI), exemplified by large language models such as ChatGPT, has the potential to further transform how audiences engage with science-related information. Although GenAI is not a search tool and faces concerns about the quality of its content, from a user perspective, interacting with GenAI through natural language, combined with the freedom to pose unlimited questions, can make GenAI a valuable source for obtaining factual and explanatory science-related information (Schäfer, 2023). Indeed, initial surveys hint at the potential of GenAI in science communication, for instance, regarding receiving answers to factual questions (Fletcher and Nielsen, 2024) or information about scientific research (Viden and Demokrati, 2023). However, our understanding of how individuals use GenAI to engage with information about science remains limited. Research so far has centered around broader applications of artificial intelligence (AI) technology, such as content creation in science journalism (e.g. Lermann Henestrosa et al., 2023; Maiden et al., 2023) or attitudes toward AI as an emerging technology (e.g. Calice et al., 2022), while tending to focus on specific countries or regions (e.g. Skjuve et al., 2023). Yet, the influence of GenAI on science communication should be considered within a cross-national perspective, taking into account the cultural, social, and economic factors that shape engagement with science communication (Gascoigne et al., 2020) and AI.

To address this need, in this study, we present results from a cross-national online survey carried out in seven countries (N_total = 4320): Australia, Denmark, Germany, Israel, South Korea, Taiwan, and the United States of America (USA). The purpose of this study is twofold. First, we offer a cross-country overview of the reported use of GenAI for searching science-related information, with specific focus on ChatGPT as it is the most widely recognized and used (Fletcher and Nielsen, 2024). Second, we delve into the demographics of ChatGPT users engaged in science-related information searches across countries, comparing their knowledge about and their level of trust in (Gen)AI¹ to nonusers and users who utilize ChatGPT for purposes other than science-related information searches.

To our knowledge, this is one of the first studies to explore GenAI’s role in science-information seeking in a cross-national setting. Rooted in the traditions of the technology acceptance model (Davis, 1989), the contributions of this study extend beyond providing empirical insights into an emerging channel for science communication. They capture a distinctive moment in the global adoption of GenAI, at a point in the early stages of use of a technology that has the potential to transform both the theory and practice of science communication.

2. Generative AI as a new intermediary for science-related information

GenAI sets itself apart from established information intermediaries in its ability to produce human-like content autonomously through both generalization and creativity employed by the AI itself. Unlike media that act primarily as channels for human-generated content, GenAI is designed and interpreted as a responsive message source. It steps into the role of a communicator capable of conveying original content in a manner that is socially meaningful and meets human communicative needs (Guzman and Lewis, 2020).

According to Fletcher and Nielsen (2024), ChatGPT, a large language model developed by OpenAI for conversational usage, is currently the most widely used GenAI technology. In a recent study, English-speaking ChatGPT users reported appreciating the tool for providing detailed information in response to complex inquiries, as well as for its ability to facilitate comprehension of long(er) responses through a clear structure and straightforward language (Skjuve et al., 2023). Similarly, in a nationwide survey, half of the German respondents indicated that they were satisfied with ChatGPT’s potential for explaining complicated scientific issues and for the opportunity to ask follow-up questions (Wissenschaft im Dialog, 2023). This suggests that users value ChatGPT’s capacity to convey otherwise complex and dense information in a manner that enhances comprehension. However, to better understand the potential impact of GenAI on how individuals inform themselves about science, it is necessary to assess both the extent of its use for this purpose and how users perceive its effectiveness compared with tools like Google Search—an aspect that is also highlighted in the technology acceptance model (Davis, 1989). The first set of research questions (RQ) that guide our study are as follows:

RQ1a: What proportion of people in the countries under study use ChatGPT to search for science-related information?

RQ1b: How do users perceive science-related information retrieval with ChatGPT compared with Google Search?

In sharp contrast to search engines, which provide users with diverse sources to choose from, ChatGPT gives tailored responses to each query. However, ChatGPT (in its 2023 iteration) is based on a large language model that generates these answers based on complex statistical patterns from training data and lacks an intrinsic understanding of the conveyed content. This creates a situation where non-factual content may convincingly appear as facts (Angelis et al., 2023). A study by Spitale et al. (2023) underscores that GPT models can both inform and misinform individuals on health-related issues. Although some users appear to be aware of this problem (Skjuve et al., 2023), detecting incorrect information requires knowledge that users often lack. To assess the AI knowledge of individuals who use ChatGPT for science-related information retrieval and see if they constitute a distinct segment compared with other subpopulations, we ask:

RQ2: What is the level of factual knowledge about (Gen)AI among ChatGPT users who engage the model for science-related information searches compared with nonusers and users who utilize ChatGPT for other purposes?

Furthermore, the design of AI technologies, particularly commercial GenAI models, is often non-transparent. The training data behind these models is frequently withheld from the public, and tech companies tend to guard the inner workings of their AI systems as proprietary information (Van Dis et al., 2023). In navigating the black-box nature of AI, trust emerges as a pivotal factor (Choung et al., 2023; Rheu et al., 2021). Trust has also been found to drive users’ intentions to accept such technologies (Kelly et al., 2023). Notably, survey data from Germany indicates a lack of trust in ChatGPT when disseminating science-related information (Wissenschaft im Dialog, 2023). This is attributed to concerns about misinformation, revealing a critical need to address trust in the context of AI-driven science communication. With this in mind, to further characterize individuals who use ChatGPT for science-related information retrieval, we ask:

RQ3: What is the level of trust in GenAI among ChatGPT users who engage the model for science-related information searches, and how does it compare with that of nonusers and users who utilize ChatGPT for other purposes?

3. Method

Sample

To address our RQs, we conducted an online survey gathering data in Australia, Denmark, Germany, Israel, South Korea, Taiwan, and the USA (N_total = 4320) between July and August, 2023. The countries examined are affluent and possess advanced technological infrastructures; however, they vary in their science communication landscapes (Gascoigne et al., 2020) and their general attitudes toward AI (Neudert et al., 2020). It is important to note that the selection of countries was not systematic but rather achieved through professional networks such as the Network for the Public Communication of Science and Technology (PCST). In each country, our samples were representative of the respective national (Internet) populations in terms of age, gender, and education.² We collected data on knowledge about, trust in, and use of GenAI through online access panels, with the questionnaire translated into the relevant primary languages. The numbers of respondents per country are as follows: n_AUS = 552, n_DEN = 504, n_GER = 566, n_ISR = 500, n_KOR = 642, n_TWN = 504, n_USA = 1052. For a demographic breakdown of the sample by country, see Supplementary B (Table S1) in the Supplemental Material. The English version of the questionnaire can be found in Supplementary A. In this research note, we only present findings on a subset of all questions asked.

Measurements

Using ChatGPT for searching science-related information

This variable is based on an original questionnaire (similar to Fletcher and Nielsen, 2024), inquiring about respondents’ general experience with five AI applications—including ChatGPT—and with Google Search. Respondents were shown the list of applications they reported having used, followed by a series of questions about their use of those applications for science-related information search, including if they were confident in finding what they needed and whether they were content with the science information they found—each measured as a 5-point scale single item. To ensure shared understanding, we offered respondents a definition of science-related information search.

Knowledge about AI technology

Drawing from the literature (Long and Magerko, 2020) and in collaboration with AI experts, we developed nine statements to assess respondents’ factual understanding of (Gen)AI technology. Respondents were asked whether those statements were true or false. In addition, they could choose the option “I don’t know” if unsure. Six of the nine statements concentrate on the functioning of (Gen)AI (sum score from 0 to 6; M = 3.5, SD = 1.5), while the remaining three address the quality of the information provided (sum score from 0 to 3; M = 1.5, SD = 1.1). The two types of knowledge are weakly related (Pearson’s r = .20, p < .001); hence, we treated them as separate dimensions. In line with prior research (Calice et al., 2022), we provided respondents with definitions of both AI and GenAI.

Trust in GenAI

Based on prior research (Choung et al., 2023; Weidmüller, 2022), we measured trust in GenAI using 15 items that encompassed aspects of both human-like trust (i.e. benevolence/helpfulness, integrity/reliability) and functionality trust (i.e. competence/functionality). In the context of science communication, we introduced dialogue and transparency as two additional dimensions deemed relevant for evaluating trust in scientists (Reif et al., 2024). The 15 items, which were measured on 5-point scales (1 = “strongly disagree,” 5 = “strongly agree”), formed a reliable scale (Cronbach’s α = .95, M = 3.4, SD = 0.8; Cronbach’s α for each dimension ranges between .76 and .80). Based on the results of an exploratory factor analysis, we used the overall score for the subsequent analyses.

4. Results

RQ1a: What proportion of people in the countries under study use ChatGPT to search for science-related information?

As displayed in Figure 1, analyzing respondents’ reported experiences with ChatGPT revealed varying usage across countries. In Australia and Denmark, 9% (95% CI = [0.06, 0.11]; 95% CI = [0.07, 0.12]) of respondents reported regular use, while Germany and the USA had 10% (95% CI = [0.08, 0.13]) and 11% (95% CI = [0.09, 0.13]) regular users, respectively. In Israel and South Korea, 16% (95% CI = [0.13, 0.19]) and 17% (95% CI = [0.14, 0.2] of respondents, respectively, reported using the chatbot on a regular basis. Taiwan had the highest proportion of regular users (26%; 95% CI = [0.22, 0.3]).

Figure 1.

Experience with ChatGPT across countries. For color figure please see online version.

In order to explore the significance of ChatGPT as a search tool for science-related information, we centered our analysis on regular users of ChatGPT (n = 555; 13% of the total sample). In total, 67% of regular users reported using ChatGPT for science-related information searches (n = 371; 9% of the total sample). The highest usage rate was again observed in Taiwan, with 84% of regular users employing ChatGPT for science-related searches. Across the remaining countries, the proportions were as follows: 69% of regular ChatGPT users in the USA, 65% in South Korea, 64% in Israel, 54% in Denmark and Germany, and 53% in Australia (see Table 1).

Table 1.

Use for science-related information searches among regular users of ChatGPT.

	Taiwan (25.6%, 95% CI = [0.22, 0.3])^a		USA (10.7%, 95% CI = [0.09, 0.13])		South Korea (17.3%, 95% CI = [0.14, 0.2])		Israel (15.5%, 95% CI = [0.13, 0.19])		Germany (10.3%, 95% CI = [0.08, 0.13])		Denmark (8.8%, 95% CI = [0.07, 0.12])		Australia (8.5%, 95% CI = [0.06, 0.11])		Total (13.4%, 95% CI = [0.12, 0.14])
	n	% [95% CI]	n	% [95% CI]	n	% [95% CI]	n	% [95% CI]	n	% [95% CI]	n	% [95% CI]	n	% [95% CI]	n	% [95% CI]
Used for science-related information searches	105	84.0 [0.77, 0.89]	72	68.6 [0.59, 0.77]	68	64.8 [0.55, 0.73]	48	64.0 [0.53, 0.74]	31	54.4 [0.42, 0.67]	23	53.5 [0.39, 0.67]	24	53.3 [0.41, 0.65]	371	66.8 [0.63, 0.71]
Not used for science-related information searches	20	16.0 [0.11, 0.23]	33	31.4 [0.23, 0.41]	37	35.2 [0.27, 0.45]	27	36.0 [0.26, 0.47]	26	45.6 [0.33, 0.58]	20	46.5 [0.33, 0.61]	21	46.7 [0.35, 0.59]	184	33.2 [0.29, 0.37]

Subsample of regular ChatGPT users.

Proportion of regular ChatGPT users by country. 95% CI calculated using Wilson score interval.

RQ1b: How do users perceive science-related information retrieval with ChatGPT compared with Google Search?

Across the seven countries, we found that users who employed ChatGPT for science-related information expressed contentment with the information provided (M = 4.0, SD = 0.9) and felt confident in their ability to find the necessary information (M = 3.6, SD = 1.0). Comparing ChatGPT with Google Search, there was a small significant difference in user satisfaction with the information provided (t(263) = −2.038, p = .043). Also, a significant difference emerged in users’ confidence in finding the information they sought (t(276) = −3.798, p < .001), with higher scores for Google Search in both cases. At the country level, however, a significant difference was particularly evident in Germany and Israel, where respondents reported significantly higher confidence in finding information using Google Search (see Table 2).

Table 2.

Confidence and contentment the last time ChatGPT and Google Search were used for science-related information searches.

		Germany		Taiwan		USA		South Korea		Denmark		Israel		Australia		Total
		n	M (SD)	n	M (SD)	n	M (SD)	n	M (SD)	n	M (SD)	n	M (SD)	n	M (SD)	n	M (SD)
Contentment with the science information that was found	ChatGPT	27	4.3 (0.9)	58	4.2 (0.7)	57	4.1 (1.2)	49	3.9 (0.7)	23	3.9 (0.9)	35	3.8 (1.0)	15	3.6 (0.6)	264	4.0 (0.9)⁴
Contentment with the science information that was found	Google Search	27	4.5 (0.6)	58	4.2 (0.6)	57	4.3 (0.9)	49	4.1 (0.7)	23	3.9 (0.9)	35	4.0 (1.0)	15	4.1 (0.8)	264	4.1 (0.8)⁴
Confidence that you can find what you need	ChatGPT	29	4.1 (0.8)^1,a,b	57	3.7 (0.8)	62	3.9 (1.1)^c	51	3.4 (1.0)^b	23	3.5 (0.8)	40	3.2 (1.0)^2,a,c	15	3.2 (1.1)	277	3.6 (1.0)³
Confidence that you can find what you need	Google Search	29	4.6 (0.6)^1,a,c,d,e	57	3.5 (0.6)^e,g	62	4.2 (1.0)^f,g	51	3.6 (1.0)^d,f	23	3.5 (1.2)^a	40	3.9 (1.1)^2,c	15	3.7 (1.0)	277	3.9 (1.0)³

Subsample of regular ChatGPT users who use the model for science-related information searches. Mean values range from 1 to 5, with 5 indicating high contentment/confidence. Mean values with a common exponent differ with p < .05 in the Bonferroni post hoc test of an ANOVA or in the paired samples t-test. Superscript numbers denote a comparison between user groups within one country (columns), while superscript letters denote a comparison between countries within one user group (rows).

RQ2: What is the level of factual knowledge about (Gen)AI among ChatGPT users who engage the model for science-related information searches, compared with nonusers and users who utilize ChatGPT for other purposes?

As given in Table 3, knowledge about the functioning of (Gen)AI varies among science-information seekers: The six-point sum score ranges from M = 3.5 (SD = 1.3) in Australia to M = 4.6 (SD = 0.9) in Taiwan. Across the seven countries studied, this group demonstrates significantly higher knowledge about how (Gen)AI functions compared with nonusers (mean difference = .7, p < .001) (F(2, 4316) = 36.66, p < .001). However, there is no significant difference between the two groups of ChatGPT users, nor are there any significant group differences in Australia and Taiwan. Regarding knowledge about the quality of information provided, among science-information seekers, the three-point sum score ranges from M = 1.0 (SD = 0.9) in Taiwan to M = 2.2 in Israel (SD = 0.9). Significant differences are only observed in Israel, where users engaging ChatGPT for science-related searches know more about the epistemic limitations of (Gen)AI than nonusers (F(2, 497) = 7.45, p < .001; mean difference = .5). Notably, across all countries, there is a prevalent use of “I don’t know” responses (see Supplementary B, Table S2).

Table 3.

Factual knowledge about AI among different user groups.

	Denmark		Taiwan		Israel		Germany		South Korea		USA		Australia		Total
	n	M (SD)	n	M (SD)	n	M (SD)	n	M (SD)	n	M (SD)	n	M (SD)	n	M (SD)	n	M (SD)
Knowledge about the functioning of AI (sum index 0-6)
ChatGPT users: science-related information searches	23	4.6 (0.9)^a,1	105	4.3 (1.0)^c	48	4.1 (0.9)¹	31	4.1 (1.2)¹	68	4.0 (1.1)¹	72	3.8 (1.1)¹	24	3.5 (1.3)^a,c	371	4.1 (1.1)¹
ChatGPT users: no science-related information searches	20	4.2 (1.3)	20	3.6 (1.8)	27	4.3 (1.2)^d,2	26	3.9 (1.5)	37	3.6 (1.3)	33	3.7 (1.3)	21	3.0 (1.6)^d	184	3.8 (1.4)²
Nonusers	461	3.4 (1.5)^e,1	379	4.0 (1.4)^e,f,h,j,k,m	425	3.5 (1.3)^f,1,2	509	3.3 (1.6)^g,h,1	537	3.6 (1.5)^g,i,k,I,1	947	3.2 (1.6)^i,j,1	507	3.2 (1.5)^l,m	3765	3.4 (1.5)^1,2
Knowledge about quality of information (sum index 0–3)
ChatGPT users: science-related information searches	23	2.1 (1.1)^a,b	105	1.0 (0.9)^b,d,f	48	2.2 (0.9)^c,d,e,1	31	1.9 (1.1)^f	68	1.3 (1.1)^a,c	72	1.5 (1.1)^e	24	1.5 (1.3)	371	1.5 (1.1)
ChatGPT users: no science-related information searches	20	2.1 (1.1)^g	20	1.0 (0.9)^g,I,j	27	2.1 (1.1)^h,i	26	1.6 (1.1)	37	1.2 (1.0)^h	33	1.9 (1.1)^j	21	1.9 (1.0)	184	1.7 (1.1)
Nonusers	461	1.7 (1.1)^k,l	379	0.8 (0.9)^l,m,p,r,s,t	425	1.7 (1.1)^m,1	509	1.9 (1.0)^k,n,o,p,q	537	1.5 (1.1)^o,s	947	1.5 (1.2)^n,r	507	1.6 (1.2)^q,t	3765	1.5 (1.1)

Sum index of correct answers; mean values with a common exponent differ with p < .05 in the Bonferroni post hoc test of an ANOVA. Superscript numbers denote a comparison between user groups within one country (columns), while superscript letters denote a comparison between countries within one user group (rows).

RQ3: What is the level of trust in GenAI among ChatGPT users who engage the model for science-related information searches, compared with nonusers and users who utilize ChatGPT for other purposes?

Among ChatGPT’s regular users who employed the tool for science-related information searches, the levels of trust in GenAI ranged from M = 3.4 (SD = 06. and SD = 0.5, respectively) in Denmark and Israel to M = 4.1 (SD = 0.5) in Taiwan (see Table 4). Across all countries (F(2, 2935) = 69.34, p < .001), trust in GenAI was higher for ChatGPT users who engage the model in science-related information search than for nonusers (p < .001); but no significant difference was observed for ChatGPT users who did not engage in science-related searches (p = .06). At the country level, the significant difference, however, was not evident in Israel (F(2, 286) = 0.1, p = .90). In all subpopulations, a substantial proportion found themselves unable to provide responses to items related to trust in GenAI, reflected in the prevalent occurrence of “I don’t know” responses.

Table 4.

Trust in generative AI among different user groups.

	Taiwan		South Korea		Germany		USA		Australia		Israel		Denmark		Total
	n	M (SD)	n	M (SD)	n	M (SD)	n	M (SD)	n	M (SD)	n	M (SD)	n	M (SD)	n	M (SD)
ChatGPT users: science-related information searches	99	4.1 (0.5)^a,b,1	62	4.0 (0.5)^c,d,1,2	27	3.9 (0.8)^e,1	63	3.8 (0.9)¹	19	3.7 (0.8)¹	36	3.4 (0.5)^a,c,e	20	3.4 (0.6)^b,d,1	326	3.9 (0.7)¹
ChatGPT users: no science-related information searches	18	4.0 (0.5)^a	34	3.6 (0.6)¹	23	4.0 (0.7)²	27	3.8 (0.8)²	15	3.3 (0.7)^a	19	3.3 (0.7)	9	3.3 (0.9)	145	3.7 (0.7)²
Nonusers	348	3.8 (0.6)^{a,b,c,d,e,f,1}	440	3.5 (0.6)^a,g,h,I,j,2	353	3.2 (0.8)^b,g,k,q,1,2	564	3.3 (0.9)^c,h,m,n,1,2	315	3.1 (0.8)^d,I,m,o,1	234	3.4 (0.7)^e,k,o,p	213	2.9 (0.7)^f,j,q,n,p,1	2467	3.4 (0.8)^1,2

Mean index (1–5; Cronbach’s α = .95); mean values with a common exponent differ with p < .05 in the Bonferroni post hoc test of an ANOVA. Superscript numbers denote a comparison between user groups within one country (columns), while superscript letters denote a comparison between countries within one user group (rows).

5. Discussion

This research note explores the evolving landscape of new intermediaries for science-related information searches, by providing insights into the cross-national use of GenAI, exemplified by ChatGPT. 9% of the total sample reported actively using this tool for information retrieval in scientific domains, which may highlight the start of a potentially transformative role of AI in shaping how individuals access and engage with scientific knowledge (Schäfer, 2023). Although differences across countries must be interpreted with caution due to the small sample sizes and the resulting sampling variability, ChatGPT users in Taiwan, the USA, South Korea, and Israel seem to display a keen interest in leveraging ChatGPT for science-related information searches. Taiwan, South Korea, and Israel are known for having a receptive environment for AI adoption (Getz et al., 2020; Johnson and Tyson, 2020). In addition, while there is widespread belief in science and technology for national prosperity in these countries, their science communication landscapes are still developing (Baram-Tsabari et al., 2020; Huang et al., 2020; Kim, 2020), potentially creating an opportunity for new intermediaries like ChatGPT to fill a gap. In the USA, however, while there is evident interest in using AI tools like ChatGPT (Fletcher and Nielsen, 2024), concerns about their impact on daily life and privacy remain prominent (Tyson and Kikuchi, 2023).

According to the technology acceptance model, the perceived ease of use and usefulness of a technology drives its adoption (Davis, 1989). Our data indicate that ChatGPT users in Germany and Israel have higher confidence in finding the desired information via Google Search, representing an established science-information search tool. However, at the country level, users evaluate ChatGPT and Google Search comparably in terms of satisfaction with the information provided. This is noteworthy because, while GenAI technologies can aid in information retrieval, they clearly distinguish themselves from traditional intermediaries such as search engines—particularly regarding the quality of factual information (Angelis et al., 2023).

For science communication research, the tension between the probabilistic nature of AI outputs—where answers are generated by predicting the most likely next words or sequences based on data patterns rather than directly retrieving facts—and the public’s need for accurate scientific information for quality decision-making is of significant theoretical importance. Our study contributes to this by showing that, in all seven countries surveyed, ChatGPT users seeking science-related information are at least as informed about the functioning and epistemic limitations of AI and GenAI as nonusers. In most countries, knowledge about how AI technology works is higher among users compared with nonusers. Moreover, apart from Israel where no significant differences were found, the level of trust in AI is significantly higher among ChatGPT users compared with nonusers. This alignment of characteristics among users potentially suggests a specific subpopulation that actively seeks out AI-powered solutions for scientific information retrieval (Bao et al., 2022). However, given the substantial proportion of respondents answering “I don’t know” in all subpopulations of our study, it appears that many people do not feel adequately informed—at least, at this stage of the adoption of GenAI tools for daily use—to express definitive opinions on trust in AI systems or to report whether a statement about (Gen)AI is true or false. Also, there were no significant differences between users employing ChatGPT for science versus other purposes, raising questions about whether science-related information is treated just like any other type of information in this context. This underscores the need for ongoing research in this area, also taking into account country-specific engagement with science and AI and additional variables; for instance, we also found that respondents who use ChatGPT for science-related information searches were on average younger, often male, and tended to have higher levels of education (see Supplementary B, Table S1).

Acknowledging both the novelty of the technology and the resulting small sample sizes within regular users of ChatGPT seeking science-related information, we refrained from further characterizing this subpopulation at a country-level. Furthermore, in the absence of established scales for GenAI, we relied on self-created items, which need to be further validated in future studies. Our selection of countries also extends beyond the traditionally well-researched areas of Europe and the USA, albeit with a continued emphasis on Western-oriented countries. As mentioned, the selection process was not guided by a systematic approach. Consequently, the findings of this study should not be generalized globally. Future research would benefit from adopting a more rigorous sampling strategy, assembling cross-national samples based on a systematic variation of criteria such as the pace of digital adoption, the presence of AI policies, or the status of the science communication landscape.

Despite these limitations, our cross-national survey provides empirical insights at an early stage of an emerging technology’s integration into everyday life, focusing on users’ science-related searches. As such, it paves the way for future research to delve more deeply into the global changes in science communication practices resulting from GenAI.

Supplemental Material

sj-docx-1-pus-10.1177_09636625241308493 – Supplemental material for The perception and use of generative AI for science-related information search: Insights from a cross-national study

Supplemental material, sj-docx-1-pus-10.1177_09636625241308493 for The perception and use of generative AI for science-related information search: Insights from a cross-national study by Esther Greussing, Lars Guenther, Ayelet Baram-Tsabari, Shakked Dabran-Zivan, Evelyn Jonas, Inbal Klein-Avraham, Monika Taddicken, Torben Esbo Agergaard, Becca Beets, Dominique Brossard, Anwesha Chakraborty, Antoinette Fage-Butler, Chun-Ju Huang, Siddharth Kankaria, Yin-Yueh Lo, Kristian H. Nielsen, Michelle Riedlinger and Hyunjin Song in Public Understanding of Science

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by Niedersächsisches Vorab, Research Cooperation Lower Saxony—Israel and Lower Saxony Ministry for Science and Culture (MWK), Germany [Grant No. 11-76251-2345/2021 (ZN 3854)]; Aarhus University Research Foundation, Grant No. AUFF-E-2019-9-13; Morgridge Institute for Research & Wisconsin Alumni Research Foundation (WARF); National Science and Technology Council, Taiwan (Grant No. MOST 110-2511-H-194-003-MY3 and Grant No. NSTC 112-2628-H-128-001-MY3); and Yonsei University Research Grant (Funding no: 2024-22-0385).

ORCID iDs

Esther Greussing

Lars Guenther

Ayelet Baram-Tsabari

Evelyn Jonas

Inbal Klein-Avraham

Monika Taddicken

Becca Beets

Dominique Brossard

Anwesha Chakraborty

Antoinette Fage-Butler

Chun-Ju Huang

Siddharth Kankaria

Yin-Yueh Lo

Kristian H. Nielsen

Michelle Riedlinger

Hyunjin Song

Supplemental material

Supplemental material for this article is available online.

Notes

Author biographies

Esther Greussing is a postdoctoral researcher at the Institute for Communication Science at Technische Universität Braunschweig in Germany. Her research focuses on the digitalization of science communication, with a particular emphasis on the role of nonhuman agents in the communication process. She explores the conditions and consequences of the use of these agents within the context of the knowledge society.

Lars Guenther (PhD, 2015, at Friedrich Schiller University Jena, Germany) is Professor of Communication Science at LMU Munich’s Department of Media and Communication in Germany, and Extraordinary Associate Professor at the Centre for Research on Evaluation, Science and Technology (CREST) at Stellenbosch University in South Africa. He is interested into public perceptions of (controversial) science, science and health journalism, trust in science, as well as the public communication about risks and scientific (un)certainty.

Ayelet Baram-Tsabari is a Professor of Science Education and Communication at the Faculty of Education in Science and Technology at the Technion—Israel Institute of Technology. Her research program focuses on the relevance of science education to public engagement with science and on training scientists for effective science communication.

Shakked Dabran-Zivan is a PhD student at the Faculty of Education in Science & Technology at the Technion—Israel Institute of Technology.

Evelyn Jonas is a Research Assistant and PhD candidate at the Institute for Communication Science at Technische Universität Braunschweig, Germany. She holds a master’s degree in Media Technology and Communication from the same institution. Currently, her PhD project focuses on user perceptions of trustworthiness and the usage of AI as an intermediary for complex and science-related information.

Inbal Klein-Avraham is a postdoctoral fellow at the Faculty of Education in Science and Technology, Technion—Israel Institute of Technology. Her current research focuses on publics’ critical engagement with science via AI-based information technologies. Her previous studies were published, inter alia, in New Media and Society, Journalism Studies, and more.

Monika Taddicken is a Professor in Communication Science at the Technische Universität Braunschweig (Germany). Her research interests include science communication with a special focus on new media environments and user engagement.

Torben Esbo Agergaard is a PhD student at Aarhus University within the field of science studies. His project concerns ethical and epistemic aspects of explainable artificial intelligence.

Becca Beets is an Assistant Professor in the Department of Communication at the University of Maryland, College Park. Her research examines emerging areas of science and technology, with a focus on public opinion, engagement, and communication in new media environments.

Dominique Brossard is Professor and Chair in the department of Life Sciences Communication at the University of Wisconsin-Madison. Her research is situated at the intersection of science, media and policy as related to new technologies.

Anwesha Chakraborty is a postdoctoral fellow at the Department of Communications, Humanities and International Studies (DISCUI) at the University of Urbino Carlo Bo in Italy. Her research focuses broadly on the study of technology as a social good. She has contributed to a wide range of projects related to digital governance, responsible innovation, and more recently, the use of generative AI in combatting disinformation on social media.

Antoinette Fage-Butler is an Associate Professor in the School of Communication and Culture at Aarhus University whose research interests center on the communication of science, risk and trust.

Chun-Ju Huang specializes in science communication and public understanding of science, with current research focusing on scientific uncertainty communication in the post-truth era.

Siddharth Kankaria works at the intersection of science communication research, practice and teaching in India. He is currently a Senior Manager for Global Research Alliances at Ashoka University and serves on the Scientific Committee of the PCST Network.

Yin-Yueh Lo focuses on cross-cultural aspects of science communication, drawing insights from international collaborative projects to examine cultural variations in public science communication practices.

Kristian H. Nielsen is an Associate Professor in science communication at Aarhus University with a research interest in trust in science, citizen science and science history.

Michelle Riedlinger is a Chief Investigator in the Digital Media Research Centre at the Queensland University of Technology. Her research interests include emerging environmental, agricultural, and health research communication practices, roles for “alternative” science communicators, online fact checking, platformized engagement with scientific research, and, most recently, generative authenticity.

Hyunjin Song is an Associate Professor in the Department of Communication at Yonsei University, Seoul, South Korea. His research centers on statistical modeling of social networks and explores how algorithmically driven information environments influence individuals’ reception of political and scientific messages, as well as the broader consequences of these effects.

References

Angelis

de Baglivo

Arzilli

Privitera

Ferragina

Tozzi

, et al. (2023) ChatGPT and the rise of large language models: The new AI-driven infodemic threat in public health. Frontiers in Public Health 11: 1166120.

Bao

Krause

Calice

Scheufele

Wirz

Brossard

, et al. (2022) Whose AI? How different publics think about AI and its social impacts. Computers in Human Behavior 130: 107182.

Baram-Tsabari

Orr

Baer

Garty

Golumbic

Halevy

, et al. (2020) Israel: Developed science, developing science communication. In: Gascoigne

Schiele

Leach

, et al. (eds) Communicating Science: A Global Perspective. Canberra, ACT, Australia: ANU Press, pp. 443–468.

Calice

Bao

Newman

Scheufele

Brossard

Xenos

(2022) U.S. public attitudes on artificial intelligence. Report. Available at: osf.io/k82d6

Choung

David

Ross

(2023) Trust in AI and its role in the acceptance of AI technologies. International Journal of Human–Computer Interaction 39(9): 1727–1739.

Davis

(1989) Perceived usefulness, perceived ease of use, and user acceptance of information technology. MIS Quarterly 13(3): 319–340.

Fletcher

Nielsen

(2024) What Does the Public in Six Countries Think of Generative AI in News? Oxford: Reuters Institute for the Study of Journalism. DOI: 10.60625/risj-4zb8-cg87

Gascoigne

Schiele

Leach

Riedlinger

Massarani

Lewenstein

, et al. (eds) (2020) Communicating Science: A Global Perspective. Canberra, ACT, Australia: ANU Press.

Getz

Buchnik

Zatcovetsky

(2020) Science, technology and innovation indicators in Israel: An international comparison—2019—part A—key figures. Available at: www.neaman.org.il/EN/Science-Technology-and-Innovation-Indicators-in-Israel-An-International-Comparison-2019-Part-a-Key-figures

10.

Guzman

Lewis

(2020) Artificial intelligence and communication: A Human–Machine Communication research agenda. New Media & Society 22(1): 70–86.

11.

Huang

C-J

Y-Y

(2020) Taiwan. From nationalising science to democratising science. In: Gascoigne

Schiele

Leach

, et al. (eds) Communicating Science: A Global Perspective. Canberra, ACT, Australia: ANU Press, pp. 849–864.

12.

Johnson

Tyson

(2020) People globally offer mixed views of the impact of artificial intelligence, job automation on society. Available at: https://www.pewresearch.org/short-reads/2020/12/15/people-globally-offer-mixed-views-of-the-impact-of-artificial-intelligence-job-automation-on-society/

13.

Kelly

Kaye

S-A

Oviedo-Trespalacios

(2023) What factors contribute to the acceptance of artificial intelligence? A systematic review. Telematics and Informatics 77: 101925.

14.

Kim

H-S

(2020) South Korea. A different exemplar. In: Gascoigne

Schiele

Leach

, et al. (eds) Communicating Science: A Global Perspective. Canberra, ACT, Australia: ANU Press, pp. 801–824.

15.

Lermann Henestrosa

Greving

Kimmerle

(2023) Automated journalism: The effects of AI authorship and evaluative information on the perception of a science journalism article. Computers in Human Behavior 138: 107445.

16.

Long

Magerko

(2020) What is AI literacy? Competencies and design considerations. In: Proceedings of the 2020 CHI conference on human factors in computing systems, Honolulu, HI, 25–30 April, pp. 1–16. New York: ACM.

17.

Maiden

Zachos

Franks

Nyre

Linden

C-G

(2023) Automating science journalism tasks: Emerging opportunities. Journalism Practice. Epub ahead of print 26 June. DOI: 10.1080/17512786.2023.2226116.

18.

Neudert

L-M

Knuutila

Howard

(2020) Global Attitudes Towards AI, Machine Learning & Automated Decision Making. Oxford: Oxford Commission on AI & Good Governance. Available at: https://oxcaigg.oii.ox.ac.uk

19.

Reif

Taddicken

Guenther

Schröder

Weingart

(2024) The public trust in science scale (puts): A multilevel and multidimensional approach. Science Communication. Epub ahead of print 20 December. DOI: 10.1177/10755470241302758.

20.

Rheu

Shin

Peng

Huh-Yoo

(2021) Systematic review: Trust-building factors and implications for conversational agent design. International Journal of Human–Computer Interaction 37(1): 81–96.

21.

Schäfer

(2023) The notorious GPT: Science communication in the age of artificial intelligence. Journal of Science Communication 22(2): Y02.

22.

Segev

Sharon

(2017) Temporal patterns of scientific information-seeking on Google and Wikipedia. Public Understanding of Science 26(8): 969–985.

23.

Skjuve

Følstad

Brandtzaeg

(2023) The user experience of ChatGPT: Findings from a questionnaire study of early users. In: Proceedings of the 5th international conference on conversational user interfaces (eds Lee

Munteanu

Porcheron

Trippas

Völkel

), Eindhoven, 19–21 July, pp. 1–10. New York: ACM.

24.

Spitale

Biller-Andorno

Germani

(2023) AI model GPT-3 (dis)informs us better than humans. Science Advances 9(26): eadh1850.

25.

Tyson

Kikuchi

(2023) Growing public concern about the role of artificial intelligence in daily life. Available at: https://www.pewresearch.org/short-reads/2023/08/28/growing-public-concern-about-the-role-of-artificial-intelligence-in-daily-life/

26.

Van Dis

EAM

Bollen

Zuidema

Van Rooij

Bockting

(2023) ChatGPT: Five priorities for research. Nature 614(7947): 224–226.

27.

Viden Demokrati (2023) V&d-barometeret: Befolkningens forhold til videnskab [V&D barometer: The population’s relationship to science]. Available at: https://videnogdemokrati.dk/app/uploads/2023/12/VidenDemokrati-barometeret-2023-dec-download.pdf

28.

Weidmüller

(2022) Human, hybrid, or machine? Exploring the trustworthiness of voice-based assistants. Human-Machine Communication 4: 85–110.

29.

Wissenschaft im Dialog (2023) Wissenschaftsbarometer 2023 [Science barometer 2023]. Available at: https://wissenschaft-im-dialog.de/documents/47/WiD-Wissenschaftsbarometer2023_Broschuere_web.pdf

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.08 MB