The survey gathered a total of 325 responses. As some participants did not go beyond the demographic questions, 39 (12%) responses were excluded before analysis, resulting in 286 (88%) valid responses (162 women, 123 men, and 1 who preferred not to say), with a mean age of 72.87 (SD 5.67) years. Partial responses, where participants completed part of the main survey, were included. Participants were recruited via word of mouth, social networking sites (eg, X and LinkedIn), flyers, and the Join Dementia Research platform, with 272 participants identifying as currently residing in England and 14 in Scotland. Education levels are summarized in Table 1. Participants reported diverse professional and vocational backgrounds, including health care, education, government, engineering, business, and skilled trades. This diversity highlights the varied socioeconomic and professional experiences represented in the sample. They were divided into 3 groups: healthy older adults aged older than 65 years without a neurological diagnosis (n=194, 67.8%, mean 72.38, SD 5.55 y; 119 women and 75 men); adults aged older than 65 years with early-stage or mild dementia or MCI (n=87, 30.4%, mean 74.1, SD 5.62 y; 42 women, 44 men, and 1 who preferred not to say); and caregivers of adults aged older than 65 years with early-stage or mild dementia or MCI, who responded on behalf of the person they cared for, providing their demographics (not their own; n=5, 1.7%, mean 70.4, SD 8.47 y; 1 woman and 4 men). Participants with cognitive impairment included 31 (33.7%) with Alzheimer disease, 44 (47.8%) with MCI, 8 (8.7%) with vascular dementia, 3 (3.3%) with frontotemporal dementia, 2 (2.2%) with Lewy body dementia, and 4 (4.3%) with other or unspecified conditions. Due to the small size of the caregiver group, these responses were combined with those of the dementia and MCI group. Sensitivity analyses excluding caregiver responses showed no meaningful changes in results, justifying their inclusion in the combined cognitive impairment group.

This study was designed as a mixed methods cross-sectional survey using independent samples. The online survey was created and conducted on Qualtrics (Qualtrics International Inc), with questions focusing on respondents’ views of online search systems, such as Google, and Gen-AI systems, particularly conversational Gen-AI, such as ChatGPT.

The survey instrument was developed collaboratively by the first author and 2 supervisors to explore awareness, experience, and perceptions of Gen-AI and online search systems in older adults with and without cognitive impairment (including MCI and early-stage dementia). As no existing validated instruments addressed perceptions of conversational Gen-AI in older adults with cognitive impairment, items were created de novo, and established frameworks such as the Technology Acceptance Model (TAM) were not used because they did not capture the specific nuances of these systems in this population. The survey included a combination of Likert-scale and open-ended questions covering prior use, interaction and question phrasing, perceived response relevance, emotional experience, perceived usefulness, and perceived ability for independent use.

For Gen-AI questions, participants were first asked whether they used this technology. Those who said yes were asked about the applications they use, how often, for how long, and through which devices. They responded to questions on a 5-point Likert scale regarding the ease or difficulty of using the applications and the reasons behind their experiences, their usefulness, ease of selecting and writing queries, the relevance of the output, and their experience with the system in terms of boredom, enjoyment, independence, complexity, engagement, and feeling overwhelmed. Those who said no to using Gen-AI tools were asked about their reasons for not using them, what might encourage them to use these tools in the future, and whether they would consider using them in the future. Most Likert-scale questions were rated between “strongly disagree” and “strongly agree.” These responses were encoded as integers from 1 through 5 and averaged across groups.

Similarly, participants were asked if they used online search systems. Those who answered yes were also asked about which applications they used (specific to general web search engines), how often, for how long, and through which devices. They were given Likert-scale questions about how easy or difficult the applications were to use and why and questions regarding the overall search process. They were asked about query formulation in terms of phrasing and rephrasing questions and relevance assessment regarding whether the search results were relevant, whether enough relevant information was provided, whether they could decide what information was relevant, and whether results were always relevant to their question. They were also questioned about helpful features, any aspects needing improvement, and their experiences with the system, including memory of their search goals, feelings of being lost, engagement, enjoyment, boredom, independence, complexity, and feeling overwhelmed. Those who said no to using online search systems were asked about the reasons for not using them, what might encourage them to use these systems, and whether they would consider using them in the future. All participants were asked at the end of the survey what they would like to see in a system they could design themselves and given an opportunity to add any further comments they wished to voice.

Five healthy older adults with no cognitive impairments were recruited through the University of Strathclyde Older Adult Research Panel. The pilot sessions were conducted both in person and online. Participants completed the survey with the researcher, providing feedback on the questions, structure, and length of the survey.

Following pilot feedback, minor refinements were made: illustrative images introducing Gen-AI and online search systems were added; it was clarified that “generative AI” referred to conversational, text-based chatbots rather than voice assistants; clarification that questions referred to the search system interface rather than website credibility was included; confusing or redundant response options were removed; and it was more clearly indicated where multiple responses could be selected. Participants reported the survey length and structure to be appropriate, and no major restructuring was required. As the study was exploratory, no formal psychometric validation or reliability testing was conducted.

Participants first selected the group they identified with, then filled out the relevant participant information sheet and consent form. If they were interested in joining the focus groups, they had the option to provide their email address for future contact. They received an ID and proceeded to the main questions in the survey, completing the Generative AI section followed by the Online Search System section. A debrief was then displayed. After the survey recruitment, invitations were sent to those who had expressed willingness to participate in the focus groups. Survey completion times were highly skewed due to a small number of very long durations. The median completion time was 825 (IQR 597.5‐1223) seconds.

A series of 2-tailed independent-samples t tests were carried out to examine each Likert-scale response between the MCI or dementia and healthy older adult groups. Although Likert-scale data are technically ordinal, independent-samples t tests can be considered robust to violations of the interval assumption, particularly for detecting group differences [44]. To account for the small and unequal sample sizes in the Gen-AI comparisons, we conducted supplementary nonparametric analyses using Mann-Whitney U tests, which produced results consistent with the t tests. Therefore, the reported t test results provide a valid representation of group differences.

Ten survey participants took part in the follow-up focus groups (5 men, 5 women; mean age 73.9, SD 5.67 y). Two (20%) of 10 participants had indicated in the survey that they had used Gen-AI before, and all participants had said yes to having used online search systems. Six (60%) participants identified as healthy older adults, while 4 (40%) had a diagnosis of Alzheimer disease, dysexecutive syndrome due to neurodegeneration, or MCI.

Three focus groups were conducted in total. Group composition was determined based on cognitive status and prior Gen-AI experience: (1) healthy older adults with no prior Gen-AI experience (n=4, 40%), (2) adults with MCI or dementia with no prior Gen-AI experience (n=2, 20%), and (3) a mixed group including both healthy older adults and adults with MCI or dementia, with varying prior Gen-AI experience (n=4, 40%). While group 2 had only 2 participants due to recruitment constraints, their perspectives were included to ensure representation of individuals with cognitive impairment. The characteristics of each group are summarized in Table 2.

Question topics were informed by initial survey responses, establishing exploration of early patterns, such as trust, which was not questioned in the survey, yet was raised in responses. A semistructured interview guide was used to ensure consistent topic coverage across groups while allowing flexibility for probing and follow-up questions. The full interview guide is provided in Multimedia Appendix 1. To minimize moderator bias, neutral questioning was used, participation from all group members was encouraged, and both positive and negative perspectives were explored. The presence of a second researcher allowed oversight of adherence to the interview guide and provided a reflexive discussion following each session. All sessions were audio recorded and transcribed verbatim, and data collection continued until thematic saturation was reached, ensuring rigor and reliability of qualitative findings. The focus groups were moderated by the lead researcher, with a senior supervising researcher present during each session.

Selected participants were contacted by email for the focus group follow-up. Participants read and completed the participant information sheet and consent form before attending the meeting, and focus groups were conducted online via Zoom (Zoom Communications). Each session lasted approximately 80 minutes, including a short mid-session break. Introductions were made, followed by a short demonstration of Gen-AI, particularly ChatGPT (OpenAI), illustrating the basic functions and capabilities of the system. This demonstration lasted approximately 5 to 10 minutes and followed a scripted sequence of predefined prompts to ensure consistency across groups (Multimedia Appendix 1). The main discussion then centered around Gen-AI, covering overall opinions, experience asking questions, its responses to questions, relevance, trust, and training. The same questions were also posed for online search systems. A brief debrief and a question-and-answer session were provided before closing.

To analyze the qualitative data from both the open-ended survey questions and focus group responses, a reflexive thematic analysis was performed [45]. Themes were iteratively developed and refined across both data sources, allowing the qualitative findings to contextualize and deepen the understanding of the quantitative survey results.

When asked whether participants currently use a Gen-AI tool, 40 (14%) of 286 participants responded “yes,” comprising 27 (13.9%) of 194 older adults and 13 (14.1%) of 92 of those with MCI or dementia, with the most common use being to find information (Table 3). Many participants had heard of this technology, but the reasons given for not having used it varied (Table 4). Despite most participants saying “no” to using Gen-AI, 169 (68.7%) of 246 participants indicated they would consider using Gen-AI tools in the future, highlighting the potential for future adoption. Among those who used these tools, usage was mostly less than monthly, followed by weekly, lasting from a few minutes up to 10 minutes, with ChatGPT being the most frequently used. Thirty-three (94.2%) of 35 Gen-AI users felt that Gen-AI tools could be helpful in their lives. When asked to rate the ease of searching for information using a Gen-AI tool, where 1 was extremely difficult and 5 was extremely easy, the mean rating was 4.15 (SD 0.96).

Independent-samples t tests revealed no significant differences between the older adult and MCI or dementia groups, except for independent use of the system, where older adults responded in greater agreement to being able to use Gen-AI independently compared to those with MCI or dementia (Table 5). Nonetheless, there was a noticeable trend of older adults rating items, such as independent use and the ease of use of Gen-AI applications, as more agreeable than the MCI or dementia group. Further investigation through qualitative responses provided additional insights.

When survey participants were asked why they did not use a Gen-AI tool, responses reflected a general lack of knowledge, such as “I don’t really understand them,” not knowing enough about their potential, such as “don’t really know enough about it, its benefits or how to access and use it,” and not understanding how to operate it, such as “don’t know how to use it.” This also extended to the belief that it is solely for education; one survey participant said:

These qualitative responses align with quantitative trends showing that awareness alone does not guarantee confident adoption. For example, the mean ratings for knowing how to phrase their questions were 3.23 (SD 1.3; MCI or dementia) and 3.86 (SD 1.04; healthy older adults), and for understanding when to use Gen-AI, the mean ratings were 3.67 (SD 1.23) and 4.14 (SD 0.89), respectively (Multimedia Appendix 2). This suggests that limited familiarity may restrict effective use, particularly for the MCI or dementia group. Focus groups provided similar insights, with participant 6 (male, aged 69 y, no impairment) stating, “...I didn’t know what it was and how to use it,” while also expressing interest in using it after hearing about its capabilities.

A minor theme identified in the survey data was the lack of perceived need to use this technology, as reflected by both the MCI or dementia group, “not felt the need,” and in the older adult group, "I have no need to generate tracks of commentary or explanation... I feel confident in my ability to read, assess, and formulate written responses.” Quantitative data support this: although 68.7% of participants indicated they would consider using Gen-AI tools in the future, current use was only 14%. Older adult survey responses offered more detail, with comments such as “have no need to and don’t know enough about it’ and “I’ve not felt a need to use one and I’m not sure how to get started.” These insights suggest that perceived lack of need may reflect limited understanding of Gen-AI applications.

A further theme was mistrust. Survey participants who had never used Gen-AI expressed security concerns (“don’t trust it. Think it can be used/hacked for nefarious and cruel purposes”) and concerns about accuracy (“probably foolishly, I don’t feel I can fully trust AI-generated tools or the information they provide”). This aligns with quantitative findings showing lower confidence in independent Gen-AI use (MCI or dementia: mean 4.00, SD 1.15 vs healthy older adults: mean 4.68, SD 0.65; P=.03), suggesting that it may limit use. This may relate to limited knowledge about data management.

Survey participants also raised concerns about regulation: “I believe they are dangerous...There is insufficient and inadequate regulation.” This aligns with responses from those who had used Gen-AI but stopped or disliked it; when asked what caused this, survey replies included accuracy concerns, (“suspicious of the accuracy of results”), privacy and security (“I need to be absolutely certain that it’s secure”) and regulation (“better monitoring of these tools by independent third parties”). Quantitative responses showed that although participants generally agreed Gen-AI provides relevant answers (MCI or dementia: mean 3.69, SD 1.32 and healthy older adults: mean 4.09, SD 0.81), lower confidence in phrasing questions (mean 3.23, SD 1.3 vs mean 3.86, SD 1.04) and understanding when to use tools (mean 3.67, SD 1.12, vs mean 4.14, SD 0.89) reflects that mistrust may affect perceived usefulness (Multimedia Appendix 2).

Focus groups discussed trust in more depth. Participants said, “…at the moment I don’t trust artificial intelligence,” and “…I actually trust very little of anything.” Focus group participants also raised privacy concerns:

Regulation was also highlighted: “what is being done with social media and how it’s abused…we allowed that because nothing was regulated,” emphasizing transparency. Participant 3 (female, aged 78 y, no impairment) expressed feeling a lack of control:

Survey responses mirrored this cautious use: "if I needed to use AI I would, but I prefer not to increase use of AI until I am sure it is controllable.” This mistrust in AI is reflected through sci-fi references by participant 9 (male, aged 74 y, Alzheimer disease), “...these extraordinary machines are not human and all our concerns about sci-fi scenarios in which these machines take over the world,” which again mirrors survey responses, “I think they could be dangerous...watch the Terminator films again. If the machines ever become self aware humanity will be in real trouble.”

Focus groups also highlighted that trusting Gen-AI depends on the user verifying results. Participant 7 (male, aged 82 y, MCI) said:

They stressed that their own knowledge is essential:

Healthy older adults echoed this:

Participant 1 (male, aged 76 y, Alzheimer disease) reinforced this:

These observations help explain why participants reported relevant outputs (MCI or dementia: mean 4.17, SD 0.83), yet lower confidence in independent use, suggesting that mistrust and cognitive concerns influence engagement. Therefore, reliance on the user to identify correct and unbiased information is a significant factor. This was echoed by a participant with MCI or dementia in the survey data, who stated:

Thus, mistrust and need for oversight appear linked to lower confidence in independent use shown in the quantitative data, highlighting the importance of distinguishing genuine from AI-generated content.

A distinct theme was participants’ perceived capability when interacting with Gen-AI, reflecting situational self-efficacy. Some focus group participants reported feeling the need to treat AI as more capable than themselves. For example, participant 7 (male, aged 82 y, MCI) said:

Conversely, focus group participant 8 (female, aged 71 y, no impairment), who uses Gen-AI, offered a contrasting viewpoint, “...I believe I treat it as a machine, as a supercomputer.” These perspectives contextualize survey responses reporting discomfort with interactions, “the robotic nature of the exchanges and limitations to interact are uncomfortable to deal with,” and may help explain variance in engagement and trust metrics across participants. Although perceived capability was not directly measured, these qualitative insights suggest that participants’ situational self-efficacy in using Gen-AI varies depending on how they perceive the AI’s abilities relative to their own.

When asked about what would encourage participants to use Gen-AI in the future, survey responses reflected training needs and a general willingness to learn. The MCI or dementia survey group often requested “more knowledge, experience, and training” and “an explanation and teaching of how to do this.” Similar responses were given by the older adult survey group, expressing a desire for information on “how it would benefit me followed by training on how to use it,” including “a demonstration and practice in using them” and access to “free training sessions possibly at a library.” Both survey groups also mentioned the importance of support from others to encourage use, such as “having someone around with the time to explain it and then help me to use it” in the MCI or dementia group, and “recommendation by a friend or family member...” in the healthy older adult group. Quantitative data support this: 68.7% (169/246) of survey respondents indicated they would consider using Gen-AI in the future. Additionally, Likert-scale responses showed moderate agreement that Gen-AI helps answer questions (MCI or dementia: mean 3.77, SD 1.01; and healthy older adults: mean 4.18, SD 0.73) and provides relevant answers (MCI or dementia: mean 4.17, SD 0.83; and healthy older adults: mean 4.18, SD 0.73). This suggests that addressing knowledge gaps and providing transparency could facilitate adoption.

Survey participants also outlined their expectations for training, including desires for “a clear idea of what it can do and be used for,” “more information and guidance in how to use and its benefits,” and practical examples. Comments also reflected concerns about trust and system performance. The MCI or dementia group mentioned, “only if I could be absolutely assured that what was produced was true and could be collated with other trusted sources of information,” while the older adult group expressed concerns about accuracy, stating, “being reassured that they don’t make up the answer if they can’t find it.” These concerns parallel survey Likert items on confidence in independent use, where lower ratings among participants with MCI or dementia (mean 4, SD 1.15 vs 4.68, SD 0.65 in healthy older adults) may reflect caution about trusting outputs and the need for training to build confidence while addressing reliability and accuracy.

Despite encouraging responses, it is important to acknowledge that some reactions showed reluctance to engage. Some participants replied “nothing” when asked what would encourage them to use these systems, while others, such as from the MCI or dementia group, expressed, “I personally don’t feel I need anything like this, at the moment,” and “Nothing. I do not trust the security of these sites.” Older adults also shared perceptions that there is too much technology, with comments such as “I don’t want to get involved with anything else,” and believed that web search was sufficient, stating “I don’t think I need to use such a tool, as I can find most information I need on Google.” Willingness to use Gen-AI appears shaped by personal beliefs and experiences, with one noting, “I believe this tool is used to help create content/write something...I have never felt the need to use it.” Some responses from the MCI or dementia group cited complexity, “it’s too complicated,” “I’m not very good with tech stuff,” “coping with technology,” and “I am a technophobe.” These hesitations correspond with quantitative patterns, such as lower confidence in phrasing questions (MCI or dementia: mean 3.23, SD 1.3 vs healthy older adults: mean 3.86, SD 1.04) and understanding when to use Gen-AI (3.67, SD 1.23 vs 4.14, SD 0.89). This indicates that not all barriers can be addressed through training alone.

Both groups also noted ease of use and user-friendliness: “usefulness outweighing frustration.” When asked about ease of use, those who used Gen-AI responded positively. The MCI or dementia group said, “...it provides a response that I understand,” “it is very responsive to questions posed to it. I often put in further questions...,” and “I am just starting with AI as I feel it could be of great help as my condition progresses.” Older adults responded similarly, “the info is quite detailed but is easy to query in more detail by using more targeted questions,” “no harder than searching in Google,” and “a bit better and precise than an internet search.” However, some negative points were raised, such as, “with some questions the AI has to be tricked into answering, or will not answer,” and concerns about complexity: “I have limited tech” experience and, as I have no family or support network, I can be slow to learn & retain new methods.” Overall, these experiences align with Likert-scale ease of use scores (mean 4.15 overall).

The focus group participants expressed varied interest in training. Participant 2 (male, aged 67 y, dysexecutive syndrome) said, “I would go and get that training because I’m curious...the problem is being able to remember... what you’ve learned,” highlighting that training might be most effective early in diagnosis or beforehand. Participant 1 (male, aged 76 y, Alzheimer disease) expressed a preference for being taught by similar individuals, stating:

Others, who were more skeptical about AI, did not show interest in training, “I just think it’s money wasted on me at 80... better given the training to the youngsters so they’re prepared for the future rather than me,” or simply conveying a lack of interest in attending. Participant 10 (female, aged 79 y, no impairment) expressed a preference to explore it in their own time, stating:

Additionally, it was discussed that targeted training for groups could be most beneficial:

Despite varied interest, some participants found the focus group demonstration intriguing, saying:

Overall, responses highlight that individual experiences, cognitive abilities, and confidence shape willingness to learn, but there is potential for training to help those interested.

Participants were asked if they currently use an online search system. Of the 286 survey participants, 280 (97.9%) responded to this question, reflecting partial survey completion. Among respondents, 275 (98.2%) of 280 participants responded “yes,” with 186 (98.9%) of 188 older adults and 89 (96.7%) of 92 participants in the MCI or dementia group reporting use. Among these users, Google was the most used, and most participants reported using search systems daily and for less than 5 minutes at a time. Participants used these tools primarily to find information and learn (Table 3). When asked how easy they found searching for information using an online search system, where 1 was extremely difficult and 5 was extremely easy, the average response was 4.38 (SD 0.8), indicating generally positive perceptions.

Independent-samples t test results reveal significant differences between groups on several survey questions, with the healthy older adult group consistently rating items as more agreeable (Table 6) than the MCI or dementia group. For example, the MCI or dementia group was more likely to report needing to reword queries, feeling lost, or feeling overwhelmed, whereas healthy older adults more often reported being able to remember what they were searching for, deciding whether information was relevant, and using search systems independently. Survey open-ended questions and focus groups provided additional insight.