Large Language Models in the UK: Public Use, Trust, and Attitudes

1.1 PUBLIC ADOPTION OF LLMS

• UK LLM use rose from 40% in 2025 to 58% in 2026, with 35% reporting at least weekly use [2, 3].Work, study, and information-seeking are common activities, while reports also note growing use for personal advice and emotional support [7, 8].Source: §1.1 · p. 5

1.4 RESEARCH CONTRIBUTION

• A survey of 2,002 UK adults examines LLM use, trust, public attitudes, and demographic differences, addressing a gap in population-level evidence about engagement.The introduction identifies limited evidence on the prevalence of social and emotional uses relative to practical use, and on the characteristics associated with different forms of interaction.Source: §1.1 · p. 5, §1.4 · p. 7
• The study compares five LLM activity types, from text generation and project planning to emotional support, companionship, and roleplay, and assesses perceived humanness and gender.It spans functional and social engagement, including uses that may simulate human relationships and social connection.Source: §1.4 · p. 7

1.2 PUBLIC TRUST IN LLMS

• The study compares trust in LLM information with experts, search engines, newspapers, and social media across knowledge domains, including politics, science, and health.The introduction notes that LLMs can provide accessible, personalized information but may also produce biased, misleading, or inaccurate outputs; the study examines demographic and social correlates of trust.Source: §1.2 · p. 6, §1.4 · p. 7

1.3 PUBLIC ATTITUDES TOWARDS LLMS

• The report assesses public perceptions of LLM benefits and risks and overall sentiment, alongside social and demographic characteristics associated with these views.Prior UK survey findings described in the introduction found 63% saw LLMs as very or fairly beneficial and 47% expressed concern [2].Source: §1.3 · pp. 6–7, §1.4 · p. 7

1.4 RESEARCH CONTRIBUTION

• The report examines demographic characteristics associated with LLM use, trust, and sentiment, including differences that may shape adoption, access to benefits, or exposure to adverse effects.The introduction notes prior evidence that women are often more cautious in digital environments and more skeptical of AI [24, 25], potentially in light of disproportionate exposure to technology-facilitated harms [26].Source: §1.4 · p. 7, §1.3 · p. 7
• Because public engagement with LLMs is new and evolving, the findings provide a current baseline for assessing future changes in use, trust, and public sentiment.The introduction links this changing context to expanding use and integration into workplaces and existing digital systems.Source: §1.4 · pp. 7–8

2.1 DATA COLLECTION AND AVAILABILITY; 2.2 SAMPLE CHARACTERISTICS

• A December 2025 online Qualtrics survey recruited through Prolific yielded 2,002 quality-checked participants selected for UK representativeness across age, gender, and region.Informed consent was obtained, and the full survey and anonymised data are available on OSF.Source: §2.2 · p. 9, §2.1 · p. 9

2.3.1 DEMOGRAPHICS AND BACKGROUND QUESTIONS

• Researchers collected age, gender, ethnicity, education, employment, political orientation, and digital-literacy measures.Digital skills were self-rated on a four-point scale, while political orientation used a 0–100 scale from extremely left to extremely right.Source: §2.3.1 · pp. 9–10

2.3.2 LLM USE

• Participants received an AI-chatbot description and reported prior chatbot and companion use, chatbot and companion frequency, and duration of chatbot use.Prior use, frequency, and duration were assessed with response options ranging from yes/no/unsure or never to daily and less than a month to more than a year.Source: §2.3.2 · p. 10

2.3.3 TYPES OF LLM INTERACTION

• Respondents rated five interaction purposes on a four-point scale, spanning practical tool use through role-play and character-based interactions.Other purposes included collaboration and planning, personal and emotional support, and deep or meaningful conversation.Source: §2.3.3 · p. 10

2.3.4 MODES OF LLM USE; 2.3.5 GENDER AND HUMANNESS PERCEPTIONS OF LLMS

• Questions covered text or voice replies, voice type, tone-setting adjustments, perceived chatbot gender, and how human-like interactions felt.Human-likeness was rated on a four-point scale from not at all to completely.Source: §2.3.5 · p. 11

2.3.6 TRUST IN LLM-GENERATED INFORMATION

• Participants rated trust in six sources, including AI chatbots, across climate change, elections, health, legal matters, personal relationships, and public services.Trust used a four-point scale from not at all to very much.Source: §2.3.6 · p. 11

2.3.7 ATTITUDES TO LLMS

• Attitudes were measured through four-point enthusiasm-for-benefits and worry-about-risks items, plus a 0–100 harms-to-benefits slider.The slider ranged from LLMs creating more harms than benefits to creating more benefits than harms.Source: §2.3.7 · p. 11

2.4 SURVEY PROCEDURE; 2.5 ANALYTICAL APPROACH

• All respondents answered trust and attitude items, while use questions depended on prior experience; analyses described use and modeled predictors among weekly users or the full sample.Trust respondents were randomly assigned two of the six information domains; trust and overall-attitude models used the full sample.Source: §2.4 · p. 11, §2.5 · pp. 11–12

3.1 USE OF LLMS

• 92.7% had used an LLM at least once; among prior users, 32.9% used LLMs several times a week, 25.7% daily, and 14.8% about weekly.Most respondents who had used LLMs did so at least several times a week.Source: §3.1 · p. 13
• 55.5% of regular LLM users had started using LLMs more than a year earlier, while 27.3% had used them for about 6–12 months.More recent uptake was less common: 15.7% reported 1–6 months of use and 1.5% said they had only recently started.Source: §3.1 · p. 13

3.2 USE OF AI COMPANIONS

• 20.2% of the sample had tried an AI companion; among prior users, 31.1% had used one only once or twice, while 20.8% used one several times a week and 7.6% daily.Another 15.6% of prior users reported using AI companions about once a week.Source: §3.2 · p. 13

C. AI Companion use frequency

• Figure 1 presents prior use among all respondents and use frequency among respondents with prior LLM or AI companion use.The supplied caption gives no specific prevalence or frequency values.Source: Figure 1 · p. 14

3.4 GENDER AND HUMANNESS PERCEPTIONS OF LLMS

• Among regular users, 44.9% described LLM interactions as slightly human-like and 36.2% as moderately human-like.Most regular users mainly used text replies (86.6%) and did not change default tone settings (74.8%); perceptions of LLM gender were most often ungendered (46.6%).Source: §3.3 · pp. 14–15, §3.4 · p. 15

3.5 TYPES OF LLM USE

• 64.1% of regular users used LLMs for practical tasks very often, while 36.2% did so for collaboration and planning.Occasional use was reported by 27.2% for practical tasks and 38.9% for collaboration and planning; personal advice, meaningful conversation, and role-play or companionship were less common, at 31%, 24.9%, and 14% respectively, including very often or occasional use.Source: §3.5 · p. 15

3.5.1 DEMOGRAPHIC PREDICTORS OF TYPES OF LLM USE

• Age was the most consistent predictor of interaction type, with each additional decade associated with 16–26% lower odds of use.Degree holders had higher odds of practical use (OR = 1.86) and collaboration or planning (OR = 1.45); men had higher odds of practical use and role-play, but lower odds of personal support. Younger respondents were more likely to report deeper conversation, and men and more right-wing users were more likely to report role-play.Source: §3.5.1 · pp. 16–17

3.6.1 SOCIAL AND DEMOGRAPHIC PREDICTORS OF TRUST IN LLMS

• Regular users were 3.27–4.54 times more likely than non-regular users to trust LLM information across all six domains.Older respondents showed 12–14% higher odds of trust per decade for elections, legal matters, and health; right-leaning respondents had around 11–12% higher odds for elections, legal matters, and relationships.Source: §3.6.1 · p. 18

3.6 TRUST IN LLM-GENERATED INFORMATION

• 63.0–70.8% of respondents trusted LLMs at least somewhat on elections, public services, climate change, legal matters, and health, but 37.6% did so for relationship advice.Trust in LLMs was higher than for social media, newspapers, and television, but generally lower than for online search, experts, and friends and family; climate change was the exception for friends and family.Source: §3.6 · p. 18, Figure 3 · p. 19

3.7 ATTITUDES TO LLMS

• 77.3% agreed they felt enthusiastic about chatbot benefits, while 69.6% agreed they felt worried about chatbot risks.On the 0–100 benefits-versus-harms scale, attitudes were slightly more positive than negative (mean = 56.7; median = 60).Source: §3.7 · pp. 19–20

3.7.1 SOCIAL AND DEMOGRAPHIC PREDICTORS OF ATTITUDES TO LLMS

• Regular users scored 18.90 points higher than non-regular users on the attitudes scale, and were more likely to report enthusiasm for benefits (OR = 6.82).Regular users were less likely to report concern about risks (OR = 0.35); older, male, and more right-leaning respondents were also less likely to report concern, while the same groups were associated with more positive attitudes or benefit enthusiasm.Source: §3.7.1 · p. 21

4.1 PUBLIC ADOPTION OF LLMS

• 31% of regular users reported turning to LLMs for personal or emotional support, while practical uses remained more common.A quarter reported meaningful conversations, 14% roleplay, and more than eight in ten described interactions as at least slightly human-like. Use declined with age across all five use types; men were more likely than women to use LLMs for practical tasks and roleplay, but less likely to use them for personal support. The findings point to a need to study when social engagement complements or displaces human and professional support.Source: §4.1 · p. 22

4.2 PUBLIC TRUST IN LLMS

• 63%–70% of respondents trusted LLM information at least somewhat across elections, public services, climate, legal, and health topics; trust in relationship advice was 38%.LLMs were generally trusted less than experts, online search, and friends and family—except for climate change—but more than newspapers and television, and substantially more than social media. Regular use was the strongest predictor of trust; because LLMs can produce inaccurate or misleading outputs, users should be aware of their limitations and exercise caution.Source: §4.2 · p. 23

4.3 PUBLIC ATTITUDES TOWARDS LLMS

• 77.3% of respondents felt enthusiastic about LLMs’ potential benefits, while 69.6% were concerned about potential risks.After LLM use itself, gender was the strongest and most consistent predictor of attitudes: men were more enthusiastic, less concerned, and more likely to expect benefits to outweigh harms. The findings underscore the importance of inclusion and further research into public expectations for AI governance and regulation.Source: §4.3 · pp. 23–24

4.4 STUDY LIMITATIONS

• The non-probability Prolific sample may overrepresent digitally engaged adults, so overall LLM-use estimates may not represent the UK population.The study included adults only, leaving children’s and teenagers’ use, trust, and developmental outcomes for future research. Because adoption and public discourse are changing, the findings capture one moment and require regular monitoring.Source: §4.4 · pp. 24–25

• Practical tasks remain the most common LLM use, alongside substantial emotional support, meaningful conversation, and companionship; trust is relatively high, and attitudes combine optimism with concern.Source: §5 · p. 26
• The findings offer a UK baseline for tracking changes in LLM use, trust, and public sentiment.Source: §5 · p. 26
• Further research should examine long-term effects on learning, cognition, and social relationships, and how trust shapes reliance on and action upon LLM information and advice.Source: §5 · p. 26
• Demographic differences by age, gender, and educational attainment in use, trust, and sentiment underscore the need to make LLMs accessible and beneficial to all.Source: §5 · p. 26

A1 DESCRIPTION OF LLMS GIVEN TO PARTICIPANTS

• Participants were told that LLMs are AI programs generating text, speech, or images in response to prompts, and were asked separately about character-based AI companions.The description excluded voice assistants such as Alexa and Siri; companions were defined as chatbots designed for emotional support, friendship, and conversation.Source: A1 DESCRIPTION OF LLMS GIVEN TO… · p. 31

A2 TABLES SHOWING FULL REGRESSION RESULTS

• Table 1 reports demographic predictors of types of LLM interaction among regular users, while Tables 2 and 3 report demographic regression results for trust in LLM-provided information and overall attitudes toward LLMs.The trust models span six information domains.Source: Table 1 · p. 32, Table 2 · p. 33, Table 3 · p. 34
• Age’s odds ratios were 0.78 (0.68–0.89), 0.82 (0.76–0.90), 0.74 (0.68–0.80), 0.84 (0.77–0.92), and 0.77 (0.69–0.86) (all p < .001) across five regular-user interaction models.Age was scaled per 10-year increase; each model compared reporting the interaction at least occasionally with very rarely or never.Source: A2 TABLES SHOWING FULL REGRESSION… · p. 32, Table 1 · p. 32