Language models have moved quickly from novelty to routine tool for many U.S. adults. A new study from the Imagining the Digital Future Center describes a technology that is being used across demographic lines, often for everyday learning, planning, voice conversation, image generation, and important life choices.
The most striking finding is about perception. Nearly half of users, 49%, believe language models are more intelligent than they are, even as many also report errors, confusion, dependence, or discomfort with how these systems shape their choices.
LLM use is no longer limited to early adopters
The study describes language models as having reached historic adoption rates among U.S. adults. It also says the tools have moved past some traditional adoption barriers, with women using them at the same rate as men, regardless of education or income levels.
ChatGPT is the leading platform among users in the study, with 72% choosing OpenAI's tool. Google's Gemini follows at 50%, Microsoft's Copilot reaches 39%, Meta's LLaMa is used by 20%, xAI's Grok by 12%, and Anthropic's Claude by 9%.
Those figures also show that many people are not treating LLMs as a single-product habit. The study says 58% of users have tried two or more language models, suggesting that comparison and experimentation are already common parts of adoption.
Paid access is less common. While 20% of users access paid versions, only 4% personally pay for those subscriptions. That gap matters because it suggests premium AI use may depend on workplace access, shared accounts, or other arrangements rather than direct consumer spending.
People use LLMs more for personal learning than work
The research finds that personal use is ahead of professional use. According to the study, 51% of users mainly turn to language models for personal learning and planning, while 24% mainly use them for work-related tasks.
That distinction helps explain why language models are becoming part of ordinary decision-making rather than remaining confined to offices or technical teams. A person may use an LLM to organize a plan, understand a topic, compare options, or think through a next step without seeing the interaction as a formal work task.
Professional adoption still varies by occupation. Managers, scientists, and licensed professionals like teachers report high usage rates. Retail and service workers also show high engagement at 62%, which challenges the assumption that these tools mainly serve desk-based or highly technical roles.
Household context appears to matter as well. The study says households with children under 18 are more likely to use language models, at 61%, compared with 50% for households without children under 18.
AI image tools and voice chat are becoming part of the same habit
LLM adoption is closely linked with AI image generation. Two-thirds of language model users, 67%, have tried image tools such as DALL-E, Midjourney, Adobe Firefly, or ImageFX.
For some users, image generation is not occasional. The study says 18% generate AI images daily, while 12% do so multiple times per week. This points to a broader AI workflow in which text systems and image systems are used together, rather than as separate categories.
Voice interaction is another major behavior. The study reports that 65% of users have spoken conversations with AI systems that answer in realistic voices. That practice is especially common among lower-income households, at 76%, and non-white users, at 83%.
Voice changes the experience of using a language model. Instead of typing into a box, a user can ask questions aloud and receive a spoken answer, which can make the system feel more immediate and conversational. The source does not explain why voice use is higher in those groups, but the pattern is clear in the study's findings.
High satisfaction comes with real friction
The study presents a mixed picture of trust and risk. On one hand, 76% of users say they are satisfied with language models. On the other, 23% report making serious mistakes because of incorrect LLM information, and 21% say they feel manipulated by these systems.
Concerns extend beyond factual errors. About half of users say LLMs make tasks too easy or encourage laziness. More than a third feel too dependent on the technology, and the same proportion describe their LLM use as a form of cheating or say the interactions leave them confused.
These findings matter because users are not only asking harmless or low-stakes questions. The study says 41% consult LLMs for career development decisions, 37% use them for health-related questions, and 28% look to them for help with changing jobs.
Financial and housing decisions are also part of the pattern. The study reports that 25% of users rely on language models for important money matters, while 18% turn to these tools when deciding where to live.
Seeing LLMs as smarter changes the stakes
The perception of intelligence may be the most important thread running through the study. Nearly half of users, 49%, believe language models exceed their own intelligence. Among female users, 30% rate LLMs as "significantly more intelligent" than themselves, compared with 20% of men.
That belief can shape how people interpret answers. If a system is seen as intellectually superior, a user may be more likely to defer to it, especially when the topic is complex, personal, or emotionally charged. The study's findings on mistakes, manipulation, dependence, and major life decisions make that perception difficult to ignore.
The research was conducted by SSRS through its Opinion Panel platform from January 21-23, 2025. It included 500 adults with language model experience, with 498 online interviews and 2 by telephone; 473 interviews were in English and 27 in Spanish. Researchers also weighted responses from 939 screened panelists to determine the overall incidence of LLM use among U.S. adults.
The result is a snapshot of a technology that many people already find useful, flexible, and satisfying. It is also a reminder that adoption is moving faster than the public's comfort with accuracy, trust, and dependence.