Conversational AI as a catalyst for informal learning: An empirical large-scale study on LLM use in everyday learning

Synthesis: Conversational AI as a catalyst for informal learning: An empirical large-scale study on Large Language Models (LLMs) use in everyday learning

Key Findings

• In a large-scale survey of 776 German participants conducted in February 2025 — the third year following the breakthrough launch of ChatGPT — 87% of respondents reported already incorporating LLMs into their everyday learning routines for a wide variety of learning tasks (11% used LLMs but not for learning, and 2% did not use them at all).
• Young adults among German-based, digitally engaged users are at the forefront of adopting LLMs, primarily to enhance their learning experiences independently of time and space: LLM learners averaged 31 years versus 35 for non-learners (F = 6.68, p < .01), men were more likely to adopt (53% of LLM learners were male, while 64% of avoiders were female; χ² = 15.5, p < .001), and ChatGPT dominated, with 93% (n = 632) of LLM learners reporting its use.
• Four types of learners emerge across learning contexts, distinguished by the tasks they perform with LLMs and the devices they use to access them: Structured Knowledge Builders (22.9%), Self-Guided Explorers (26.5%), Analytical Problem Solvers, and Adaptive Power Users, identified via latent class analysis with balanced class shares (23.8–25.7%).
• Respondents exhibit paradoxical trust behaviors: they rely on LLMs while simultaneously distrusting their accuracy and privacy protection measures — 88% (n = 601) perceived improved learning or productivity, yet misinformation was the most commonly cited challenge (68%, n = 460) and the majority (n = 543) reported taking no privacy measures.
• The study's implications emphasize the importance of including different media types for learning, enabling Collaborative Learning, providing sources, and meeting the needs of different types of learners — "learning by design."

Study Design & Method

The study addresses who is embracing LLMs for self-directed learning, who remains hesitant, their reasons for adoption or avoidance, and the learning patterns that emerge with this novel technological landscape. Data come from an online survey administered via Prolific in February 2025 to 776 German participants (mean age 31.6, SD = 9.87, max 72), including non-adopters; 678 respondents were actively using LLMs for learning. Analysis combined descriptive and inferential statistics (ANOVA, chi-square), a latent class analysis from which the four-learner typology was derived, multinomial logistic regression predicting class membership from privacy perceptions, effectiveness, over-reliance, and demographics, and qualitative analysis of open-text responses. Adoption drivers included curiosity (n = 517), recommendations from social connections (n = 254), and media coverage (n = 217), while avoidance was led by mistrust of factual output and a preference for traditional methods (52% of non-LLM learners).

What this means for practice

• Learners. Notice that your LLM learning is already a routine, not a novelty: the median learner spent about 20 minutes per weekday on it, 63.5% had used LLMs for more than six months, 58% (n = 395) said they were extremely likely to keep using them for learning, and 72% would recommend them to friends.
• Learners. Calibrate trust rather than switching it on or off: reported reliance came alongside distrust of accuracy and privacy, so check outputs and check what a tool does with what you tell it.
• Designers. Support that calibration directly with source provision and transparency features, and build for varied task-device combinations instead of one interface for everyone — the four learner types the study identifies argue against a single design.
• Designers. Consider multimodal and collaborative features for self-regulated, informal use, since so much AI-mediated learning now happens outside formal curricula and the shape of that learning differs from classroom study.

Limitations

• Single-country, self-selected sample. Recruited exclusively from Germany via Prolific, the sample may overrepresent tech-savvy, English-fluent individuals (mean AI literacy score = 3.06), and cultural attitudes toward AI vary, so findings may not generalize to non-Western or less tech-savvy populations.
• Small non-user subsample. The subgroup of non-users was small, so results for that group should be treated as exploratory and descriptive.
• Self-reported use with no outcome measures. The study measured self-reported LLM use and perceived purposes without measuring concrete learning outcomes, so no claims can be made about whether more frequent or highly satisfactory LLM use translates into deeper conceptual learning.

Citation

Terzimehić, N., Bühler, B., & Kasneci, E. (2026). Conversational AI as a catalyst for informal learning: An empirical large-scale study on LLM use in everyday learning.