AI Can Leave Language Learners Feeling Empowered or Ashamed – New Study Reveals Why

Generative artificial intelligence has swept into language classrooms around the world, promising instant feedback, endless conversation practice, and personalized support for students learning English as a foreign language. But a new study published in Current Psychology suggests that the emotional story behind this technological revolution is far more complicated than a simple tale of enhanced learning. Researchers Wentao Liu of Hunan Normal University and Yuehai Xiao of Guangzhou University have found that when Chinese university students use generative AI tools in their English learning, their emotional experiences split into strikingly different patterns depending on how well the technology satisfies three fundamental psychological needs: autonomy, competence, and relatedness.

The research, published on September 3, 2026, tackles a question that has become increasingly urgent as AI chatbots and writing assistants become fixtures in education. While previous studies have established that generative AI tools are associated with learners’ emotional experiences in English as a foreign language, or EFL, learning, the role of basic psychological needs in shaping those emotions remained largely unexplored. Even more importantly, earlier research typically treated learners as a homogeneous group, using variable-centered statistical approaches that average out individual differences. Liu and Xiao took a different path, employing a person-centered technique called latent profile analysis to uncover hidden subgroups of learners within their sample of 346 Chinese university EFL students.

Latent profile analysis is a statistical method that identifies unobserved subpopulations within a larger population by clustering individuals based on their response patterns across multiple variables. Rather than asking how variables relate to each other on average across all participants, this approach asks whether distinct types of people exist within the data. In this study, the researchers measured students’ satisfaction with three basic psychological needs – autonomy, competence, and relatedness – in the context of generative AI-assisted English learning, and then examined how achievement emotions varied across the resulting profiles.

The three needs under investigation come from self-determination theory, a foundational framework in motivational psychology developed by Edward Deci and Richard Ryan. According to this theory, human beings have three innate psychological requirements that must be met for optimal functioning and well-being. Autonomy refers to the feeling that one’s actions are self-chosen and volitional. Competence involves the sense of being effective and capable in one’s interactions with the environment. Relatedness captures the feeling of connection to and caring for others. When these needs are satisfied, people tend to thrive; when they are thwarted, motivation and well-being suffer.

The analysis revealed three distinct profiles of basic psychological needs satisfaction among the 346 learners. The largest group, comprising 79.2 percent of participants, was labeled the satisfaction-moderate profile, characterized by moderate levels of satisfaction across all three needs. A second group, making up 13.9 percent of the sample, showed high levels of both autonomy and relatedness and was therefore named the autonomy-relatedness-high profile. The smallest group, at 6.9 percent, was the autonomy-deficient profile, in which students reported notably low satisfaction with their sense of choice and control over how they used generative AI in their English learning.

When the researchers examined how achievement emotions differed across these profiles, they uncovered a pattern that carries significant implications for educators. Positive emotions followed a clear gradient: students in the autonomy-relatedness-high profile reported the highest levels of enjoyment, hope, and pride in their AI-assisted English learning, followed by those in the satisfaction-moderate profile, with the autonomy-deficient profile reporting the lowest levels. This finding suggests that when learners feel they have genuine control over how they use AI tools and experience a sense of connection with others through those tools, they are far more likely to feel pleasure, optimism, and pride in their language learning journey.

Negative emotions, however, told a more nuanced story. Boredom, anger, and hopelessness were significantly lower in the autonomy-relatedness-high profile than in the other two profiles, but there were no significant differences between the autonomy-deficient and satisfaction-moderate profiles on these emotions. This means that simply having moderate levels of need satisfaction did not protect students from boredom, frustration, or despair any more than having deficient autonomy did. Only the combination of high autonomy and high relatedness appeared to buffer learners against these negative emotional states.

Shame followed yet another pattern, one that the researchers describe as graded. Students in the autonomy-deficient profile reported the highest levels of shame, followed by those in the satisfaction-moderate profile, with the autonomy-relatedness-high group reporting the lowest. Shame, in the context of language learning, often arises when students feel embarrassed about their perceived lack of ability despite having access to powerful tools. The graded pattern suggests that the more satisfied learners are with their basic psychological needs, the less likely they are to experience this particularly painful self-conscious emotion when using AI to learn English.

Interestingly, anxiety did not differ significantly across the three profiles. This finding is notable because anxiety is one of the most extensively studied emotions in foreign language research, often linked to poorer performance and avoidance behaviors. The fact that anxiety levels were similar regardless of need satisfaction profile suggests that anxiety in AI-assisted language learning may be driven by factors other than basic psychological needs – perhaps by the novelty of the technology itself, by concerns about academic evaluation, or by individual differences in trait anxiety that were not captured in this analysis.

The researchers also examined whether learner characteristics predicted membership in the different profiles. Their analysis revealed that generative AI competence – essentially, how skilled students were at using AI tools – was significantly associated with profile membership. Students with higher AI competence were more likely to belong to profiles with greater need satisfaction. In contrast, gender and English proficiency were not significantly related to which profile a student belonged to. This finding challenges the assumption that stronger language learners automatically have better psychological experiences with AI tools, and instead points to digital literacy as the more decisive factor.

The theoretical significance of this work lies partly in its methodological approach. By using latent profile analysis rather than traditional variable-centered methods, the researchers were able to reveal diversity that would otherwise remain hidden. Prior research on basic psychological needs satisfaction in language learning had largely ignored individual differences, treating the population as if all learners experienced their needs in the same way. The identification of three distinct profiles – with the vast majority falling into the moderate category but meaningful minorities showing either high autonomy and relatedness or deficient autonomy – provides a more realistic picture of how students actually experience AI-assisted learning.

The study also connects to a growing body of research on achievement emotions in educational settings. Achievement emotions are emotions tied to academic activities and their outcomes, and they are theorized to influence learning through their effects on attention, motivation, and the use of learning strategies. The control-value theory of achievement emotions, developed by Reinhard Pekrun and colleagues, holds that these emotions arise from appraisals of control and value in academic situations. The current study adds an important dimension to this framework by showing that basic psychological needs satisfaction – a construct from a different theoretical tradition – is systematically associated with the emotional experiences learners report when working with generative AI.

The practical implications of the findings are what the researchers describe as preliminary but potentially significant for differentiated instructional support. Because the profiles differ in their emotional signatures, educators may need to tailor their approaches depending on which group their students fall into. For students in the autonomy-deficient profile, interventions that increase their sense of choice and control over AI use may be particularly important. For students in the satisfaction-moderate group, who still experience significant boredom, anger, and hopelessness despite moderate need satisfaction, additional strategies may be needed to prevent disengagement. And for the autonomy-relatedness-high group, educators might focus on maintaining the conditions that support their positive emotional experiences.

The finding that generative AI competence, rather than English proficiency, predicts profile membership carries a particular message for institutions rolling out AI tools in language programs. It suggests that training students in how to use AI effectively – what some researchers call AI literacy – may be more important for their psychological well-being than their existing language skills. This aligns with broader frameworks such as the European Union’s DigComp 2.2 digital competence framework and UNESCO’s 2023 guidance for generative AI in education and research, both of which emphasize the importance of developing learners’ capacity to engage with AI tools knowledgeably and critically.

The study was conducted with ethical approval from the Ethics Committee of Hunan Normal University, and all participants provided informed consent. The researchers note that their data are available from the corresponding author upon reasonable request. As generative AI continues to reshape how languages are learned around the world, studies like this one offer a reminder that the technology’s impact cannot be measured solely in terms of test scores or proficiency gains. The emotional lives of learners – their enjoyment, their shame, their boredom, their hope – are shaped by how well these tools meet deep-seated psychological needs, and understanding those needs may be the key to ensuring that AI enhances rather than undermines the human experience of learning a language.

Subject of Research: Basic psychological needs satisfaction profiles and achievement emotions among Chinese EFL learners using generative AI tools

Article Title: Autonomy, competence, and relatedness in the age of AI: decoding the emotional responses of EFL learners

Article References: Liu, W., & Xiao, Y. (2026). Autonomy, competence, and relatedness in the age of AI: decoding the emotional responses of EFL learners. Current Psychology, 45(17), Article 1445. https://doi.org/10.1007/s12144-026-10008-9

Image Credits: AI Generated

DOI: 10.1007/s12144-026-10008-9

Keywords: generative AI, EFL learning, basic psychological needs, achievement emotions, latent profile analysis, self-determination theory, autonomy, competence, relatedness, AI literacy, Chinese university students, educational psychology