Who is Using AI “Correctly”?

These differences are not always visible on the surface. Two students may submit equally polished work while relying on different evaluation and verification practices behind the scenes. One may actively question AI-generated information throughout the process, while another may accept outputs with minimal scrutiny because the responses appear plausible enough.

A lot of public discussions about AI focus on access: Who has access to advanced AI tools? Who is falling behind? These are important questions. But access (accessibility) to generative AI tools alone does not explain how people engage with AI-generated information. What matters just as much is how users decide whether a generated output is “good enough.” In AI-mediated environments, people may develop different thresholds for what counts as a “good enough” answer. For some users, a quick and usable answer may be sufficient. For others, information may only feel acceptable after careful verification, revision, or comparison with outside sources.

In a recent interview study I conducted with university students, I found that users often approached AI-generated information differently depending on their goals, expectations, and broader contexts of AI use through the lens of Activity Theory. Some participants actively verified AI-generated responses, while others accepted answers if they appeared “good enough” for the task at hand.

Importantly, these differences were not simply about whether students cared about accuracy and credibility of AI outputs. In many cases, users adapted their evaluation strategies depending on what they were trying to accomplish. Students working on quick summarization tasks or coding-related tasks often described more pragmatic approaches, focusing on efficiency and convenience. In contrast, students engaged in more complex writing or research-oriented tasks were more likely to question, refine, and verify AI-generated outputs. This difference may seem small, but it has important consequences.

When “Good Enough” Depends on the Task

As generative AI becomes embedded in education, work, and everyday information environments, people may develop very different habits of verification and reliance. Some users may routinely cross-check information, recognize the limitations of AI systems, and critically revise AI-generated outputs. Others may prioritize speed, convenience, or immediate usefulness, accepting responses that merely appear plausible or helpful.

Over time, these patterns may contribute to uneven vulnerability to misinformation and low-quality information. The risks of AI-generated content are not likely to be distributed evenly across users. Some people may become skilled at negotiating when AI should be trusted, questioned, revised, or ignored. Others may become more dependent on systems that produce fast and convincing answers without engaging deeply with the information itself.

This issue extends beyond the classroom. AI systems are no longer simply tools for retrieving information; they participate in the production, summarization, and circulation of public knowledge. As AI-generated content becomes more common across digital platforms, workplaces, and online information environments, differences in evaluation practices may shape broader patterns of trust and credibility in society. If that is the case, then fairness in the age of AI is not only about access to technology. It is also about differences in how people evaluate what AI systems produce.

When AI Evaluation Shapes Public Knowledge

Discussions surrounding AI in education and policy often focus on access to tools or technical prompting skills. However, future AI literacy may depend not only on the ability to use AI systems effectively, but also on the ability to critically evaluate AI-generated outputs across different tasks and contexts. Beyond these debates, we may also need to pay greater attention to how AI systems are designed to support users’ evaluation and interpretation of AI content across varying information environments.

Yet, we still stand at the beginning of this AI transition, and many current legal and policy debates surrounding AI continue to focus on training data, copyright, fair use, and sourcing practices. At the same time, as AI-generated content becomes embedded in public knowledge environments, the broader challenge may involve how AI-generated outputs circulate, shape public knowledge, and are evaluated across different contexts.

The future challenge of AI literacy is not simply teaching people how to use AI tools but helping people understand when AI-generated information should be trusted, verified, revised, or rejected. In an AI-mediated society, public knowledge depends not only on access to information but also on differences in how people evaluate it. AI systems also need to evolve to better support users’ evaluative literacy in diverse information environments.