Articles

›

Article · AI Literacy

AI Literacy in High School and Higher Education

By Allison Westgate

·

Partnerships @ Collage AI

·

·

6 min read

Illustration of a student's handwritten notes, a laptop showing AI feedback with checked and rejected items, a magnifying glass with a check mark, and an open textbook

Students can produce polished work with AI without being able to explain the reasoning behind it. For educators, the central question is not simply whether students use AI, but whether that use supports understanding they can demonstrate independently.

Students can produce polished work with AI without being able to explain the reasoning behind it. For educators, the central question is not simply whether students use AI, but whether that use supports understanding they can demonstrate independently.

Key takeaways

  • AI literacy should be assessed through students’ decisions and explanations, not tool-use frequency.

  • Structured guidance can help students examine AI outputs more critically.

  • Research on AI-supported reading comprehension and reflective AI use highlights the importance of matching assistance to learners and tasks.

  • High school and college assignments can make permitted use, verification, disclosure, and independent understanding explicit.

  • Educators retain responsibility for learning objectives, assessment, and the role AI plays.

AI literacy combines technical understanding, critical evaluation, practical application, and ethical judgment. Frequent tool use alone is not competence. In high school and higher education, meaningful AI literacy requires students to evaluate assistance, make responsible decisions, and remain accountable for their work.

What is AI literacy in education?

AI literacy is the ability to understand what AI can and cannot do, evaluate its outputs, use it for a defined purpose, and judge when its use is appropriate. This article organizes those responsibilities into four connected areas: technical understanding, critical evaluation, practical application, and ethical judgment. These are a practical teaching framework, not a claim that one study established a universal definition.

Recent North American research supports treating AI literacy as more than operational skill. Prabhudesai et al. (2025) examined how university students evaluated a large language model, the type of system behind many conversational AI tools. Students without basic AI literacy struggled to identify and investigate bias independently. Structured auditing guidance helped them engage more purposefully with the model. The study suggests that telling students to “check the answer” is not equivalent to teaching them how.

Why frequent AI use is not the same as competence

Using AI often demonstrates familiarity, not necessarily understanding or judgment. A student may write effective prompts while accepting unsupported claims, overlooking bias, or relying on explanations they cannot reproduce. A stronger assessment asks what the student understood, checked, changed, and decided.

Technical understanding: explain the tool’s limits

Students should be able to explain, in practical terms, why a plausible response still needs verification. Instructors can ask students to distinguish a generated explanation from evidence that supports it, identify when the tool lacks relevant course context, and explain what additional information they need before relying on an answer.

A useful task is to ask students to annotate an AI-generated explanation: What does it claim? What evidence would establish whether it is correct? Which parts remain uncertain? The aim is not to require expertise in model engineering, but to make the limits of the tool visible.

Critical evaluation: investigate rather than accept

Critical evaluation means testing accuracy, relevance, assumptions, and possible bias. Prabhudesai et al. (2025) found that students’ interactions became more focused when an auditing activity provided explicit structure. Their small, quasi-experimental study supports guided evaluation activities.

In practice, students can compare an AI response with assigned readings, verify that cited sources exist, and identify whose perspectives are missing. An assessment should reward the quality of that investigation, including a well-supported decision to reject the response.

Practical application: choose help that serves the learning goal

Practical competence involves deciding which assistance is useful and which would bypass the skill being learned. Asking for feedback on an existing argument differs from asking AI to supply the argument itself.

For example, a student might draft an explanation, request a challenge to one assumption, and then revise using course evidence. The student should be able to explain why the revision is stronger. The relevant evidence of competence is the reasoning behind the choice, not the number of prompts submitted.

Ethical judgment: take responsibility for use

Ethical judgment includes following assignment rules, protecting sensitive information, acknowledging assistance, and considering unfair or exclusionary outputs. Educators can make these responsibilities concrete through scenarios: May a student upload a classmate’s unpublished work? Does this assignment permit generated prose? What disclosure would accurately describe the assistance received?

Together, these considerations make a useful distinction: AI literacy necessitates responsible decisions about a tool, as well as the ability to operate it.

What recent research shows about AI-assisted learning

Recent research suggests that the educational impact of AI depends not simply on access to the technology, but on how it is designed and used to support learning. Learning outcomes, critical engagement, and reported confidence are different measures; evidence for one should not be presented as proof of another.

AI support can affect readers differently

Etkin et al. (2025) examined GPT-based reading tools in a preregistered, randomized crossover study with college-aged adults living in the United States. Participants completed standardized reading-comprehension tasks with no AI assistance and with four forms of support: a generated summary, an outline, a question-and-answer tutor, and a Socratic discussion chatbot.

The findings differed by baseline performance and tool. In analyses grouping participants by prior SAT/ACT scores, the Socratic chatbot improved comprehension for the lower-scoring group, while reading an AI-generated summary instead of the original passage reduced comprehension for the higher-scoring group. Analyses grouping participants by their no-AI passage scores showed broader differences across tools. This variation matters: the study does not support treating every form of AI assistance as equally beneficial for every reader (Etkin et al., 2025).

For educators, the implication is to evaluate the fit between a tool, a learner, and a task, rather than assume that adding AI support improves understanding.

Reflective AI use differs from delegating the work

Hakola et al. (2026), the study behind the supplied EurekAlert release, examined undergraduate take-home exams at the University of Eastern Finland. The researchers identified three approaches:

  • Instrumental use: delegating task completion with limited evaluation.

  • Ideational use: generating and selectively incorporating ideas.

  • Reflective use: using AI for critical evaluation and feedback-driven refinement.

Reflective engagement was associated with stronger and more consistent performance, while instrumental use was associated with weaker performance on some tasks. A classroom adaptation is to have students draft first, seek feedback second, and justify any revision with disciplinary evidence (Hakola et al., 2026).

How to teach and assess AI literacy

Make AI use visible, teach verification explicitly, and assess what students can explain without assistance. The following practices translate the research into instructional options.

In high school: provide bounded practice and clear expectations

High school teachers can define permitted assistance at the assignment level. Instead of a general statement that AI is “allowed,” specify whether students may use it for hints, brainstorming, feedback, or generated text, and what they must complete independently.

A bounded activity might ask students to write an initial explanation, request feedback through a school-approved tool, and compare that feedback with a teacher-selected source. Students then submit a revision note explaining what they accepted, rejected, or corrected. A short unassisted explanation can show whether they understand the concept.

In higher education: align use with disciplinary judgment

College and university instructors can connect permitted AI use to the thinking a course is designed to develop. A literature course might examine unsupported interpretation; a methods course might investigate whether an answer fits the research design; a professional program might consider confidentiality and accountability.

A useful sequence is attempt, consult, verify, revise, explain:

  1. Attempt: Produce an initial response or solution independently.

  2. Consult: Request a bounded form of assistance, such as a hint or critique.

  3. Verify: Check substantive claims against course materials or credible original sources.

  4. Revise: Make changes based on evidence rather than the tool’s confidence.

  5. Explain: Describe the reasoning and demonstrate understanding without relying on the generated response.

This proposed sequence combines the guided evaluation emphasized by Prabhudesai et al. (2025) with the distinction between reflective and instrumental use described by Hakola et al. (2026). It should be adapted to the learning objective as needed.

Assess judgment as well as the final product

An AI-literacy assessment can ask students to explain tool limitations, verify a consequential claim, select appropriate assistance, and disclose use accurately. Formative assessment provides feedback during learning; summative assessment evaluates achievement at a defined endpoint. Both can preserve a place for independent demonstration.

Instructors can also offer an equivalent route for students who cannot or should not use a particular tool. The aim is to assess the intended learning.

AI literacy is ultimately a shared educational responsibility. Students practice accountable use; educators define meaningful learning and evidence of achievement; institutions establish appropriate conditions for access and oversight. The lasting measure is what students can understand, evaluate, explain, and apply for themselves.

About the author

Allison Westgate

Partnerships @ Collage AI

References

  1. Estaiteyeh, M., & Mindzak, M. (2025). Building AI literacy in pre-service teacher education in Canada: A case study of two cohorts. Journal of Teaching and Learning, 19(4), 135–154. https://doi.org/10.22329/jtl.v19i4.10041

  2. Etkin, H. K., Etkin, K. J., Carter, R. J., & Rolle, C. E. (2025). Differential effects of GPT-based tools on comprehension of standardized passages. Frontiers in Education, 10, Article 1506752. https://doi.org/10.3389/feduc.2025.1506752

  3. Hakola, O. J., Kukkonen, J., & Valtonen, T. (2026). Self-regulated approaches to students’ generative AI use in higher education. Computers in Human Behavior Reports, 24, Article 101340. https://doi.org/10.1016/j.chbr.2026.101340

  4. Prabhudesai, S., Kasi, A. P., Mansingh, A., Antar, A. D., Shen, H., & Banović, N. (2025). “Here the GPT made a choice, and every choice can be biased”: How students critically engage with LLMs through end-user auditing activity. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems. Association for Computing Machinery. https://doi.org/10.1145/3706598.3713714