Generative AI literacy competencies are often reduced to knowing how to write a good prompt. Ravinithesh Annapureddy, Alessandro Fornaroli, and Daniel Gatica-Perez offer a much broader account.
Their framework identifies twelve competencies that people may need to understand, evaluate, and use generative AI responsibly. It begins with basic AI knowledge and moves through practical use, critical evaluation, technical skills, contextual judgment, ethics, law, and continuous learning.
The authors developed the framework after searching five academic databases for work explicitly addressing “generative AI literacy.” That search produced only six records. Because the research base was still small, they expanded their inquiry through a non-systematic review of related literature on AI literacy, generative AI, prompt engineering, and competency models.
This origin is important. The framework brings together informed ideas from several areas, but it has not been empirically validated as a developmental model.
What do the twelve competencies include?
The first group concerns foundational understanding. People should possess basic AI literacy, understand at a general level how generative models work, and recognize what these tools can and cannot do. This includes knowing that a language model generates responses from statistical patterns and does not operate as a conventional search engine.
The second group concerns practical and evaluative skills. Users should be able to operate generative AI tools, assess their outputs, identify inaccuracies and biases, and write effective prompts. The ability to evaluate output is especially important because fluent language can make weak or fabricated information appear convincing.
The framework then introduces a more advanced technical competency: the ability to program and fine-tune models. The authors acknowledge that most users will never need this level of expertise and can still become AI literate without it.
The final competencies concern situational judgment. Users should understand where generative AI is appropriate, consider its ethical implications, know the relevant legal rules, and continue learning as models, policies, and practices change.
For me, the strongest part of this framework is its refusal to treat prompting as the centre of AI literacy. A technically impressive prompt has little value when the user cannot judge the response, recognize a harmful application, or understand the legal and ethical conditions surrounding its use.
This broader view fits with previous research showing that AI literacy includes more than knowing how to use ChatGPT.
Does everyone need the same competencies?
The framework combines general competencies with highly specialized ones. Basic understanding, output evaluation, contextual judgment, ethics, and continuous learning are relevant to almost everyone. Programming and model fine-tuning belong mainly to developers and technical specialists.
I would therefore use the twelve competencies as a menu or layered framework. A teacher, student, policymaker, researcher, and software developer will need different levels of proficiency. Research on how AI literacy competencies vary among learner groups supports this need for differentiation.
The proposed order also deserves discussion. Ethics and law appear near the end of the progression. The authors placed them there to emphasize their importance, but their position may unintentionally suggest that they come after technical proficiency. In educational settings, ethical and contextual questions should appear from the beginning and continue throughout the learning process.
One competency requires particular caution: the ability to detect AI-generated content. The paper discusses detection tools as possible ways to establish whether content was machine-generated. More recent research has shown that AI detectors in education can produce unreliable and unfair judgments. Detection literacy should therefore include understanding why definitive identification is often impossible. Knowing the limitations of detectors may be more important than knowing how to operate them.
A framework awaiting validation
The authors are clear that some of the proposed benefits and risks are speculative. The twelve competencies emerged from literature review and conceptual analysis. They were not tested with students, teachers, public employees, or expert panels.
We still need research that asks whether the competencies are distinct, whether the proposed order reflects how people actually learn, and how proficiency should be assessed. A person may know that AI can hallucinate while still accepting inaccurate answers. Knowledge, demonstrated skill, and behaviour are different outcomes.
Even with these limitations, the framework gives educators a useful starting point. It can help audit an AI literacy curriculum and reveal what has been overlooked. Many programs teach tool use and prompting while giving far less attention to judgment, context, law, and continuing development.
The framework’s final competency may be the one that holds everything together. Generative AI literacy cannot be completed once and placed on a résumé. Tools change, evidence changes, and laws change. Our understanding has to change with them.
Reference
Annapureddy, R., Fornaroli, A., & Gatica-Perez, D. (2025). Generative AI literacy: Twelve defining competencies. Digital Government: Research and Practice, 6(1), Article 13. https://doi.org/10.1145/3685680
