Generative AI Competence Assessment.

Generative AI tools write, summarise, draw and answer on request, and most people now use one at work or in study without ever having checked what they actually know about it. This test measures that knowledge directly. Instead of asking how confident you feel, it asks twenty questions with one correct answer each, covering what generative AI is and how it works, how to use it well, how to judge what it produces, and the ethical questions it raises. A twenty-first question asks you to write a prompt in your own words.

The twenty keyed items are the Generative AI Literacy Assessment Test (GLAT), an openly licensed instrument developed and validated by education researchers in 2024. The open task is our own addition and is not part of your result; it is there so that the test also captures something you do with generative AI, not only what you know about it.

The test has 21 questions and takes about fifteen minutes. Your result is the number of the twenty keyed items you answered correctly, placed in one of three reference bands. There is no pass mark. Answer without an AI assistant or a search engine open: a result produced with help measures the help, not you.

The test measures generative AI competence as one overall score: how much of the working knowledge and judgement that competent everyday use requires you actually have. The items were written against a four-part blueprint. Know and understand covers what generative models are, how they are trained and why they can produce fluent but false output. Use and apply covers how to get useful results, including how a prompt shapes the answer. Evaluate and create covers judging the quality and reliability of what a tool produces and deciding when to rely on it. Ethics covers privacy, bias, authorship and responsible use.

Those four areas are coverage categories, not separate scores: they make sure the twenty items span the construct rather than clustering on definitions, and they are reported as a single total. The open task, writing a prompt for a realistic job, is recorded with your answers but does not count towards the score. The band you receive says where your total sits against the reference mean of the validation sample, roughly one standard deviation either side; it is a guideline for reading your score, not a percentile and not a certificate of competence.

The GLAT was published in 2024 by Jin, Martinez-Maldonado, Gasevic and Yan as an objective test of generative AI literacy, built on the AI-literacy competency framework described earlier by Long and Magerko. It is an objective alternative to self-report questionnaires, which measure how competent people feel with AI rather than what they know. In the validation sample of 355 higher-education students, tested in English, the twenty items gave a reliability of 0.80 (Cronbach's alpha), with an average item difficulty of about 0.67, meaning the average participant answered roughly two thirds of the questions correctly.

The reference bands this page uses are derived from those published item statistics, not from our own general-population sample. They say whether your total sits more than about one standard deviation below the reference mean, within it, or above it; they do not say what percentage of people you outscored. The test measures knowledge and judgement at one moment and can be improved by learning. It is not a certificate, it does not assess your skill in any particular tool, and it says nothing about your intelligence or personality.