AI in education – how does it change learning?

AI in education changes the way we learn primarily because it can tailor content, pace and the format of exercises to a specific student, instead of sticking to one scheme for the whole class. This makes it easier to maintain the right level of difficulty, spot gaps more quickly and learn more effectively, not just “more”. At the same time, generative tools (e.g. chatbots) open up new modes of work. From exercises guided step by step to preparation for tests and the matura exam with immediate feedback. However, these same solutions require sensible habits, because models can make mistakes and build answers that only appear reliable. In this article, we show practical, tried-and-tested applications of AI in learning, without promising “magic” and without taking shortcuts. Keep reading if you want to understand how to use AI so that it strengthens real skills, rather than just producing ready-made answers.

How does AI personalise learning in education?

AI personalises learning by matching the level of difficulty and the type of exercises to your answers and response times in tasks. Adaptive systems (e.g. in Duolingo or Khan Academy/Khanmigo) can mean that one person keeps practising fractions for longer with more examples, while another moves on to equations more quickly because the basics are already mastered. In practice, this reduces the risk of getting stuck with material that is too easy or too difficult. This kind of matching works best when the student completes tasks regularly and the system has data on which to base conclusions about progress.

AI also supports “1:1” learning, because chatbots and tutors can ask guiding questions, explain step by step and ask for justification of an answer. As a result, the student does not receive only the outcome, but is guided through the reasoning, much like in tutoring. In addition, “knowledge tracing” models analyse mistakes and point to specific skill gaps (e.g. dividing fractions, rather than just “maths” in general). On this basis, the app can suggest a tailored plan, e.g. 15-minute revision sessions for 5 days, instead of general revision of the whole topic.

AI helps students learn at a pace suited to them by breaking the material into micro-lessons and matching the format (text, video, quiz) to what works in a given case. For example, someone with a shorter attention span may get 6 short exercises of 3 minutes each instead of one 20-minute test. In long-term revision, tools such as Anki or Quizlet use spaced repetition, and AI can generate flashcards from notes and set intervals based on real mistakes. To make sure personalisation does not “lock” the student into a bubble of easy tasks, it is worth using mastery thresholds (e.g. 80–90% correct answers) and regularly adding mixed tasks that test knowledge transfer.

1. 01Analysis of progress & behaviourBased on the student’s answers and response time
2. 02Content adaptationMatching the level of difficulty and type of exercise. Fundamentals mastered → Challenges & equations
3. 03Individual 1:1 supportTutors guide the reasoning process

Key result: Reduced risk of getting stuck with material that is too easy or too difficult. Key takeaway: AI personalises learning through continuous analysis, dynamic content adjustment and individual support, allowing students to learn effectively at their own pace.

Using generative tools for effective student work

Generative tools can genuinely improve a student’s work, provided they are used for explanations, exercises and checking the reasoning process, rather than for mindlessly copying ready-made answers. Students most often turn to solutions such as ChatGPT/Claude/Gemini for summaries, step-by-step explanations and sample tasks, with the best results appearing when they add their own notes and ask for check questions. The same applies to language learning. Tools such as ChatGPT, DeepL, Grammarly or LanguageTool help with proofreading, but their greatest value comes from simulating conversations and giving specific feedback on errors (tenses, prepositions, word order). A student can practise a dialogue “at a hotel reception” and receive a list of the most common mistakes together with corrections and short exercises.

Generative AI can support essay writing without plagiarism if it is used to prepare an outline, thesis, counterarguments and refine style, rather than to generate the whole text 1:1. The safe standard is to treat AI as an “editor and sparring partner”: it helps structure the work, but the arguments and examples must come from the reading and the student’s own work. In maths and the sciences, tools such as Wolfram|Alpha, GeoGebra or Photomath are best used to check correctness and analyse mistakes (“which step did I transform incorrectly?”), rather than to bypass the solving process itself. A good practice is to provide AI with your own reasoning and ask it to identify the first incorrect step, rather than asking for the final answer alone.

• Add your own notes and ask for check questions, instead of relying solely on a summary.
• When writing essays, use AI for the thesis, outline and counterarguments, and base the content and examples on the reading.
• For science and maths tasks, ask for a diagnosis of the error in your solution, rather than the final answer.
• When preparing for tests and the matura exam, generate task sets “according to the key” as well as 45-minute mini mock papers and work with an “examiner” mode that requires justification.
• Limit the risk of hallucinations: require sources, ask “what do I need to check in the textbook/primary source” and confirm factual information in at least 2 independent sources (e.g. using tools that provide links, such as Perplexity).

Generative tools also work well in projects, because they can break a topic down into roles, set out a schedule (e.g. 2 weeks, 4 people) and prepare quality checklists that organise the team’s work and reduce chaos. They also support critical thinking when a student asks for a list of sources, identification of assumptions and hypothesis testing, rather than expecting “one answer”. Image and video generators are most useful when they are used to visualise concepts (process diagrams, mind maps) and projects, rather than simply to create “pretty pictures”. In practice, the teacher and the student should assess factual accuracy and the way arguments are justified, because these are the elements that build real competence.

The role of AI in assessment and feedback for students

AI supports assessment and feedback because it can automatically check part of the work and quickly indicate exactly what the student should improve. This works best for multiple-choice quizzes and short answers, where a rubric and a set of examples of correct solutions can be clearly defined. In practice, the teacher can prepare a test in Moodle, Canvas or Forms, and AI will help analyse the results and spot questions that are too difficult or ambiguously worded. This shortens response time and makes it easier to plan revision in class.

AI can also speed up feedback on essays by analysing the structure of the argument, coherence and language, and by generating specific comments such as “add a counterargument” or “there is no example”. The most useful feedback is produced when assessment is based on a rubric (e.g. criteria scored 0–5 points), because the student can see what they are losing points for and exactly what they need to improve. In open-ended tasks, tools can additionally identify typical mistakes (e.g. in physics, confusing units; in chemistry, problems balancing equations), which means the feedback remains concrete rather than vague. This approach supports formative assessment because it directs attention to skills, not just the “final result”.

• Set a rubric and examples of correct answers before using AI to mark work or prepare comments.
• Introduce short diagnostic tasks, then formulate micro-goals (e.g. “definition mastered, application missing”).
• Analyse solution strategies and ask for typical errors to be identified, rather than limiting yourself to points alone.
• Use learning analytics from platforms (e.g. Google Classroom, Microsoft Teams, Moodle, Librus) to detect risk patterns, such as a drop in work submissions for 3 weeks.
• Treat AI/plagiarism detection as a supporting signal and assess the process: drafts, sources, notes, oral defence of the work and writing in class.

AI also helps measure progress in skills by mapping tasks to specific competencies (e.g. “argumentation”, “reasoning”, “chart interpretation”) and showing changes over time. In class, it can support immediate feedback when tools such as Kahoot, Socrative, Nearpod or Edpuzzle collect responses live and signal which concept should be discussed again. When creating tests, AI can be useful for generating question banks at different levels of difficulty and for distinguishing between memory-based questions and application questions (a reference to Bloom’s taxonomy). This means assessment can be both faster and more diagnostic.

1. 01Automated checkingQuizzes and short answers
2. 02Results analysisDetecting difficulties, saving time
3. 03Faster feedbackStructure analysis, improvement suggestions

AI supports teaching best when assessment is based on clear rubrics and examples.

Supporting student accessibility with AI

AI supports student accessibility because it makes content easier to access, helps organise work and enables active participation in lessons for people with diverse needs. For dyslexia and reading difficulties, TTS (text-to-speech) and simplified paraphrases work well, while tools such as Immersive Reader (Microsoft) can split words into syllables and highlight parts of speech. A student can listen to a set text and read a version with a larger font and increased line spacing at the same time. This works best when the support is a constant part of the learning process, rather than an “emergency fix” just before a test.

AI also supports students with ADHD by breaking tasks into short stages and creating specific timed checklists (e.g. 3 blocks of 12 minutes instead of 1 hour). For students with sensory disabilities, speech recognition (STT), the option to dictate answers and automatic captions for videos are important. In practice, Teams and YouTube offer live captions, and the student can receive a transcript of the lesson for revision. This increases independence, because there is no need to rely solely on notes taken during the lesson.

AI can also level the playing field by providing access to materials and a “tutor” 24/7 at no travel cost, especially when the school uses free resources (e.g. Khan Academy, open courses). In multilingual classes, contextual translation (DeepL, Google Translate) and simple explanations in the student’s language help them get into the curriculum more quickly, on the assumption that they gradually return to Polish. In special education (SPE), AI can be helpful, but it only works safely with clear instructions, short and predictable interactions, and content control by the teacher or carer. When designing materials compliant with accessibility (WCAG/UDL), AI can assess vocabulary difficulty, suggest alt text for images and prepare an “easy to read” version, and in practice it is also worth making sure there is sufficient contrast, short paragraphs, and captions and transcripts for videos.

The new role of the teacher and the development of future skills

AI most often does not replace the teacher, but shifts their role from “delivering content” to designing educational experiences and training students’ thinking. In practice, this means more moderating discussions, checking lines of reasoning, and teaching argumentation and information verification, which AI does not provide in the relationship. The key principle becomes: AI can propose, but the didactic decisions and assessments remain on the human side. This model also supports work on competencies, rather than solely on covering topics.

In the AI era, critical evaluation of materials, the ability to structure problem-based tasks, and formative assessment are becoming increasingly important. Added to this is “prompting” in a didactic sense, that is, phrasing instructions so that AI creates exercises aligned with the objectives and success criteria. The teacher can also shape responsible use of AI by introducing rules for citation, describing AI’s contribution, and requiring fact-checking, much like when teaching how to work with the internet. A simple standard helps here: the student adds a section “How I used AI” and provides sources confirming the key claims.

AI also reduces the teacher’s workload during preparation, because it can generate lesson plans, worksheet variants, marking rubrics or messages to parents, but such materials should be treated as a draft that needs to be checked and adapted to the realities of the class. The purpose of homework is changing too: since an answer can be generated easily, process-focused tasks work better, such as drafts, reflection, or a recording explaining the line of reasoning. To avoid AI use becoming too absorbing, station rotation works well: some activities with AI (practice and feedback), some without AI (discussion, problem-solving, experience), and then a synthesis of understanding at the end. In assessment, a clear separation of when AI may and may not be used, as well as more frequent assessment in controlled conditions (e.g. in-class work, oral response, project with defence), strengthens the sense of fairness.

1. 01Designing educational experiencesFrom delivering content to creating learning
2. 02Moderating discussions and verificationTraining thinking and critical evaluation
3. 03Human as the final decision-makerAI proposes, the teacher assesses
4. 04Development of future skillsWorking on skills, not just topics

The teacher’s role is evolving towards that of a mentor who supports students in critical thinking and using AI as a tool.

Creating didactic content with the help of AI

AI makes creating content for teaching easier when the teacher starts from the requirements and learning outcomes, and only then chooses the tool. You can paste in the requirements (e.g. the “functions” unit in secondary school) and ask for a lesson plan with operational objectives, success criteria and tasks for different levels. The condition for quality is a manual check against the syllabus, because models may get details wrong or suggest content outside the scope. This approach saves time while maintaining subject matter control.

AI can also significantly shorten the time needed to prepare worksheets and exercises, because it generates sets of tasks with an answer key and variants for weaker and stronger pupils. When differentiating instructions, it can rephrase a task in simpler language, add an example and clarify the assessment criteria, which reduces the number of misunderstandings in the classroom. It can also support multimedia materials by preparing scripts for short videos, quizzes for video (e.g. in Edpuzzle), slides with check questions and suggestions for experiments. With this type of content, the teacher should additionally ensure accuracy, copyright compliance and suitability for the age of the pupils.

In project-based and problem-based teaching, AI can prepare scenarios (e.g. a trip budget, smog analysis), data sets and research questions that lead to conclusions, while the actual data collection and defence of the results still belong to the students. In developing writing and argumentation, generative tools can act as a “reviewer”: they spot gaps in reasoning, the lack of counterarguments and make it easier to refine a text step by step, especially when the teacher expects versions 1.0 and 2.0 together with a description of the changes introduced. In programming teaching, AI can support debugging by helping to find bugs, suggesting unit tests and explaining code line by line, supporting debugging rather than unthinking copying. To keep AI as a tool rather than an end in itself, it is worth structuring the lesson so that it takes up around 10–30% of the time, and the rest is student work: discussion, experimentation, writing or problem-solving.

Ethics, privacy and security in education with AI

Ethics, privacy and security in education with AI come down primarily to protecting student data, limiting the risk of errors and having transparent rules for using tools. The most sensitive data are those identifying the child (name, e-mail, register number, image), as well as information about grades, attendance and health issues, because they can enable profiling. In practice, it is worth sticking to a simple rule: do not paste personal data or full assignments marked with a surname and school name into external chatbots. GDPR requires the school to have a legal basis, appropriate data processing agreements and clearly defined purposes of processing, and with tools outside the EU there is a risk of data transfer.

Safe use of AI also includes a conscious approach to copyright, bias and the quality of answers. Content generated by AI can infringe copyright, especially when the model reproduces fragments similar to protected materials or when the user pastes other people’s texts without a licence, so good practice is to use OER, cite sources and check image licences. Models can reinforce stereotypes (bias), so it is worth testing the tool across different pupil profiles and remembering that AI is there to suggest, while teaching decisions and grades are made by a human. Hallucinations and factual errors require a consistent verification procedure against sources, e.g. by listing several claims from the AI’s answer and checking them in a textbook, encyclopaedia or statistical data (GUS, Eurostat). AI can also generate age-inappropriate content, so schools use managed accounts, allowlists of tools and work on specific tasks under supervision.

Academic integrity in the AI era requires a clearly written policy: what is allowed (e.g. language correction, preparing an outline), what is forbidden (e.g. generating the entire assignment), how the use of the tool is disclosed, and what consequences apply. To make the rules realistically enforceable, it is worth expecting traces of the process: notes, drafts, bibliographies and a short oral defence of the key arguments. Where the priority remains data protection and control (e.g. analysing pupils’ work or sensitive data), the school can consider local or private models such as Azure OpenAI, private hosting or solutions with a “no training on your data” option, which reduces risk but requires budget and IT expertise. Security also means cyber hygiene: pupils should learn to recognise phishing and fake tool websites, and before creating an account they should check the domain, certificate and privacy policy.

Introducing AI in schools and its impact on the labour market

Introducing AI in schools works best when you start with 2–3 specific use cases and only then select the tools and integrations. Good starting points are scenarios such as feedback on essays, diagnostic quizzes or differentiating worksheets, launched as a pilot in one class for 6–8 weeks. In a pilot, it is a good idea to define metrics straight away: teacher time, results, pupil satisfaction and the number of corrections, so you do not end up with an implementation “on faith”. Ecosystems such as Google Workspace for Education and Microsoft 365/Teams, LMS platforms (Moodle, Canvas) and content tools (H5P, Nearpod, Edpuzzle) are most often combined, because this reduces the number of logins and makes it easier to control data and permissions.

Implementation costs depend on licences (sometimes 0 PLN in educational versions, sometimes several dozen PLN per teacher per month), hardware and training, with the biggest “hidden cost” often being the time needed to get everything up and running. Schools often start with free features in tools they already have, and only later buy premium licences for specific subject teams. Good training should not end with “how to use the buttons”, but should cover practical scenarios: creating a rubric, generating tasks at three levels, planning project work and checking hallucinations. A practical mini-standard is for the teacher to prepare one lesson with AI, deliver it, and then discuss the results at a training council.

The impact of AI on the labour market is above all the growing importance of information-handling skills and working alongside tools, not merely “using applications”. Increasingly important are: checking the reliability of sources, formulating precise prompts, data analysis, communication and collaboration with tools, because in many professions there will be a “copilot” for documents, code or analysis. A pupil who can evaluate an AI output and refine it will be more competitive than someone who accepts the answer uncritically. In assessment, process-based forms are becoming more likely (portfolios, projects with a defence, live tasks and short in-class competency checks), while trends include AI-supported tutoring, “on demand” generated content, learning analytics and accessibility tools. At the same time, as a passing trend, solutions without quality control may appear, promising “full automation of the school”.