AI is changing online education faster than many schools expected. Students can ask an AI system to explain a difficult concept, generate exercises, translate a text or give instant feedback. That raises an awkward question: if a machine can explain something in seconds, what is a human tutor for? Research from 2025 and 2026 suggests tutors are not becoming less important. Their role is shifting from delivering information to diagnosing learning problems, personalising support and creating meaningful interaction.
Online learning is now mainstream in Europe
According to Eurostat, 34.8% of internet users in the EU used an online course or online learning materials in 2025, up from 21.4% in 2019. That is a rise of 13.4 percentage points in six years, or roughly 63% relative to the 2019 level. Countries differ widely. In 2025, 60.2% of internet users in the Netherlands reported online learning, followed by 59.7% in Ireland, 50.7% in Finland and 50.6% in Sweden. At the lower end, the figure was 11.8% in Romania, 18.4% in Bulgaria and 21.0% in Cyprus.
Online education also extends beyond formal courses. In 2025, 30.5% of EU internet users used online learning materials, while 17.3% took an online course. Few people still ask whether online education will become part of everyday learning. The more pressing question is which kind of online education produces meaningful results.
AI adoption is growing, and so are the concerns
The OECD Digital Education Outlook 2026 reports that 37% of lower-secondary teachers used AI for their work in 2024. Some 57% agreed that AI could help them write or improve lesson plans, while 72% believed it could make it easier for students to present someone else’s work as their own. The OECD’s underlying data give slightly different figures: around 36% of lower-secondary teachers reported using AI in their teaching, 53% said it helps with lesson planning and 52% said it can help adapt materials to students’ different abilities. Teachers appear to be using AI as a support layer rather than a replacement for teaching.
The OECD adds a caution. Completing a task successfully with GenAI does not mean a student has learned the underlying material, and without pedagogical guidance AI can improve task performance without producing learning gains. That distinction is likely to define the next phase of online education.
Students are moving faster than teachers
The trend goes well beyond teachers. A European Commission report published in September 2026 found that four in ten young people aged 16 to 24 in the EU used generative AI for formal education in 2025. The Commission also notes that educational AI tools generally produce better learning outcomes when they support students’ own thinking instead of doing the task for them. Yet only 29.7% of EU teachers surveyed in TALIS 2024 had taken part in professional development on AI for teaching and learning.
Students, in other words, are adopting AI faster than many teachers are being prepared to use it well. Part of a tutor’s job may therefore be showing students how to use AI without handing the learning over to it.
What the research says about online tutoring
The evidence on online tutoring is more mixed than EdTech marketing tends to suggest. A 2026 randomised controlled trial in Germany followed 839 secondary-school students on an online tutoring programme for low-performing students. Six months after the programme began, the effect on mathematics grades was 0.06 standard deviations, which was not statistically significant. For students with no previous tutoring experience, however, the effect was 0.14 standard deviations.
The study also looked beyond grades. Among students in non-academic school tracks, taking part in the programme was associated with a 5-percentage-point increase in vocational training take-up 18 months later. Among those without previous tutoring, the difference reached 12 percentage points. Averages can hide large differences between groups, and online tutoring clearly helped some learners more than others, particularly those who had never had access to tutoring before.
Research with younger children points the same way. A randomised controlled trial published in Educational Evaluation and Policy Analysis studied children from kindergarten to Grade 2 who were assigned to one-to-one tutoring, two-to-one tutoring or a control group. Virtual tutoring raised early literacy skills by 0.05 standard deviations across the full sample. When English learners and students with disabilities were excluded, the estimated effect rose to 0.08. The clearest gains came among students who started well below the expected benchmark, where one-to-one tutoring produced improvements of about 0.15 standard deviations. The researchers also noted that effects were smaller than those usually found in comparable in-person literacy tutoring. “Online” is not automatically better or worse than “in person”; results depend on how the intervention is designed and who the learner is.
AI in English learning: growing evidence, one big gap
A 2026 systematic review in Computers and Education: Artificial Intelligence examined 708 publications on AI and English as a foreign language from 2019 to 2024. After screening, the researchers kept 103 empirical studies. The findings were generally positive, particularly for integrated language skills, writing and speaking. But 75% of the studies (77 of 103) focused on university-level learners, against only 7% on secondary education and 8% on elementary education. AI-powered chatbots were the most common tool, appearing in 56% of tool-focused studies.
Much of the evidence base therefore concerns adults. For providers working with children, that is a real research gap, because technologies that work for independent adult learners cannot be assumed to work the same way for younger students.
Where the tutor adds most value
These findings suggest a different picture of the online tutor. The traditional tutor was mainly a source of information. The tutor in the AI era is closer to a learning diagnostician and coach.
Take a child who makes five grammar mistakes. An AI system can spot all five, but the educational question is whether they are five separate errors or symptoms of a single misunderstanding. If the child does not grasp when to use a particular tense, another 20 exercises may only repeat the same problem. A tutor can find the underlying difficulty, change the explanation, offer a different example and judge when the learner is ready to move on. That is the point where personalisation stops being an algorithmic recommendation and becomes teaching.
AI also changes the economics of tutoring. A tutor’s time is expensive and limited, and AI can reduce the routine work needed before, during and after a lesson. It can help produce differentiated practice materials, vocabulary exercises, extra examples, lesson summaries, formative feedback, revision activities and versions of the same task at different difficulty levels. That does not make the tutor less important. It frees time for the activities where human judgement counts most. The OECD’s 2026 evidence is consistent with this: teachers report using AI for lesson planning and adapting materials, and the OECD describes AI as potentially working as a tutor, partner and assistant when it is embedded in sound pedagogy.
Language learning still depends on human interaction
English learning makes the distinction especially clear. An AI system can generate vocabulary lists, correct written sentences and simulate a conversation. A child learning English also needs real communication: waiting for a reply, asking for clarification, misunderstanding something, trying again and slowly gaining confidence. Those experiences are part of the learning itself, not technical extras. This is one reason online language providers are increasingly combining digital tools with live teaching instead of choosing between them. Platforms such as LearnLink, for example, offer online access and individual English lessons for children. The model reflects a wider shift in the sector: technology can remove geographical and scheduling barriers, while the tutor remains responsible for the human and pedagogical side of the lesson.
Europe is redefining what teachers need to know
European education policy is heading in the same direction. In June 2026, the European Commission and the OECD introduced an AI Literacy Framework for primary and secondary education, with 4 dimensions and 19 competences covering how learners engage with, create with, manage and shape AI. The Commission has also reported that 81% of EU citizens believe teachers should have the skills to use and understand AI, and almost 90% believe teachers should be able to help students recognise online disinformation. Its 2026 review of the teaching profession found that only around one in three EU teachers reported using AI in their work, and that AI-related training remains limited.
The tutor of the future will need two kinds of expertise. One is traditional subject knowledge. The other is AI and digital literacy: knowing when technology is useful, when it misleads and when a learner needs human guidance instead.
From “teacher versus AI” to “teacher plus AI”
Asking whether AI will replace online tutors is probably the wrong question. A more useful one is which parts of tutoring should be automated and which should stay human.
The emerging evidence points to a division of labour. AI suits scale and routine: it can generate materials, analyse patterns, give immediate feedback and support extra practice. Tutors are better placed for interpretation and interaction, since they diagnose misunderstandings, adjust their teaching, sustain motivation and create meaningful communication. Students still own the learning itself, and if AI does the cognitive work for them, better-looking answers will not necessarily mean better learning.
The next generation of online schools may be defined less by how much AI they use than by how well they combine it with human teaching. The online tutor of 2026 is not disappearing, but the job is changing. Less time goes on delivering information and more on understanding the learner. Less goes on creating repetitive exercises and more on deciding which exercise matters. The most effective tutors may be those who know when technology should take over and when it should step aside.
Adam Mulligan, a psychology graduate from the University of Hertfordshire, has a keen interest in the fields of mental health, wellness, and lifestyle.
