AI isn't saving teachers time - it's another thing to mark according to research by Up Learn
Almost three quarters of teachers rank accuracy as their top AI concern, and more than half say checking its outputs outweighs the time saved.
LONDON, Sept. 10, 2026 /PRNewswire/ -- The debate around AI in schools has largely centred on plagiarism and cheating, yet new research, from online learning platform Up Learn, shows students and teachers are more concerned about accuracy and whether AI can be trusted to get things right.
In a survey of over 2,800 students and teachers, accuracy or reliability of outputs ranked as the single biggest concern for teachers using AI, cited by 73%. More than half (56%) of teachers surveyed said the checking required means it may not save them time at all, and a similar proportion (54%) doubted the quality would match their own teaching standards.
Beyond accuracy, nearly half (48%) flagged misalignment with exam boards and specifications and a third (33%) cited privacy and safeguarding concerns.
Yet both groups keep using AI at scale. More than two thirds of teachers leverage the technology in their day-to-day work – most often on at least a weekly basis. Only 8% reported no concerns at all, with the remainder pointing to concerns such as lack of guidance or training (19%), safeguarding issues (33%), and overreliance (40%).
Students report the same doubts. Accuracy topped their list too, at 74%, making it a rare point of agreement across the classroom. Cheating was cited by only one in four students, well behind fears of becoming too reliant on AI (65%) and losing the ability to think for themselves (47%).
How students say they use it explains the gap. They were far more likely to turn to AI to explain difficult concepts (80%) and create summaries (61%) than for essay-writing support (40%).
If students are turning to AI to be taught rather than to cheat, the question is not about discipline but the tool's fitness for purpose.
Guy Riese, CEO and founder of Up Learn, says: "AI is solving an education gap. Students are using it to understand things they've been taught but haven't grasped and they are rightly sceptical about the answers they get. This is where we need to support students - by helping them turn that scepticism into a skill: knowing when AI has got it right, and when to look again. AI is part of the new normal, students need to be equipped with tools and techniques that enable them to use it with trust."
Riese suggests three questions teachers can apply to any AI-generated resource before it reaches a classroom, and teach students to apply themselves:
Can it be checked? Before using any resource for their learning, ask where did the answer come from, and was it built by subject experts?
Is it aligned to the specification? General knowledge about a subject is not the same as being prepared for a specific exam board. General-purpose AI will only return what a student thinks to ask about, which leaves gaps that only surface in the exam hall. Anything set as independent work should be built to the specification being taught.
Who is doing the thinking? Learning should feel hard. This is known as 'desirable difficulty' and it's how you know it's working. Watching a model produce a worked answer is not the same as producing one. Independent work needs to require retrieval, not recognition, if it's going to hold until results day.
Notes to editors
Methodology
Research was conducted by Up Learn via an online UK survey of 2,591 students and 248 teachers in its contact base between 1 May and 17 August 2026.
About Up Learn
Up Learn is an adaptive attainment platform for GCSE and A Level, built by teachers and educational scientists. It combines expert teaching, adaptive learning and cognitive science, with AI supporting rather than replacing learning. It is trusted by 685+ schools and used by 1 in 3 A Level students in the UK.
Share this article