AI Helps Trainee Teachers Learn to Question the Evidence

Across Europe, governments and universities are pushing hard to make teaching a genuinely evidence-informed profession. England’s Early Career Framework promises new teachers support linked to the best available research evidence, the Netherlands expects university lecturers to apply recent evidence from their domain, and the European Commission has urged that student teachers be supported to use and engage in research in their practice. Yet a stubborn gap has opened up between the ambition and the reality: while trainee teachers are routinely taught to use digital technology for lesson planning and classroom tasks, far less is known about how they can use technology to interrogate the research evidence itself. A new study from the University of Southampton, published in the Journal of New Approaches in Educational Research, set out to close that gap.

The research, led by Christian Bokhove with Lucy Hoyes and Ros Hyde, formed the English case within the Erasmus+ project ‘Research in Teacher Education’, in which five teams from four European countries collaborated to help beginning teachers in STEM subjects develop evidence-informed practice. The team redesigned a postgraduate secondary mathematics teacher preparation programme around two complementary theoretical frameworks: a set of six evidence-use mechanisms distilled by Breckon and Dodson from a meta-review of dozens of systematic reviews, and Engeström’s expansive learning cycle, a model in which learners question existing practice, analyse its contradictions, model a new solution, implement it, reflect and consolidate. Crucially, the researchers identified the questioning phase of that cycle as the moment when beginning teachers have the best opportunity to engage with evidence, and they built technology into that phase deliberately.

The team defined evidence broadly, as the best available information on a topic, gathered and interpreted by researchers, practitioners or experts through intentional and systematic procedures, with format and focus varying according to purpose. That breadth matters because, as Gorard and colleagues found in their review of evidence-use in education, most studies in the field have not shown positive results, and it is not always clear whether influential studies are high quality or how their findings should be implemented in schools. Poorly judged implementation, the reviewers warned, can lead to wasted opportunities and even harm. The Southampton redesign therefore focused on building two of Breckon and Dodson’s mechanisms: awareness, meaning positive attitudes towards evidence use, and skills, meaning the ability to access and make sense of evidence.

The programme under revision leads to a postgraduate qualification for secondary mathematics teachers and contains three Master’s level assignments, two of which are specific to mathematics education. The first asks trainees to critically evaluate a mathematics education topic, choosing between mathematical resilience or mathematical misconceptions; the second requires them to design a small research project on that topic, collect empirical data, analyse it and draw conclusions. Support for both assignments was remapped onto the expansive learning cycle, with technology use highlighted at every phase. The aim was not technology for its own sake but technology as a tool for critical evidence-use, which the researchers described as a means to an end.

The questioning phase is where the redesign was most inventive. To ensure all trainees started from a comparable knowledge base, the tutors compiled curated literature lists on the two topics and produced in-depth annotated PDF versions of key articles, complete with critical commentary on conceptual and methodological aspects, alongside screencast video presentations. About a month later, the tutors deliberately challenged that baseline. For the resilience topic, they introduced a psychological article arguing that the claims underpinning Mindset Theory appear stronger than the evidence supporting them, presenting it through an annotated PDF and a screencast recorded by the first author. For misconceptions, they exploited a genuine tension in the literature: one body of work argues that errors should be kept out of instruction, while other studies contend that explicitly including them helps learning. This engineered contradiction mirrors Engeström’s idea that expansive learning begins by criticising or rejecting aspects of accepted practice and existing wisdom.

Using an educational design research approach, the team refined the programme through formative evaluation, expert review within the wider project, and classroom implementation with a cohort of 23 secondary mathematics trainees in 2020-21 and another of around 21 in 2021-22. The researchers then conducted five one-hour voluntary interviews with trainees, transcribed the recordings and coded them through thematic analysis, focusing only on passages about technology. Three themes emerged. The first concerned technology platforms as sources of evidence. Trainees described relying on news websites such as BBC News alongside magazines and research articles, and several stressed the need to consult multiple sources to get, as one put it, a picture of the potential truth, particularly when reports said slightly different things or failed to reference their sources.

The trainees were equally clear about which sources they distrusted. Social media came in for sustained criticism, with one interviewee saying she would not trust anything that came up on platforms like Facebook as evidence, and another describing social media as a channel for potentially harmful misinformation that needs critical challenge. One trainee remarked that she now took everything on the news with a pinch of salt because reports said one thing one day and something different the next. The interviews also revealed a temporal dimension to the problem: recent developments in video editing mean a video can be tampered with, further complicating trust in digital evidence. Trainees preferred documented or hard evidence, such as emails and quantitative data, but acknowledged that qualitative information like student opinions also holds value, even if it is more difficult to measure.

The second theme, challenging evidence, surfaced a tension between critical ideals and classroom practicality. Trainees recognised the need to engage critically with evidence as part of their Master’s level work, but questioned whether complex methods such as large-scale surveys could translate into daily teaching, especially given the time constraints of a busy profession that leaves little room to read everything or conduct proper research. Views on the programme’s two assignments diverged in instructive ways. One trainee said she knew nothing about mathematical anxiety or resilience when she started the first assignment and counted its screencast-supported critical evaluation among the things she had taken from the whole year, while suspecting the more practical data-collection assignment would stay with her less. Another trainee valued collecting his own evidence through small-scale projects, saying it helped him back up what he was doing against external directives. The third theme was more instrumental: trainees described spreadsheets for visualising data, surveys and polls for gathering student feedback, school information systems tracking marks, behaviour and homework, and electronic lesson plans and observation notes serving as evidence of professional practice.

The study’s most forward-looking element is its next redesign, built around generative AI. The authors stress that the human tutors, with decades of teaching experience and deep knowledge of the mathematics education literature, remain an essential ingredient that cannot be understated, but no teacher can have such a knowledgeable other on call throughout a career. As a proof of concept, the team took an evidence grid of critical questions used in Master’s level modules and fed it, together with the mindset-critique article, into NotebookLM, a tool powered by Google’s Gemini large language model. The result was an interactive critical dialogue: the agent could summarise sources, explain the rationale for a study on request, shorten answers to fifty words, identify the article’s biggest limitation, namely that its results mainly apply to university students, and generate study guides, briefing documents and even audio podcasts. The researchers see potential for genAI to support Socratic dialogue between trainee and machine, while cautioning that no data on such use yet exist.

The authors are candid about the study’s limits and about technology’s double edge. With only five interviews from a single country, generalisability is constrained, though the findings align with the wider four-country project. They warn that pre-service teachers are novices who benefit from automaticity in classroom routines, which reduces cognitive load, yet habit formation can also limit growth in teacher effectiveness, so critical thinking about evidence must remain on the radar across a teacher’s whole career rather than only during training. They call for proper field experiments comparing their approach with business as usual. Most pointedly, they caution that technology is no silver bullet: after decades of perennial promise, its effectiveness depends on context, and recent research suggests that cognitive offloading onto AI tools may even erode critical thinking. The lesson, they argue, is neither to dismiss the technology nor to embrace it uncritically, but to weigh its strengths and limitations, which is precisely the questioning stance the programme tries to instil.

Subject of Research: Technology-supported critical evidence-use in pre-service secondary mathematics teacher education

Article Title: Using technology for questioning evidence in pre-service teacher education

Article References: Using technology for questioning evidence in pre-service teacher education. (n.d.). https://doi.org/10.1007/s44322-025-00045-w

Image Credits: AI Generated

DOI: 10.1007/s44322-025-00045-w

Keywords: pre-service teacher education, evidence-informed teaching, expansive learning cycle, generative AI, mathematics education, educational design research, digital literacy, critical thinking, teacher training, evidence-use mechanisms, technology, questioning