Ask a group of primary teachers whether they have used generative AI in the past year, and the honest answer from most is: yes, occasionally, mostly for planning, and with varying degrees of confidence. That is broadly what I found in this small-scale survey, which asked 45 primary educators about their use of AI tools in science teaching.
The responses, which came mainly from trainee teachers alongside classroom teachers, science subject leads and a small number of senior leaders, offer a snapshot of where things currently stand. What they describe is a profession that is finding AI genuinely useful, but that is largely making it up as it goes along, without much institutional support to fall back on.
What teachers are actually doing
ChatGPT is by some distance the most widely used tool, cited by 39 of the 45 respondents. Microsoft Copilot came second at 23, followed by Google Gemini and TeachMate AI both on 13. Very few respondents had used tools such as Perplexity, Claude or MagicSchool, suggesting that the typical teacher’s AI experience is fairly narrow.
That is not necessarily a problem. Teachers are busy, and reaching for a familiar tool is entirely reasonable. But it is worth noting that general-purpose AI models were not built with primary science in mind. They do not know your scheme of work, your year group’s prior knowledge, or the specific vocabulary you want pupils to be using. Subject-specific prompting can compensate for this to a degree, but that skill takes time to develop.
In terms of frequency, most respondents are using AI occasionally rather than as a daily habit. Fifteen said they use it one to three times a month, ten said one to three times a week, and just four said daily or almost daily. Nine said they never use it for science teaching at all. This mirrors broader patterns. Research from the Tony Blair Institute found that only around one in four UK teachers are using AI daily, with use largely limited to a narrow set of tasks, most commonly generating lesson materials and supporting curriculum planning (Crowley-Carbery and Johnson, 2025).
Within science specifically, the most common uses reported in this survey were planning lessons or sequences (16 respondents), creating quiz or retrieval practice questions (14), differentiating resources (12), designing or adapting practical investigations (11), and creating worksheets (11). Fourteen respondents said they had not used AI for science tasks at all, which is a useful reminder that adoption is still far from universal, even among those who responded to a survey on the topic.
One figure that stands out is that ten respondents had used AI to check or extend their own science subject knowledge. This is a practical use case that rarely features in discussions about AI in education, which tend to focus either on lesson planning or on pupil-facing applications. For a generalist primary teacher who may not feel especially confident with forces, sound or electricity, being able to ask a quick question and get a clear explanation is genuinely useful. It is not a substitute for sustained subject knowledge development, but as a low-stakes starting point, it has real value. This does of course run the risk of the AI hallucinating information. It would be better to check more authoritative sources of information.
What Trainee Teachers are doing with AI
The trainee teacher data deserves particular attention from ITE providers. Of the 26 trainee respondents, 73% said they had used AI to summarise course reading. That is a significant figure, and it raises a reasonable question about depth of engagement: a summary generated by an AI model is shaped by what the model prioritises, which may not align with what a tutor intended the trainee to take from a text, or with the nuance that comes from reading primary sources directly. Similarly, 42% reported using AI for writing reflections. It is not clear from the survey whether this means generating a reflection wholesale or using AI to structure and develop an outline, and that distinction matters considerably. Reflective writing in ITE is not simply a task to complete; it is a professional learning process, and if AI is doing the cognitive work rather than supporting it, something important is being short-circuited. The 20% using AI to write their assignments should also raise some concerns. None of these findings is a reason to prohibit AI use in initial teacher training, but they do suggest that providers need to be explicit about where and how AI is appropriate within their programmes, rather than leaving trainees to work that out for themselves.
The workload question
Thirty-seven of the 40 respondents who use AI said it has either slightly or significantly reduced their workload. That is consistent with what larger surveys have found. The DfE’s own guidance on generative AI acknowledges that AI has demonstrated it can help the education workforce by reducing some of the administrative burdens that hard-working teachers face in their day-to-day roles (DFE 2025). The honest caveat is that time savings at the generation stage can be offset by time spent checking, editing and adapting output. Several respondents in this survey flagged that as a consideration, and it is borne out by how people are actually behaving.
Accuracy: a real concern, not just a theoretical one
Thirty-three of the 45 respondents said they always check or edit AI-generated content before using it with pupils. That is encouraging. Six said they often do so, and four said sometimes. The concern is not that teachers are being careless; it is that not everyone knows what they are looking for when they check.
Six respondents had already identified scientific inaccuracies or misconceptions in AI-generated content. A further 17 said they were not sure whether they had encountered any. That second group is worth pausing on. If a teacher does not spot an error, they will not count it as an error, but the misconception may still reach pupils. Inaccurate or misleading science content was the single most widely cited concern in the survey, selected by 29 respondents. Over-reliance on AI came second, followed closely by worries about age-appropriateness for primary pupils. An AI-generated story for a lesson hook is relatively low risk. An AI-generated explanation of what happens to particles when a substance changes state is another matter entirely and deserves closer scrutiny. Knowing which parts of the output need the most careful checking is a skill that develops with experience, but it can also be taught explicitly.
The policy gap
One of the clearest findings from this survey is how patchy the institutional picture is. Only nine respondents said their school has a clear, well-communicated policy on staff use of generative AI. Sixteen said there was no policy yet, ten said any policy that existed was not very clear, and a further ten were not sure whether a policy existed at all. This is not unique to the schools in this sample. A 2025 study by Doss et al. found that across their schools, professional development for teachers, training for students on how to use AI, and school and district policies all lag behind the rapid increase in actual use. In the UK context, as of late 2025, many educators reported that their institutions had minimal oversight of how staff were using generative AI, with teachers left to experiment independently (DFE 2025). The DfE’s guidance exists, but it is deliberately non-prescriptive, which means that schools are largely left to work out the details themselves. For individual teachers, working without a clear policy creates uncertainty about what is and is not appropriate. It can also mean that good practice is not shared, and that teachers who are more cautious or less confident are left without the reassurance they need.
Training: most have had little or none
The training picture is stark. Almost half of respondents (19) had received no training or guidance on using generative AI in education at all. Fifteen had received only a brief overview, and six had received in-depth training. Five had pursued self-directed learning only.
Pearson’s School Report 2025, which gathered views from more than 14,000 people across the UK education sector, found that almost a quarter of teachers said they were not confident using AI, and only 9% felt confident teaching it. In response, 42% said AI should be included in teacher training. A separate report for the National Literacy Trust found that three in four teachers said they needed more training, support and resources to use generative AI tools effectively (Picton and Clark, 2024).
What teachers in this survey said they actually want is fairly specific. Practical examples of effective use in primary science came top, selected by 28 respondents. Subject-specific prompts and templates for science topics came second at 24. Training on checking the accuracy and bias of AI outputs was chosen by 20 respondents, followed by time to experiment and collaborate with colleagues (17), clear school or MAT policies (15), and guidance on safeguarding and data protection (15).
What schools could do
The requests teachers are making are not especially difficult to act on, and they do not require a school to have resolved every policy question before starting.
The most practical starting point is sharing examples of good practice. A science subject lead who has used AI to generate quiz questions for a particular topic, or to produce differentiated versions of a resource, could share those examples with colleagues alongside a brief note about what they changed or removed. That kind of peer-led approach is less intimidating than formal training, and it builds a shared understanding of both the possibilities and the pitfalls.
Prompt templates are similarly accessible. A short bank of tested prompts, designed specifically for primary science tasks, saves time and helps staff get more consistent results. They can be built collaboratively, updated as people find what works, and shared across a staff team or a MAT without significant cost. See this post for examples of prompts you can use.
A simple fix – spend five minutes in every staff meeting sharing useful ways you have used AI this week. Also share the ways it went wrong – as examples to avoid!
Training on the use of AI is arguably the most important gap to address. The Frontiers in Education review of generative AI in K-12 settings found that nearly 90% of reviewed papers emphasise the need for teachers’ professional development for AI-integrated classrooms, specifically highlighting the importance of developing subject-specific AI literacy through authentic learning tasks (Alfarwan, 2025). For science, that means helping teachers understand where AI tends to get things wrong: oversimplified models, conflated concepts, imprecise language, or outdated information. These are not random errors; they have patterns, and knowing the patterns makes checking faster and more reliable. ScienceFix can help with this – contact us for our AI training options.
Finally, a clear AI policy, even a brief one, removes a lot of low-level uncertainty. It does not need to cover every conceivable scenario. It just needs to say what is permitted, what is not, who to ask, and what to do if AI-generated content turns out to be inaccurate. The DfE’s own guidance is clear that any content produced by AI requires critical judgement to check for appropriateness and accuracy, and that the final responsibility rests with the teacher and their school. (DFE 2025). A school policy that reflects that principle in plain language gives teachers something to work from. For guidance on how to write an AI policy, see this post on the Whiteboard Blog.
A practical tool, not an automatic one
The teachers in this survey are, on the whole, approaching AI with reasonable caution. Most are checking what they use. Most find it useful. Most want more support to use it well, not less. What they are describing is a tool that works best when the person using it knows their subject, knows their pupils, and knows what questions to ask of the output.
For primary science, that last point matters more than it might in some other contexts. Science is cumulative, and misconceptions formed in Key Stage 1 or 2 can be surprisingly persistent. An AI model that confidently explains something incorrectly, or uses language that is technically imprecise, is not a helpful starting point for teaching. Used well, with appropriate checking and a clear sense of where the tool is likely to fall short, generative AI can save time and open up new possibilities. Getting there requires support that, for most teachers in this survey, has not yet arrived.
Note on the survey: this survey was carried out in early 2026 and received 45 responses from primary educators in England, with a significant proportion of trainee teachers among the respondents. It should be read as indicative rather than representative.
Alfarwan, A. (2025) ‘Generative AI use in K-12 education: a systematic review’, Frontiers in Education, 10, article 1647573. Available at: https://doi.org/10.3389/feduc.2025.1647573 (Accessed: 31 March 2026).
Crowley-Carbery, K. and Johnson, R. (2025) Generation ready: building the foundations for AI-proficient education in England’s schools. London: Tony Blair Institute for Global Change. Published 1 September 2025. Available at: https://institute.global/insights/public-services/generation-ready-building-the-foundations-for-ai-proficient-education-in-englands-schools (Accessed: 31 March 2026).
Department for Education (2025) Generative artificial intelligence (AI) in education. Policy paper. London: DfE. First published 29 March 2023, last updated 12 August 2025. Available at: https://www.gov.uk/government/publications/generative-artificial-intelligence-in-education (Accessed: 31 March 2026).
Doss, C.J., Bozick, R., Schwartz, H.L., Chu, L., Rainey, L.R., Woo, A., Reich, J. and Dukes, J. (2025) AI use in schools is quickly increasing but guidance lags behind: findings from the RAND survey panels. Santa Monica, CA: RAND Corporation. Available at: https://www.rand.org/pubs/research_reports/RRA4180-1.html (Accessed: 31 March 2026).
Pearson School Report (2025) Learning for Life Available at: https://www.pearson.com/en-gb/schools/insights-and-events/topics/school-and-college-report/2025.html (Accessed 31 March 2026)
Picton, I. and Clark, C. (2024) Children, young people and teachers’ use of generative AI to support literacy in 2024. London: National Literacy Trust. Available at: https://literacytrust.org.uk/research-services/research-reports/children-young-people-and-teachers-use-of-generative-ai-to-support-literacy-in-2024/ (Accessed: 31 March 2026).
The post How Primary Science Teachers Are Using Generative AI: Findings appeared first on Danny Nics Science Fix. Written by Danny Nicholson