GenAI in Education: What the DfE Requires — and How Willow Delivers It
A briefing for school leaders, headteachers, and MAT executives
Part One: The Reality of AI in Your School Right Now
Whether schools are ready or not, generative AI has arrived. Students are using it. Teachers are using it. In many cases, it is being used right now — on devices in your pupils’ pockets, in free time, in homework sessions, and increasingly in the classroom itself. The question for school leaders is no longer whether AI is present in your school. It is whether the AI your students are accessing is safe, pedagogically sound, and aligned to their learning.
This section draws on the OECD Digital Education Outlook 2026 — the most comprehensive international review of generative AI in education published to date — to give you a clear picture of the landscape your school is operating in.
1.1 Student Use Is Already Widespread
The OECD’s 2026 report is unambiguous: student use of generative AI is not a future concern. It is a present reality, growing across every year group.
| Metric | Finding |
|---|---|
| GenAI use in lower secondary | Moderate and growing rapidly |
| Upper secondary students | Majority use GenAI regularly |
| Lower secondary teachers who used AI in their work (2024) | 37% (OECD TALIS) |
The OECD is clear that use starts small in primary school, grows in lower secondary, and reaches majority usage by upper secondary. Perhaps more striking: the tools students are using are largely consumer-facing, unmoderated, and not designed for learning. ChatGPT, Google Gemini, Microsoft Copilot, and even Snapchat’s embedded AI are all freely accessible to young people without parental consent, age verification, or any pedagogical structure.
This is not a niche behaviour. A seven-country European survey cited in the OECD report found that 31% of students report using AI to provide complete solutions to tasks — in other words, to do the work for them. Only 20% use it for anything that resembles self-regulated learning.
1.2 The Performance vs Learning Gap
Here is the uncomfortable truth at the heart of the OECD’s findings, and one that every school leader should understand:
“GenAI systems may enhance the apparent quality of student work — that is, their performance at educational tasks — without improving their actual learning — that is, their knowledge and skill acquisition.”
— OECD Digital Education Outlook 2026
The OECD reviewed multiple studies where students using general-purpose AI tools produced higher-quality outputs than their peers — but that advantage disappeared, and sometimes reversed, when AI access was removed in exams. One striking neuroscience study across five universities found that within an hour of completing a task, 89% of students who wrote their own work could quote something from it. In the group that used general-purpose AI, only 12% could recall anything from the text they had submitted.
Their essays were rated highly. They had learned very little.
The OECD calls this ‘metacognitive laziness’ — when students outsource the cognitive effort to AI, they lose the deep processing that drives durable understanding. Offloading thinking to a chatbot is not learning. It is the appearance of learning.
| Statistic | Source |
|---|---|
| 72% of lower secondary teachers are concerned AI lets students pass off work as their own | OECD TALIS 2024 |
| 31% of students in European survey use AI to provide complete solutions to tasks | OECD, 2026 |
| 12% of GenAI-assisted students could recall content from their own work an hour later | OECD, 2026 |
1.3 Why Schools Need to Act — Schoolwide
The OECD is direct on what must happen next. Governments and school leaders should not attempt to ban AI — the ship has sailed. Instead, they should ensure students have access to AI tools that are:
- Pedagogically-first in their design — built to support learning, not just task completion
- Protective against cognitive offloading — prompting students to think, not bypassing thinking altogether
- Personalised to the learner’s level — adapting to each student’s needs and gaps
- Safe and age-appropriate — with robust content filtering, safeguarding protocols, and data protection
- Curriculum-aligned — reinforcing what teachers have already introduced in class
A single teacher adopting a good AI tool is a start. But it is not enough. To meaningfully shape how students in your school interact with AI — and to provide consistent safeguarding, data protection, and learning quality — schools need a whole-school approach. That means a tool that works for every classroom, every year group, every teacher, and every student.
That is precisely what Willow Learn is built for.
Part Two: The DfE’s Generative AI Product Safety Standards
In January 2026, the Department for Education published its Generative AI Product Safety Standards — a detailed framework setting out the minimum requirements for any AI product deployed in an English school or college. These standards apply to edtech suppliers and, crucially, give school leaders a clear checklist for evaluating the tools they procure.
In this section, we map every standard from the DfE guidance against Willow’s features and design principles. We include the DfE’s own language so you can see, line by line, exactly where Willow stands.
How to use this section: Each sub-section presents the DfE’s requirement in italics, followed by a clear explanation of how Willow addresses it. Use this document when evaluating Willow for procurement, presenting to governors, or building your school’s AI policy.
2.1 Stated Purpose
“Generative AI products should clearly state their intended purpose and use cases, including target demographic and learning focus. Suppliers should not exaggerate the impact or capabilities of their tools. Any claims should be supported by robust and transparent evidence.”
Willow’s response: Willow Learn is designed exclusively for primary and secondary education in the UK. Its purpose is unambiguous: to reduce teacher workload through intelligent, curriculum-aligned activity generation, and to provide students with personalised, adaptive learning conversations that reinforce classroom teaching. Willow does not claim to replace teachers or guarantee exam outcomes. All capability claims are grounded in user evidence and pedagogical research.
2.2 Educational Use Cases
“Developers and suppliers should indicate the intended educational use cases of their product.”
Willow addresses four of the DfE’s defined use cases simultaneously:
- Personalised learning and accessibility: Willow adapts conversations and activities to each student’s level, pace and learning gaps.
- Assessment and analytics: Willow provides teachers with detailed insights on student progress, understanding and ability — going far beyond what standard assessments reveal.
- Digital assistant: Willow acts as an AI-powered learning companion, guiding students through tasks and prompting thinking — never simply delivering answers.
- Learner engagement and interaction: Willow creates interactive, adaptive learning conversations that keep students actively engaged with curriculum content.
“Willow isn’t just a tool — it’s changed how my students engage with their learning outside the classroom. I can see exactly where they’re struggling and what they’re getting right, in real time.”
— Secondary school teacher, Willow Learn user
2.3 Filtering
“Generative AI products must effectively and reliably prevent users from accessing harmful or inappropriate content. Filtering mechanisms should be embedded within products. Filtering will be adjusted based on different levels of risk, age, appropriateness and the user’s needs.”
Willow’s response: Content filtering is built into Willow at the architecture level — not bolted on as an afterthought. All student interactions are bounded within curriculum-aligned, age-appropriate contexts. Students cannot use Willow to generate content outside its educational scope. Filtering remains active throughout every conversation, across all devices including BYOD and smartphones, and is calibrated appropriately for the age group of the user.
Willow’s design means students interact within a structured, school-controlled environment — not a general-purpose chatbot. The system understands context throughout a conversation, ensuring that generated content remains appropriate and educationally purposeful.
2.4 Monitoring and Reporting
“Products should identify and alert local supervisors to searches for, or access to, harmful or inappropriate content… identify and alert local supervisors of disclosures that indicate a possible safeguarding issue… track and report when learners offload thinking to the system.”
Willow’s response: Willow’s teacher dashboard gives educators real-time visibility of student progress, engagement and any areas of concern. Teachers receive detailed insights on student activity far beyond what typical assessments reveal. The system is designed to support safeguarding responsibilities by flagging interactions that may require attention. Willow also tracks and surfaces data on student engagement patterns, giving teachers the information they need to intervene early and effectively.
Reporting is built for non-expert users. Insights are presented in plain language, in formats that do not burden busy teachers — because a dashboard that nobody reads helps nobody.
“The insights Willow gives me go way beyond a quiz score. I can see how a student is actually thinking through a problem — and that changes what I do next in the classroom.”
— Primary school teacher, Willow Learn user
2.5 Security
“Products should offer robust protection against ‘jailbreaking’… allow administrators to set different permission levels for different users… ensure regular bug fixes and updates are promptly implemented… have robust password protection or authentication methods.”
Willow’s response: Willow operates within a school-administered environment. Administrators control access levels, user permissions and settings. The platform is built with security-first principles and maintained with regular updates. Unlike consumer AI tools, Willow cannot be repurposed or ‘jailbroken’ by students attempting to bypass its educational scope — because the product’s architecture does not permit open-ended use in the first place. Authentication is robust, and the system is designed to be compatible with the DfE’s Cyber Security Standards for Schools and Colleges.
2.6 Privacy and Data Protection
“Products should not collect, store, share or use personal data for any commercial purposes, including further model training and fine-tuning, without confirmation of appropriate lawful basis… conduct a Data Protection Impact Assessment (DPIA) during the tool’s development and full life cycle.”
Willow’s response: Student data is held securely and never used for commercial purposes, model training, or third-party sharing. Willow is compliant with UK GDPR, the Data Protection Act 2018, and the ICO’s Children’s Code. A DPIA is conducted and maintained across the product’s life cycle. Privacy notices are written in age-appropriate language. Data is stored within appropriate jurisdictions with the appropriate safeguards in place. Schools retain control of their data, and Willow’s data practices are transparent and auditable.
2.7 Intellectual Property
“The product must not store, collect or use intellectual property created by learners or teachers for any commercial purposes, such as training or fine-tuning of models, unless consented to by the copyright owner.”
Willow’s response: Student and teacher inputs into Willow are not used for model training, product development, or any commercial purpose. The intellectual property of students — who are minors — is protected without exception. Schools and teachers retain full ownership of the content they create within the platform.
2.8 Design and Testing
“The product must prioritise transparency and children’s safety in its design… sufficient testing with a diverse and realistic range of potential users and use cases is completed… the product should consistently perform as intended.”
Willow’s response: Willow is purpose-built for education, not adapted from a general-purpose AI tool. Its design process involves educators, and the product has been developed and refined with real teachers and real students in UK schools. The platform is consistently maintained, tested, and updated. Willow’s single purpose — to support learning — means every design decision is evaluated against that goal.
2.9 Governance
“A clear risk assessment is conducted for every product to assure safety for educational use… a formal complaints mechanism is in place… policies and processes governing AI safety decisions are made available.”
Willow’s response: Willow maintains clear, accessible governance documentation including risk assessments, complaints procedures, and AI safety policies. School leaders and administrators have clear escalation paths. Policies are not buried in dense legalese — they are written to be understood and actioned by the educators who rely on them.
2.10 Cognitive Development
This is the most critical standard — and the one that most consumer AI tools fail.
“Products should not provide final answers, full solutions, or complete worked examples by default… provide responses that follow progressive disclosure of information, starting with hints or partial steps… prompt learners for input before providing answers… track and report when learners offload thinking to the system… detect cognitive offloading actions.”
Willow’s response: This is where Willow is fundamentally different from general-purpose AI. Willow does not give students answers. It gives them the right next question.
Willow’s conversations are designed to scaffold learning progressively. When a student asks for help, Willow prompts them to attempt a step first, explains their current understanding, and provides hints before moving to more detailed guidance. Full solutions are not the default behaviour. Students are challenged to think, not bypassed. This directly addresses the cognitive offloading risk identified by both the OECD and the DfE as the primary danger of poorly designed AI in education.
Willow’s teacher dashboard surfaces engagement data, including patterns that might indicate a student is seeking shortcuts rather than engaging with the learning. This gives teachers early warning — the insight they need, before it becomes a problem.
| Statistic | Source |
|---|---|
| 89% of students who wrote their own work could recall it an hour later | OECD, 2026 |
| 12% of AI-assisted students could recall their work an hour later | OECD, 2026 |
| 1st principle of Willow’s design | Never replace thinking — scaffold it |
“What I love about Willow is that it doesn’t do the work for my students. It makes them do the work. They actually learn — and I can see it happening.”
— Head of Year, secondary school, Willow Learn user
2.11 Emotional and Social Development
“Developers should not anthropomorphise products or create products that imply emotions, consciousness or personhood… products should remind users that AI cannot replace real human relationships… include default time limits on usage… avoid interactions which attempt to artificially extend engagement.”
Willow’s response: Willow uses function-based, task-bounded language throughout. It does not present itself as a friend, companion, or entity with feelings. All prompts and interactions are curriculum-focused and bounded to the learning task. Willow does not engage students in personal conversations or seek to cultivate emotional relationships.
Session structures include appropriate engagement limits. The system is not designed to extend usage for its own sake — it is designed to support learning effectively and then hand students back to their teachers, their classmates, and their own thinking. This is a deliberate design philosophy, not an afterthought.
2.12 Mental Health
“Products should detect signs of learner distress… follow an appropriate pathway when distress is detected, including soft signposting to support and raising a safeguarding flag to the institution’s safeguarding lead… use safe and supportive response language that always directs the learner to human help.”
Willow’s response: Willow’s monitoring capabilities include safeguarding-relevant detection. Where patterns of interaction raise concern — whether related to emotional distress or other safeguarding signals — the appropriate staff are alerted. Willow always directs students towards human support, never positioning itself as a substitute for teachers, family, or professional services. Responses are supportive, non-pathologising, and designed to encourage students to reach out to trusted adults.
2.13 Manipulation
“Products should not use manipulative or persuasive strategies… including sycophancy and flattery, deceiving or misleading the user, applying pressure to socially conform, stimulating negative emotions for motivational purposes… should not exploit users, including designing interactions to prolong use or steering users towards paid options.”
Willow’s response: Willow does not use gamification tricks, emotional manipulation, or artificial urgency to drive engagement. It does not tell students they are brilliant when they haven’t earned it, nor does it create anxiety to motivate action. Any motivational design in Willow is transparent, low-stakes, and educationally justified. Willow is a school tool, not a consumer app optimised for screen time.
Critically, Willow does not blend advertising with pedagogy, steer users towards upselling, or use dark patterns of any kind. Schools purchase Willow as a service. Students use Willow to learn. That is the entirety of the relationship.
Part Three: At-a-Glance Compliance Summary
| DfE Standard | DfE Requirement | How Willow Delivers |
|---|---|---|
| Stated Purpose | Clear statement of purpose, target demographic, and evidence-backed capability claims | Willow clearly states it is designed for UK primary and secondary education. Purpose, use cases, and claims are transparent and evidence-grounded |
| Educational Use Cases | Indicate intended educational use cases from DfE’s defined list | Willow addresses personalised learning, assessment & analytics, digital assistant, and learner engagement — four of the eight DfE categories |
| Filtering | Effectively prevent access to harmful content throughout every conversation, across all devices | Filtering embedded at architecture level, active throughout conversations, calibrated by age group, maintained across BYOD and smartphones |
| Monitoring & Reporting | Alert supervisors to harmful content and safeguarding disclosures; track and report cognitive offloading | Real-time teacher dashboard, safeguarding alerts, engagement analytics including cognitive offloading patterns |
| Security | Robust jailbreak protection, admin permission levels, regular updates, authentication, cyber security compliance | School-administered with role-based permissions, regular security updates, architecture prevents open-ended repurposing, compatible with DfE Cyber Security Standards |
| Privacy & Data | GDPR compliance, no commercial use of data, DPIA conducted and maintained, age-appropriate privacy notices | UK GDPR and ICO Children’s Code compliant, no commercial data use, DPIA maintained, age-appropriate notices, school data ownership |
| Intellectual Property | Student and teacher IP not used for training, product development, or commercial purposes without consent | Zero commercial use of student or teacher inputs. No model training on school content |
| Design & Testing | Child-safety-first design, diverse user testing, consistent performance | Purpose-built for education, developed with UK teachers and students, regularly tested and maintained |
| Governance | Risk assessments, complaints mechanism, accessible AI safety policies | Clear risk assessments, complaints procedures, and AI safety governance documentation available to school leaders |
| Cognitive Development | No full answers by default; progressive disclosure; detect and report cognitive offloading | Core design principle: Willow scaffolds thinking, never replaces it. Progressive hints, not answers. Cognitive offloading tracked and surfaced to teachers |
| Emotional & Social | No anthropomorphism; no artificial engagement extension; time limits; remind users AI cannot replace human relationships | Function-based language, no persona or emotional framing, task-bounded interactions, built-in session limits, no engagement-maximising design |
| Mental Health | Detect distress signals; safeguarding escalation pathway; direct students to human help | Safeguarding-relevant monitoring, DSL alerts, always directs students to human support |
| Manipulation | No sycophancy, deception, social pressure, negative emotional stimulation, or dark patterns | No manipulative design, no advertising, no gamification exploitation. Educationally-justified motivation only |
Conclusion: A Whole-School Choice
The DfE’s standards are not a bureaucratic checklist. They reflect a genuine understanding of the risks that poorly designed AI poses to the young people in your care — risks to their cognitive development, their emotional wellbeing, their privacy, and their long-term learning.
The standards also reflect something encouraging: that educational AI, done properly, can be genuinely transformative. The OECD’s evidence is clear that when AI is designed with pedagogical intent, students learn more, teachers are better supported, and the gap between performance and actual learning narrows.
The challenge for school leaders is not to avoid AI. It is to choose it wisely — and to choose it schoolwide, so that every student benefits from the same safe, effective, curriculum-aligned experience, and every teacher has the insights they need to teach brilliantly.
Willow Learn is built to meet the DfE’s standards in full — and to give every student in your school access to AI that makes them think harder, not less.
If you’d like to see Willow in action, speak to a member of our team or visit willowlearn.com.
WILLOW LEARN | willowlearn.com
Sources: OECD Digital Education Outlook 2026 | DfE Generative AI Product Safety Standards (January 2026)