Will Ellis: EdTech Accountability Framework

Introduction

English schools spend approximately £900 million every year on educational technology. No statutory requirement exists for any product within that spend to demonstrate measurable educational benefit over the non-digital alternative before procurement.

Since the widespread adoption of classroom devices over the past fifteen years, a prevailing narrative held that technology would prepare children for a digital world, personalise learning, improve assessment, and free teachers for the relational and intellectual work that technology cannot replicate. Some of that was true, in specific contexts. But any assessment of a platform’s educational value must be understood within the market dynamics of the EdTech industry, where platforms that maximise engagement and return visits generate stronger commercial returns than platforms that produce durable learning gains. The most widely used platforms in English schools operate within those dynamics.

Independent research now provides a sufficient basis to evaluate which platforms serve learning and which undermine it. For many of the most widely deployed platforms across English schools, the negative effects on learning relate closely to their engagement-optimised architecture. Pupil-facing AI raises architectural concerns serious enough to warrant prohibition regardless of supervision level or claimed use case, as the evidence tables set out.

Many of these platforms collect children’s data, track their behaviour, and operate within a commercial model that rewards engagement over learning. That is what makes the evidential standard this framework applies proportionate rather than prejudicial.

This framework is pro-learning. It recognises that many platforms have a place within a child-centred, evidence-based curriculum. This document identifies what passes, what does not, and the reasoning behind its distinctions.

Standards

Two Rules

Two Tests

Two prior compliance tests apply before any platform reaches the pedagogical question. Failure on either is grounds for non-deployment.

Two Grounds for Classification

A platform may be prohibited on either ground independently of the other.

Prohibited

The following are prohibited regardless of supervision level, claimed use case, or age group.

Applying This Framework

Apply the following steps in order. Stop at the first failure. A tool that fails any step should not be used, regardless of claimed educational benefit.

Notes

The two rules are policy positions grounded in safeguarding rationale, not learning outcomes evidence. The relevant question is not what filtering software can achieve at home but where responsibility sits. The no digitised homework rule also recognises that a child whose only internet access is a school-issued device is not less deserving of protection. A school cannot discharge safeguarding responsibility for an environment it does not control.

YouTube is a useful example of why this distinction matters. A teacher selecting a specific video and playing it to the class on a shared screen passes the instrument test: the child watches, the tool does not respond to the child. That specific use is conditional. But the surrounding architecture is prohibited: YouTube’s recommendation algorithm is engagement-optimised and profiling-based, and cannot be meaningfully disabled on a standard school account. Any individual pupil access to YouTube on a device fails the architectural test entirely. YouTube is therefore conditional in theory and prohibited in practice for most classroom settings, because the conditions required to make it safe cannot be reliably met.

Oversight does not make a prohibited tool acceptable. A conversational AI system supervised by a teacher is still a conversational AI system. Supervision matters for permitted tools. It cannot rehabilitate prohibited ones.

Efficiency is not a justification for technology use. If a tool saves teacher time but reduces the quality of a child’s learning experience, the tool should not be used. The question is never whether technology makes a task faster or easier. The question is whether it makes learning deeper.

SEND and assistive technology. Compensatory tools for pupils with identified needs (text-to-speech, speech-to-text, AAC devices) operate under different principles. The distinguishing principle is this: a compensatory tool addresses a barrier to access that prevents a child from engaging with learning at all; it does not substitute for a cognitive process the child is capable of performing. A speech-to-text tool for a child with dysgraphia enables written output the child cannot otherwise produce. An AI writing assistant for a child without that barrier removes the cognitive labour of composition that constitutes the learning. These are not equivalent uses of responsive technology. The former is assessed separately using professional judgment and appropriate specialist input, and is not subject to the blanket prohibitions above.

Reference Index

Source: Will Ellis / Reclaim Childhood – reclaimchildhoodmedia.substack.com