Explore Learning warns: Weak evidence in AI tools could risk student development
AI is moving into classrooms fast. A new white paper from Explore Learning says the rush may be putting student outcomes at risk if tools aren't backed by solid evidence and sound educational theory.
The message is simple: personalization is promising, but poorly designed systems can weaken critical thinking, dent confidence, and create overreliance on automated prompts. As governments, schools, and vendors scale AI across teaching, assessment, and support, standards need to rise-quickly.
Evidence and assessment come first
Personalization only works if it's anchored in accurate, ongoing assessment-not just surface-level performance metrics. The report warns against "metric fixation," where progress is reduced to simplified outputs instead of deeper learning.
"The UK's education system is under greater pressure than ever, and AI has significant potential to alleviate these challenges, but only when backed by strong evidence and proven to improve outcomes, with the same rigour we expect of any educational intervention," says Lisa Haycox, CEO of Explore Learning.
Avoid false signals of progress
Tools without rigorous testing can create a "mirage of false mastery." Short-term gains look good on a dashboard, then disappear the moment the tool is removed.
That's the risk when systems optimize for what's easy to measure rather than what matters for long-term retention, transfer, and reasoning.
Potential support for SEND learners
There is upside. Explore Learning reports a 35 percent increase in SEND students accessing its tuition services between 2024 and 2025. Evidence-based AI can help identify learning needs earlier and adapt tasks to each learner's profile.
Used well, these tools can scale targeted scaffolds, reduce cognitive load, and create timely feedback without replacing professional judgment.
Explore Learning's approach
"At Explore Learning, the question has never been whether to use AI, but how," says Dr Hisham Ihshaish, Head of Data and AI. "Our technology is grounded in established learning theories and informed by 25 years of longitudinal learner data."
Recent updates to Compass 2.0 aim to model not just what a student knows, but how they learn and the pace at which they develop-recalibrating scaffolding in real time.
Human educators stay central
AI can support foundational skills and routine feedback, but it cannot replace teacher judgment. Human oversight is still essential to interpret progress, ensure fairness, and correct algorithmic blind spots.
As Haycox puts it: "We cannot forfeit children's futures for hype, and we should continue to encourage healthy debate to guard against this, while embracing the full potential of transformative technology."
What school leaders and teachers can do now
- Demand evidence. Ask vendors for peer-reviewed studies, RCTs, or at minimum transparent pre/post analyses with clear baselines. See the EEF guidance on digital technology.
- Prioritize formative assessment. Choose tools that diagnose misconceptions, adapt tasks, and make thinking visible-not just those that boost quick scores.
- Guard against metric fixation. Look for measures of transfer, depth of understanding, and student explanation quality, not only accuracy and speed.
- Pilot small before scaling. Set clear success criteria, run time-bound trials, collect teacher and student feedback, and compare against a control.
- Keep teachers in the loop. Ensure teachers can review, override, and tailor recommendations. No black boxes.
- Support SEND at the core. Check for adjustable reading levels, multimodal input/output, and early flags for possible learning needs.
- Protect privacy and equity. Review data policies, bias testing, and accessibility standards upfront.
- Invest in staff training. Build capability so tools are used well. A practical starting point: AI Learning Path for Teachers.
Bottom line
AI can help, but only if it is accountable to learning science and classroom reality. Evidence before enthusiasm. Pilot before purchase. Human judgment over dashboards.
For system-wide guidance, see UNESCO guidance on generative AI in education. The opportunity is real-so are the risks. Choose like outcomes depend on it, because they do.
Your membership also unlocks:
