Dacia Toll is co-founder and co-CEO of Coursemojo, which embeds AI support for students and teachers in top-rated ELA curricula (HQIM). Mojo transforms the assignments into interactive, differentiated activities for students and real-time data insights for teachers, including every student’s understanding of every question and trends in classwide misconceptions. Mojo-supported schools achieved 5-10% point gains on state ELA tests with greater gains for students with IEPs and multilingual learners. Coursemojo is one of only two AI reading products that have multiple independent research studies that meet the Tier 2 ESSA standard for evidence of impact.
Over the last three years, a tsunami of AI tools has entered schools, with more arriving every day. School districts know these tools carry both promise and peril. Making smart guidance and guardrail decisions requires scrutinizing each tool’s design: does it enhance or replace the thinking required for learning?
When most people think of AI, they picture ChatGPT, Claude, or Gemini – Large Language Models (LLMs) many of us use daily for answers or assistance. Erin Mote, CEO of InnovateEDU, calls these “indulge-bots.” They tell us we ask great questions, then confidently answer – mostly, but not always, correctly.
Unfortunately, early evidence suggests giving students direct access to these answer-bots has a net negative impact on learning. Some students use AI well by asking it to quiz them or to help them dig deeper into an area of interest. Too many others use it to do the assignment for them: what experts call “outsourcing” or “automating” thinking.
For example, a recent MIT study found that participants using an LLM during essay writing exhibited the weakest brain connectivity of the three groups studied, reported the lowest sense of ownership over their essays, and struggled to accurately quote their work.
The good news: a growing number of purpose-built AI education tools do not allow open LLM access. Instead, they use AI like electricity to power Socratic learning experiences grounded in how the brain learns and in strong teaching practice.
How can districts differentiate answer-bots from high-quality, purpose-built AI? Drawing on key principles in Deans for Impact’s The Science of Learning, here are six cognitive science insights that districts can use to ensure students get the outcomes and experience they need and deserve.
- Learning Requires Productive Struggle
Students don’t learn by simply hearing new information. Learning happens when they actively process, retrieve, explain, and apply it – the mental work that builds neural pathways and moves new learning into long-term memory. That’s why productive learning often feels hard: difficulty is an essential part of the process, not a side effect.
If AI simply hands students the answer, re-levels the text, or summarizes the reading, it may boost efficiency – but it also strips out the cognitive work that drives learning. Bellwether’s research on productive struggle backs this up: students learn best by working through challenge with the right amount of support.
Great teachers have always understood this balance. When students struggle, they ask another question, redirect students back to the text, and prompt them to try again with a different approach. Educational AI should do the same by supporting and directing students to keep digging and achieve the key insight on their own.
- Knowledge Matters
Some prognosticators claim AI makes knowledge obsolete since it’s now at your fingertips. That claim misunderstands cognitive science. We learn by connecting new knowledge to what we already know – information with nothing to anchor to simply doesn’t stick. New understanding is built on the foundation of prior knowledge.
This is why knowledge-rich curricula matter so much: by systematically building students’ knowledge and vocabulary, they give students the foundation they need to tackle the next level of complex ideas and texts. Educational AI should reinforce that process by deepening students’ engagement with curriculum content, not replacing it with disconnected learning.
- Each Student Must Think for Themselves
AI’s greatest opportunity in the classroom isn’t replacing instruction – it’s giving every student more chances to engage in it.
Research shows that when teachers ask a question to the whole class, it doesn’t guarantee every student meaningfully engages with the content. When a teacher calls on only a handful of students, engagement becomes optional for everyone else – they can sit quietly, look attentive, and still drift in and out of focus. Students only learn when they do the cognitive work themselves: taking in new information, retrieving background knowledge, reaching a conclusion, and explaining their thinking.
This is where thoughtfully designed AI can complement excellent teaching. A teacher can’t engage one-on-one with 26 students at once, but purpose-built AI can give every student a shot at the teacher’s good questions, provide real-time feedback, and give every student a chance to sharpen their thinking before contributing to the group. The technology doesn’t replace discussion – it catalyzes it by helping prepare each student to contribute.
- Feedback Must Lead to Revision
Feedback alone doesn’t improve learning. In fact, without the chance to act on it and revise, feedback actually has almost no cognitive benefit. Practice doesn’t make perfect; practice makes permanent. That’s why students must use feedback to revise their thinking and writing, so they encode success rather than misconceptions.
Effective feedback is specific and points to what to do next, not just whether an answer is right. Revising while still engaged with a task gives students another chance to deepen understanding. The goal isn’t the right answer, but what they learn getting there. This matters most in writing, where revision helps you sharpen your comprehension and conclusion. Ideally, a teacher would give every student specific feedback, have them revise, and repeat that cycle multiple times per class – but that’s humanly impossible.
This is where purpose-built AI makes a real difference. Grounded in the assignment and curriculum-specific rubrics, it can deliver that precise feedback so students revise and move forward – while the teacher gets real-time visibility into individual progress and class trends, and steps in whenever needed.
- Students Must Become Independent Learners
The goal of educational AI isn’t to make learning effortless – it’s to make students more capable, confident, and independent.
Research shows that students develop deeper understanding when they learn to monitor their own thinking, reflect on their progress and persist through challenging work rather than expecting immediate answers. Self-reflection should be baked into the learning process.
Educational AI should reinforce these habits. It should encourage them to think strategically, continue working through challenges, and then reflect on their learning and the process. Success should be measured not by how quickly students reach the correct answer, but by whether they become more effective, independent thinkers over time.
- Independent Learning Still Requires Interaction
Becoming an independent thinker doesn’t mean learning in isolation. Students deepen their understanding by articulating their thinking, hearing different perspectives, and working through ideas with others. Research from the Education Endowment Foundation shows collaborative learning boosts student achievement, noting: “It is important to ensure that all pupils talk and articulate their thinking in collaborative tasks to ensure they benefit fully.”
Educational AI should enhance, not replace, these interactions. The last generation of edtech often meant students working alone on screens. The next generation of AI-powered products should create more opportunities for students to think and work together, while giving teachers real-time insight to guide full-class discussions.
From AI Features to Learning Design
As AI moves from experimentation to everyday classroom practice, schools need to get sharper at evaluating educational technology.
Bellwether’s Built for Learning report highlights that the most promising instructional AI tools are intentionally designed around learning science: they preserve productive struggle, encourage revision, and keep students engaged in grade-level thinking rather than replacing it. That’s an important shift, moving the conversation beyond features and efficiency toward instructional design and impact.
In this time of tight budgets and competing priorities, here’s the test for any new AI tool in front of your district: does it power more and better thinking for more students? Get that question right, and AI becomes a skilled teaching assistant for every teacher. Get it wrong, and you’ve just bought a more expensive way to do less learning.