Smartschool: why AI for school exams is harder than building a chatbot

American startup Smartschool is developing an AI system for preparing school students for the SAT and ACT exams — and its team insists: building educational AI is fundamentally more complex than creating a regular chatbot. The issue is not the power of the language model, but the nature of learning itself as a process.

Why an AI teacher is not a chatbot

Language models excel at one task: quickly finding and reproducing information. You ask a question — you get an answer. For reference queries, this works. For learning — it doesn't.

A good teacher doesn't just provide the correct answer. They diagnose exactly where a specific student has a knowledge gap, explain from the right angle, select calibrated practice, track long-term progress, and adapt each next step based on the student's response. This is an iterative process: each subsequent assignment depends on the result of the previous one, not randomly selected from a general question database.

In the view of Smartschool's founders, most educational AI products simply transfer chatbot logic into an educational context without changing the underlying architecture. The result is predictable: the model answers individual questions well, but does not teach systematically — and therefore does not prepare for the exam.

This problem is not unique to Smartschool. Large EdTech platforms — Khan Academy, Duolingo, Chegg — have already integrated AI, and each encountered the same challenge: users expect not just answers, but progress. Taking a test within the platform is easy; showing consistent growth in real grades — that's a different story.

What makes SAT and ACT a special case

SAT and ACT are standardized tests, and test scores significantly determine admission to American universities. A high score opens access to leading universities and scholarships; a low score significantly narrows possibilities. This makes the market for preparation for these exams a segment with zero tolerance for failure: parents and students won't forgive a platform that "explains well" but doesn't deliver measurable score growth.

• SAT is administered by College Board, ACT — by the eponymous nonprofit organization
• Both exams are accepted by most US universities for admission
• The US private tutoring market for SAT/ACT is valued in the billions of dollars
• The average cost of a private SAT tutor ranges from $100 to $300 per hour

For an AI platform, such requirements set specific tasks. It's not enough to correctly explain a topic. You need to diagnose weak areas for each student, organize targeted practice specifically in those areas, and then build skills for working under strict time constraints — the same as on the actual exam. A mistake in preparation strategy or a missed type of problem can directly affect the final score.

How Smartschool builds a different approach

Smartschool positions its product as AI focused on results rather than the accuracy of answers to individual questions. According to the startup's team, a pedagogically effective AI requires design from scratch — taking into account how people actually learn, not just consume information. In this architecture, the language model is just one of the tools, not the entire product.

The company operates in a segment where competition with traditional tutors requires not only accurate answers but also pedagogical effectiveness. The question is how much an AI tutor will be able to actually complement or replace an experienced mentor who knows the specific student and builds a trusting relationship with them.

What this means

The educational AI market is flooded with tools that "explain well" but don't deliver measurable growth in results. Smartschool is betting that correct pedagogical architecture matters more than the power of the underlying language model. This thesis will have to be proven with real statistics — exam scores, not demo videos.