Improving SAT Test Performance by Having Students Engage in SelfDirected Learning Using Self-assessment and Remediation With METYFoundation.org-Generated Templates

Abstract

Millions of students take the SAT and other high-stakes tests that influence what colleges or other programs they get into or graduate from. Test preparation has become a multibillion dollar global business, but evidence suggests that SAT programs lead to minimal improvement on average. Our previous research shows that teaching students to self-assess and remediate their own knowledge using Cognitive Structure Analysis (Leddo et al., 1990) improves performance for a variety of ages and academic subjects, including on SAT tests (Mishra and Leddo, 2026). Because our previous research involved manually creating and distributing selfassessment templates, a non-scalable practice, we developed a website, www.metytechnology.org, that offers free AI-generated templates for any grade level, subject and language. Chandhu and Leddo (2026) demonstrated that high school students using these AI-generated templates greatly improved their performance in math. The present study seeks to extend these findings using AI-generated self-assessment templates in SAT math. 20 students from Mt. Hebron High School in Maryland, USA took a 15-minute SAT math pre-test, consisting of 2 medium difficulty and 13 hard questions from the College Board item bank. Upon completion, half were allowed to review the material on their own using online resources, while the other half were shown how to generate self-assessment templates on the metyfoundation.org website and then reviewed the material using online resources after conducting self-assessment using the templated generated from metyfoundation.org. Following the review, both groups took a new 15-item SAT math post-test that also consisted of 2 medium difficulty and 13 hard math problems. Results showed no significant improvement in performance in the control group, whereas students using the self-assessment templates increased their performance from an average of 4.3 correct questions on the pretest to 7.5 correct questions on the post-test.