Provider: Goblins
Program Description
Goblins is an online AI math learning platform designed for students in grades 5-12. Students work individually with an AI tutor that can respond to spoken, written, or typed work and provide real-time, step-by-step feedback. Teachers can assign pre-built or customized content to whole classes, small groups, or individual students, and students can also practice independently. In Arizona’s Deer Valley Unified School District, leaders recommended using Goblins as a supplemental support rather than for core instruction, but teachers ultimately chose whether and how to use the platform.
Findings
- Significant positive effect, but not yet causal evidence. Students who solved at least two problems per week on Goblins achieved nearly three months of additional learning (0.11 SD on spring MAP math assessments) than observationally similar students who used Goblins less or not at all, after adjusting for fall achievement and other observed characteristics. Because students were not randomly assigned to use Goblins, however, the study cannot establish that Goblins itself caused the difference.
- Students used Goblins for an average of 5.3 hours over 27 weeks. Just over half of students in the study, 3,088 of 6,065 students, solved at least two problems per week, a usage threshold used in the analysis.
- Exploratory analyses showed Goblins may have had larger positive effects on students with IEPs, Hispanic students, and English learners
- The comparison group included students who did not use Goblins as well as students who used it below the two-problem-per-week threshold. District personnel also reported that many students who did not use Goblins used other supplemental math technology. No cost data was reported.
- Teachers and students chose whether and how much to use Goblins, meaning heavier users may have differed from lower-use students in ways the study could not measure. A randomized trial would provide stronger evidence about whether Goblins causes improvements in student learning.
| Research Partner | Center for Educational Data Science and Innovation at the University of Maryland |
| Study Year | 2025-26 |
| Study Location | Arizona; Deer Valley Unified School District, a district of about 32,000 students; 35 schools with students in the sample. |
| Modality | AI Platform |
| Study Design | Quasi-experimental study using propensity-score weighting |
| Total Sample | 6,065 students in grades 6–8 in the final analytic sample. 3,088 met the study’s usage threshold of at least two problems solved per week. 2,977 were in the lower- or no-usage comparison group. |
| Characteristics of Treatment Group | Female: 49% Asian: 4% Black: 3% White: 62% Latino/a: 24% Other Race(s): 8% English Language Learner: 6% Students with Disabilities (IEP): 8% Economically Disadvantaged: Not Reported |
| Subject | Math |
| Grades | 6, 7, 8 |
| Average Actual Dosage | 5.3 hours over 27 weeks |
| Scheduled Dosage | None Reported |
For more about Accelerate’s takeaways from this study, check out the Accelerate blog.
