San Francisco has its elementary school students read to software that records them, without proof of effectiveness

The district has purchased 23,000 subscriptions to Amira for nearly $400,000 per year and requires 30 minutes of weekly use per child; teachers dispute the lack of demonstrated results.

The San Francisco Unified School District licenses Amira for all of its students from kindergarten to fifth grade: a contract for 23,000 subscriptions for nearly $400,000 per year. The software records the child reading aloud, evaluates his or her performance using machine learning, and then sends back adapted reminders. The district wants each student to use it for at least thirty minutes a week, headphones on, on the laptops provided.

The report published on September 25, 2026 documents the friction between this generalization and teachers in the field, who blame the district for pushing the tool when the publisher has not yet demonstrated that it works. The issue is not anecdotal: reading at the beginning of primary school is one of the few areas where an early deficit is difficult to make up.

The case sums up a broader shift. Automated tutoring tools are no longer experimented with at the margins but deployed on the scale of an entire district, and the question of proof of effectiveness comes after the purchase rather than before.