BIGGER PICTURE Learning & work
A text message that leads back to the classroom
Sometimes a student needs a reminder, a clear answer or the address of the tutoring service. Georgia State tested whether an AI-supported conversation could make that help arrive before a student falls behind.

- Updated evidence
- June 2026 working paper
- Grade result
- +4 percentage points earning A or B
- Technology
- Non-generative chatbot with human backup
How the idea works
Spot the next course action
Course schedules and performance information guide outreach.
Answer from approved content
A non-generative chatbot handles routine questions.
Connect to human support
Teaching assistants take over unanswered or complex issues.
Help before the missed deadline
The project targeted large courses where students could struggle to connect with teachers or peers. Messages drew attention to coursework and available support. Students could reply at any hour rather than navigate another office or website. The proposition was simple: make the next useful action easier to find and take.
Sources: [1]
A prepared knowledge base, a real teaching assistant
This was not a generative tutor inventing explanations. Faculty and administrators built a course knowledge base; the chatbot selected suitable responses. Questions it could not answer went to a teaching assistant. Performance information made outreach more relevant, while staff monitored messages and intervened when a student needed help beyond a routine answer.
Sources: [1]
The updated study found more strong grades
The June 2026 working paper reports a four-percentage-point increase in earning an A or B across government and microeconomics courses. Tutoring attendance also increased. The analysis generally found no statistically significant differences between demographic groups; women in microeconomics were an exception, with stronger effects. The outcome concerns course grades, not graduation.
Sources: [1]
Scaling a relationship takes preparation
Georgia State subsequently launched further trials in mathematics and English with partner universities. Those announcements are plans, not additional proven results. This case suggests an accessible design principle: connect students to existing human support at the moment they need it. It still takes reliable course information, staff time and monitoring to keep that connection trustworthy.
Sources: [2]
The original sources
- Annenberg Institute / study authorsLet’s Chat: Leveraging Chatbot Outreach for Improved Course Performance, June 2026 version
- Georgia State University, National Institute for Student SuccessChatbot Launch