Google tones down AI in schools, but Search can't be fully switched off

Google announced changes on November 3, 2026, to how AI is used in schools. According to Chalkbeat, Gemini, Google's AI chatbot, will be turned off by default in Google Classroom. Chalkbeat reports that Google Classroom is used by 150 million students and teachers worldwide.

When Gemini is turned on, it runs in Guided Learning mode. It walks through examples and questions with the student instead of giving the answer directly. On school accounts, Google Search also shows Guided Learning instead of AI Overviews, the AI-written answers at the top of the search page.

One hard limit

Schools that want AI turned off completely will have to give up access to Google Search. There is no separate switch that turns off only AI in search. Chris Phillips, Google's education director, says, according to Chalkbeat, that AI is deeply integrated into Google Search and can't simply be turned off.

Eisha Buch, head of teaching and learning at Common Sense Media, estimates that the chance of schools avoiding Google Search is very small to nonexistent.

Kevin McKee of Google DeepMind says, according to Chalkbeat, that students in education need to do the work themselves in order to learn.

Testing with selected US schools begins in November 2026. The full rollout is planned for early 2027.

Criticism from Common Sense Media

Common Sense Media's reviewers assess that Guided Learning can still solve homework, and that the line between helping and doing the work is unclear. Large districts such as New York City and Los Angeles already restrict the use of AI in schools.

In an earlier assessment, written before the announcement, the organization ran more than 2,600 searches with accounts for 11- and 15-year-olds and gave Google Search the lowest rating, "Unacceptable Risk." According to the organization, AI Overview solved homework assignments in 180 out of 180 tests, and the same questions produced significantly different answers in 43 percent of cases.

In mental health crisis scenarios, the AI picked up warning signs in only 58 percent of cases, compared with an acceptable threshold of 95 percent. 29 percent of source citations came from user-generated sites such as Reddit and Facebook.

Google has acknowledged a problem with a discontinued eating disorder helpline and said it will be fixed. Google also says some answers are adjusted to the user's age. The results are based on Common Sense Media's own methodology.