Schools Can Say No To Google AI. But They’ll Lose Google Search Too.

Google is introducing new AI learning tools, tighter controls for schools and fresh research into their effectiveness. Yet one decision about Google Search raises uncomfortable questions about how much choice schools really have.

Amid growing criticism of Google's AI tools in schools , the company has announced plans to make access to Gemini, NotebookLM and Google Search an explicit choice for schools. Under the new arrangements, these services will be switched off for school-managed K-12 student accounts unless administrators decide to enable them.

When I met with Chris Phillips, Google's Vice President and General Manager of Education, yesterday, my immediate question concerned Google's AI Overviews, the AI-generated answers that now appear at the top of many Google searches.

Educators I have spoken to are increasingly concerned that these summaries present students with confident answers before they have had an opportunity to investigate the information themselves. The friction of finding sources, comparing evidence and reaching conclusions is often where much of the learning happens.

I asked Phillips whether Google's new opt-in system would allow schools to remove AI Overviews while retaining conventional Google Search.

The answer was no. Let me explain.

Under Google's proposed changes, expected to begin rolling out in early 2027, schools that do not want students using AI within Search will have to disable Google Search entirely.

"Search and AI are deeply integrated," Phillips explained.

So no AI means no Google Search for schools. A school might be comfortable with students searching for information, comparing sources and developing research skills, while having reservations about an AI-generated summary doing some of that thinking for them.

Google is telling us that Search and AI can no longer be separated. This is a tension that parents and educators will feel, and it sits at the heart of an otherwise promising set of announcements.

Google Wants AI To Help Students Think

At the heart of today's announcements is Google's Guided Learning AI, which begins rolling out within Google Classroom today, for schools that have already enabled Gemini access.

Rather than giving students the answer, Guided Learning is designed to help them arrive at one. It can ask questions, offer explanations, and can guide students through problems, using principles of learning science.

Crucially, students cannot simply turn off Guided Learning within those activities or open another conversation to bypass its restrictions.

Kevin McKee, a research scientist at Google DeepMind, acknowledged during the meeting that AI models have traditionally been developed to complete tasks for users, but schools require something different. Students need to do the intellectual work themselves.

Parents and educators both know the value of that moment when a child wrestles with an idea, makes a mistake, reconsiders and finally understands. Remove that process entirely, and you risk removing much of the learning with it.

Google is also bringing Guided Learning to AI Overviews in Search. A pilot beginning this month in selected US schools will guide students through certain academic questions rather than immediately presenting answers. So schools might not be able to switch off AI Overviews without switching off Google Search, but they may be able to choose the Guided Learning AI in AI Overviews, if the pilot succeeds.

Whether Guided Learning consistently refuses to simply answer questions is another story. There are anecdotal tales of Guided Learning sometimes completing student work. Google acknowledged that its general-purpose version operates differently, insisting that the Classroom integration has stricter safeguards.

What Does The Research Actually Tell Us?

Google has also presented new evidence suggesting its Guided Learning AI can improve educational outcomes.

According to preliminary findings shared by Google DeepMind during a press briefing this week, a randomised controlled trial conducted with the University of Tokyo found that students given access to Guided Learning improved their test scores by 8.1 percentage points, compared with 5.3 percentage points among students using conventional online resources. That represents roughly 50% greater improvement in test scores during the study session. Google expects to publish the full technical report later this week.

Google also pointed to a published randomised controlled trial in Sierra Leone , involving 1,763 students. Those using AI-supported mathematics instruction made gains that researchers equated to 1.2 to 1.7 years of typical learning progress over eight weeks.

But there are limitations here. The Tokyo study involved just half a day of independent study, and its full technical report had not been published at the time of the briefing. We also need to distinguish between improving performance immediately after using a tool and developing knowledge that survives once that tool is removed.

I asked McKee whether Google was planning longitudinal research examining AI's impact on learning over several years. He confirmed that longer-term studies were on the roadmap, although the company's research has largely concentrated on shorter periods so far.

That leaves an important question unanswered. Does AI help students become better learners, or simply more successful at learning tasks while assistance is available? The distinction matters, and it will take more than a single study session to resolve it.

There is genuine value in developing AI that challenges students to think rather than simply generating their work. Giving teachers greater insight into how students learn could also be useful, provided schools carefully consider the privacy implications of collecting and reviewing those conversations.

But schools need more than promises of better AI. They need the freedom to decide where it belongs. Google's latest announcements suggest the company is continuing to take the educational consequences of AI seriously.

This article was originally published on Forbes.com