Schools should be able to decide when, how and whether their students use generative AI. Google’s latest announcement shows how complicated that freedom becomes when the company supplying an everyday educational service also decides what must come with it.

On 3 November, The New York Times reported that Google would turn off student access to Search for schools that reject its AI entirely. The announcement has provoked an understandable backlash. But the details matter, because the strongest criticism is not quite that schools must adopt the Gemini chatbot or lose the internet. It is that Google is defining the choices available to educators—and excluding a choice many of them want: access to Google Search without its generative-AI features (Goldstein & Singer, 2026; Duffy, 2026).

This is a debate about AI in education. More fundamentally, it is a debate about who gets to make educational decisions.

What Google has actually announced

Google says access to Gemini, Gemini Notebook, Gemini in Classroom and Google Search will be switched off for K–12 student accounts until school administrators explicitly enable the services. The changes are expected in early 2027, according to the Times. Google’s announcement says further implementation details will follow in the coming months (Phillips, 2026; Goldstein & Singer, 2026).

Moving AI products from automatic availability to explicit administrative approval is a welcome response to calls for greater control. That part should not be overlooked.

However, Google’s subsequent clarification makes clear that schools can enable Gemini, Gemini Notebook and Search individually. A school will not have to activate the separate Gemini chatbot to retain Search. The unresolved choice is inside Search itself: Google has not offered an administrator option to retain that service while excluding AI Overviews and AI Mode, its generative-answer features (News from Google, 2026; Duffy, 2026).

The announced restriction concerns access through school-managed student accounts. It is not an announcement that Google will disconnect schools from the internet or prevent them from using other search services. Nor should the initial US Search pilot, the global availability of Guided Learning in Classroom, and the forthcoming account-control changes be treated as one identical rollout. Google has not yet published all the implementation details (Phillips, 2026).

Those qualifications do not remove the concern. They locate it precisely: schools can choose between products, but not necessarily between educationally different ways of using a product.

An opt-in is only as meaningful as the options

Imagine a school deciding that younger pupils should learn to find information, compare sources and construct an explanation before routinely receiving a machine-written synthesis. It might allow teachers to experiment with AI lesson planning, permit a carefully supervised tutoring activity, and still want conventional web search for students.

That is not an incoherent position. It is a differentiated educational judgement.

Yet if Google’s available configuration is Search with generative AI or no Search through the student account, that judgement cannot be implemented within the service. Administrators may have control over the switch, but the supplier determines what the switch controls.

This is why the company’s claim that the announcement offers more choice does not fully answer its critics. There is more choice over access to separate products. There is not the particular choice to separate web search from generative responses (News from Google, 2026; Duffy, 2026).

Embedding AI in an already familiar service is a powerful route to adoption. Users do not have to seek out a new tool: it becomes part of an established routine. That makes commercial sense. It does not, by itself, provide an educational justification.

We do not need to establish that commercial interests are Google’s only motivation to question this arrangement. The issue is whether a supplier’s product strategy should determine the limits of a school’s pedagogical discretion.

Search was never neutral—but generated answers are still a change

One response to the controversy is that Search already uses AI, so asking for AI-free search is unrealistic.

There is an important truth here. Google’s conventional ranking systems use AI technologies including BERT, neural matching and RankBrain to interpret queries and identify relevant content. Search has long mediated what becomes visible. A ranked results page is not a neutral window onto the web (Google, 2025a).

But this does not make all uses of AI equivalent.

Using machine learning to retrieve and rank existing pages is different from presenting an AI-generated account of what those pages mean. Google describes AI Overviews as offering the gist of a topic, while AI Mode generates responses to more complex questions, with links to supporting websites (Google, 2025b).

The meaningful distinction is therefore not between AI and no AI anywhere in the system. It is between AI-assisted retrieval and a search experience that also supplies machine-written explanations.

Both need critical scrutiny. Neither automatically determines how much a student learns. A generated explanation can be useful material to examine; a conventional results page can encourage superficial clicking. But the differences matter when teachers decide what intellectual work students should undertake.

Nor does the presence of supporting links settle the educational question. A citation creates an opportunity to inspect a source. It does not establish that a student opened it, understood it or checked whether it supports the claim. Turning off generated responses would not make the wider web free of AI-produced content, but it could preserve a different starting point for inquiry

Promising tutoring is not a mandate for universal adoption

Google’s response is not simply that AI is inevitable. It is also making a pedagogical argument. Its Guided Learning approach is intended to support productive struggle: coaching students through problems rather than supplying final answers. The company is introducing this into Search, initially through a pilot in selected US schools (Phillips, 2026).

That deserves serious investigation, not dismissal. Step-by-step support is a more educationally plausible proposition than a tool that simply completes assignments.

Yet the evidence needs to be read with attention to context. In Google’s October account of a ten-week UK mathematics trial involving more than 1,100 students, AI tutoring was delivered through Eedi’s structured platform. Human experts reviewed every AI response before students received it. The company reported benefits on classroom tasks and intermediate quizzes, but said those gains did not appear in the independent STAR standardised test scores during the trial (McKee & Rosenburg, 2026).

This is not evidence that AI tutoring cannot help. It is evidence that the design of the activity, the support surrounding it and the measure of learning matter. Carefully supervised tutoring in a defined mathematics programme is not interchangeable with an AI feature encountered during everyday searching.

Even strong evidence for one use would not justify removing the option to decline another. Schools need to assess learning, age appropriateness and privacy in their own context. They also need to consider what students can do independently when the tool is absent.

The educational question should be whether a particular use is worthwhile—not whether accepting it is the price of retaining a familiar service.

Critical digital competence also requires institutional agency

In my previous post, I discussed UCL’s adolescent chatbot research and the CoDiCri Critical Digital Competence Handbook. One connection between them is the importance of agency: helping learners make informed decisions rather than simply become more proficient users.

CoDiCri makes that aspiration practical by asking educators to consider intellectual tasks, resources and the organisation of participation together. Its handbook treats critical digital competence as something developed within disciplinary learning, not a separate lecture on digital dangers (CoDiCri, 2026).

Google’s announcement adds an institutional question to that discussion. We cannot ask students to make thoughtful decisions about technology while treating the institution’s adoption decisions as inevitable.

A school needs the freedom to design an AI-supported activity where it makes educational sense. It should also be able to design an activity without generative assistance, postpone adoption or stop using a feature after evaluation. Non-use is not necessarily technological illiteracy. Sometimes it is the outcome of critical judgement.

A useful classroom inquiry could examine this controversy itself: what options does the platform actually offer, whose interests do they serve, and how do they differ from the options educators want? That would connect critical digital competence to the conditions under which learning takes place—not merely the quality of a prompt.

But a lesson cannot compensate for a lack of institutional choice. Students’ agency and schools’ agency are connected.

Europe’s sovereignty question is practical, not rhetorical

The immediate controversy centres on American schooling, but it exposes a problem European education systems should recognise. When schools depend on services they do not control, a supplier’s change of direction can require them to reconsider decisions they thought were theirs.

Denmark provides a concrete example. In January 2024, its data protection authority issued an order concerning 53 local authorities’ use of Google Workspace in schools. The issue included disclosures for specified secondary purposes that lacked a sufficiently clear basis in Danish school legislation. A January 2026 follow-up expressed serious criticism of 51 authorities’ compliance documentation and issued conditional warnings about configurations and sub-processor arrangements (Danish Data Protection Agency, 2024, 2026).

This should not be described as a permanent nationwide ban on Google. Nor is it evidence that Google’s new AI policy is unlawful. It demonstrates something more useful: governing school infrastructure requires detailed work on contracts, data flows, configurations and accountability. Locating some data in Europe does not settle the entire processing chain.

In the Netherlands, SIVON and Kennisnet reported in June that they were developing exit scenarios for schools facing supplier outages, changed conditions or services that no longer meet educational needs. Their work includes practical playbooks and feasibility testing. Importantly, SIVON also acknowledged that full alternatives remain limited or may not yet meet schools’ requirements (SIVON, 2026).

That is digital sovereignty stated in operational terms. Can a school leave? What can it take with it? Who will support the transition? What disruption can it afford?

These questions do not imply that rejecting Google’s Search configuration requires replacing every school system. Changing a search service and migrating an entire digital environment are different tasks. But both require institutions to understand their dependencies and maintain credible alternatives.

There are signs of investment. On 3 November, IONOS announced a four-year framework contract for a Nextcloud-based school collaboration service covering six German states, also reported by Reuters. The offer includes setup, migration and operation, with gradual adoption possible (IONOS Group, 2026; Reuters, 2026).

It is an alternative procurement route, not proof that all those schools have migrated or that every Google function has been replaced. Its scope is collaboration and storage, not web search. Nevertheless, it illustrates the kind of investment that makes choice more than a declaration.

European ownership or open-source software alone does not guarantee educational freedom. A replacement can create new dependencies. What matters is accountable governance, interoperability, usable exports, support and the practical ability to change direction. Those capabilities need funding before a crisis, not just a promise to investigate alternatives afterwards.

The right to say “not this, not now”

Google’s announcement should not push the discussion into a choice between embracing all AI and banning it everywhere. That binary is part of the problem.

Schools may want to experiment selectively, restrict particular features or use different approaches with different age groups. They should not have to justify those decisions solely in terms of whether they fit a supplier’s current product architecture.

For Google, a convincing response would be to offer a supported way for schools to retain Search without generative-answer features. For education authorities, the response should be to build procurement and infrastructure arrangements that preserve genuine choices when a supplier changes its terms.

AI may have valuable roles in education. Whether it does, for whom and under what conditions should remain open questions for research and professional judgement.

Educational choice must include the ability to say “not this, not now”—without surrendering unrelated capabilities as the price of refusal.

References

CoDiCri. (2026). Critical digital competence: User guide / handbook (Version 1). Zenodo. https://doi.org/10.5281/zenodo.22869113

Danish Data Protection Agency. (2024, January 30). The Danish Data Protection Agency issues an order in the Chromebook case. https://www.datatilsynet.dk/Media/639287758061619989/cb-january-2024.pdf

Danish Data Protection Agency. (2026, January 29). The Danish Data Protection Agency issues serious criticism of 51 local authorities in the Chromebook case. https://www.datatilsynet.dk/Media/639286166007034414/cb-2026.pdf

Duffy, C. (2026, November 3). Google will let schools turn off AI, but they may also lose Search. CNN. https://www.cnn.com/2026/11/03/tech/google-ai-search-policy

Goldstein, D., & Singer, N. (2026, November 3). Google will turn off Search for schools that reject A.I. The New York Times. https://www.nytimes.com/2026/11/03/us/google-search-schools-restrict-ai.html

Google. (2025a, December 10). A guide to Google Search ranking systems. Google Search Central. https://developers.google.com/search/docs/appearance/ranking-systems-guide

Google. (2025b, December 10). AI features and your website. Google Search Central. https://developers.google.com/search/docs/appearance/ai-features

IONOS Group. (2026, November 3). IONOS builds cross-state school cloud for six federal states. https://www.ionos-group.com/investor-relations/newsroom/ionos-baut-laenderuebergreifende-schul-cloud-fuer-sechs-bundeslaender.html

McKee, K., & Rosenburg, A. (2026, October 29). How Gemini is helping students and teachers in the classroom. Google. https://blog.google/products-and-platforms/products/education/impact-studies-uk-italy/

News from Google [@NewsFromGoogle]. (2026, November 3). We want to clarify what we announced today about our education services. Today’s announcements give schools more choice, not less [Post]. X. https://x.com/NewsFromGoogle/status/2117714815477059722

Phillips, C. (2026, November 3). Designing responsible AI for learning. Google. https://blog.google/products-and-platforms/products/education/ai-learning-era-updates-2026/

Reuters. (2026, November 3). IONOS wins contract to build sovereign school cloud platform for six German states. https://www.reuters.com/technology/ionos-wins-contract-build-sovereign-school-cloud-platform-six-german-states-2026-11-03/

SIVON. (2026, June 30). SIVON en Kennisnet werken aan exitscenario’s bij uitval Big Tech-diensten [SIVON and Kennisnet are working on exit scenarios for outages of Big Tech services]. https://sivon.nl/2026/06/ontwikkeling-exitscenarios-bij-uitval-big-tech-diensten-2/

About the Image

This image is a satirical interpretation of “AI slop”: the low-quality, often nonsensical AI-generated content that floods our feeds. Inspired by traditional vanitas paintings — still life compositions that show the futility of pleasure and the certainty of death — this piece symbolises the inevitability of our content being scraped and regurgitated as slop. The monitor displaying OpenAI’s logo alludes to the source of the problem, while the surrounding objects reference specific examples of “AI slop” and the various domains it can be found in: art, literature and music.