Can schools keep Google Search without accepting its AI features?
Google is giving school administrators separate controls for Gemini and other classroom AI, but its own account says Search is now off by default for younger pupils because AI is integrated into it. A small, first-party randomised study found a larger short-term statistics gain with Guided Learning, while leaving important methods and uncertainty unpublished.
The short version
- What happened
- Google says Search is now off by default for K–12 student accounts until administrators enable it, because AI features are integrated into Search; Gemini, Gemini Notebook and Classroom AI have separate controls.
- Who it affects
- Teachers may encounter AI inside Search even when they have not chosen a separate chatbot, so lesson instructions and safeguarding guidance need to reflect the service pupils actually see.
- What remains unproven
- The product controls are described by Google and have not been independently audited across regions, account types and age settings.
Keep in mind
The product controls are described by Google and have not been independently audited across regions, account types and age settings.
The Japanese study is available only as a first-party summary; no full paper, protocol or preregistration was linked when checked.
The exact randomised arm sizes, baseline balance, attrition, assessment instrument and uncertainty intervals were not reported publicly.
A half-day university statistics course cannot establish durable effects for school-age learners or other subjects.
The UK pilot covered three schools and four weeks but disclosed no participant denominator, control group or measured attainment effect.
The sources do not establish a competition-law breach, a privacy violation or that all schools receive identical controls.
Multi-source analysis · 3 linked sources. Primary analysed source date: 3 November 2026. Latest linked supporting source: 4 November 2026.
Keep following the evidence
Join The Impact Brief founding list for selected AI reports, their effect on people and the evidence limits. Read a sample first.
Read the full report · 11 min The evidence, methods, implications and limitations
The direct answer: schools can separate several Google AI products, but Search itself is no longer an AI-free option
Google has introduced more granular administrator controls for AI used through school accounts. Its 3 November announcement says administrators can switch access to Gemini, Gemini Notebook and Gemini in Classroom independently. For younger pupils, Google Search is now off by default until an administrator turns it on. The reason matters: Google says AI is integrated into Search, so a school cannot treat ordinary Search as a wholly separate, non-AI service for those managed accounts.
That is a meaningful governance change, not merely a new classroom feature. A school can withhold Gemini while retaining some other Google services, but enabling Search also accepts the AI systems embedded in Search. Associated Press reported that administrators can choose services separately, while a school seeking to reject Google AI altogether would have to forgo Google Search. The distinction should be clear to parents and teachers because a settings page labelled by product may not reflect the way AI now runs across the underlying service.
Google says Guided Learning in Classroom is available globally now. A separate Guided Learning experience in Search will begin with selected US schools later in November, with wider rollout details to follow. Those are different products and deployment stages. The announcement does not show that every school or every student has received the same controls, and availability does not establish educational benefit. Schools still need to test their own configuration, age rules, data handling and lesson design before treating the defaults as a safety guarantee.[1][3]
What the University of Tokyo study found
Google for Education and Google DeepMind also released a summary of a randomised study involving more than 300 University of Tokyo students. Participants completed a half-day statistics course using either standard AI tools or Guided Learning. Google reports that both groups improved on an assessment: the standard-tools group gained 5.3 percentage points, while the Guided Learning group gained 8.1 points. The reported difference between those changes is 2.8 percentage points; describing 8.1 as about 1.5 times 5.3 is arithmetically correct but can sound larger than the absolute gap.
The intervention is designed to prompt active participation rather than deliver a finished answer. Google says Guided Learning asks questions, offers hints and adjusts explanations while preserving a path through the topic. That makes the comparison educationally relevant: it tests two ways of using AI, rather than AI against a teacher or against no support. It does not show that Guided Learning outperforms an experienced tutor, a well-designed textbook, classroom instruction or practice without generative AI.
The public summary leaves several quantities unresolved. It says more than 300 students were randomised but does not give the exact denominator in each arm, how many completed the assessment or whether any responses were excluded. It does not report baseline means, standard deviations, confidence intervals, a statistical test, the assessment instrument or how randomisation was implemented. No linked paper, preregistration or protocol was available in the source reviewed. The direction is promising; the precision and robustness of the difference cannot be independently assessed from the summary alone.[2]
The distinction that matters
What is controlled, and what the evidence actually shows
Product controls and learning evidence answer different questions. This comparison keeps the denominator and claim boundary visible.
Search access
Off by default for K–12 managed student accounts until an administrator enables it
Enabling Search also enables the AI features integrated into Search
Separate tools
Administrators can control Gemini, Gemini Notebook and Gemini in Classroom independently
Schools gain product-level choices, subject to their exact account configuration
Japan study
More than 300 students randomised for a half-day statistics course
Reported gains were 5.3 points with standard tools and 8.1 with Guided Learning
UK pilot
Three secondary schools, four subjects and four weeks
Implementation feedback was reported; no controlled attainment effect or pupil denominator was given
A UK school pilot offers implementation evidence, not a controlled learning effect
The same company report describes a four-week UK pilot in three secondary schools across four subjects. Teachers used Guided Learning in Classroom and contributed feedback about how it fitted lessons. Google presents examples of students receiving staged questions and teachers using the system to support revision. This is useful evidence about feasibility and user experience: a product can work technically yet fail if teachers cannot fit it into planning or pupils cannot understand the prompts.
It is not equivalent to the Japanese randomised comparison. Google does not report the number of participating teachers or pupils, the proportion who completed the pilot, a control group, assessment outcomes or a response rate for the quoted feedback. With three schools, local leadership, devices, curriculum and staff enthusiasm could strongly shape the experience. Positive quotations can illuminate how a product was used but cannot estimate how common the experience was or whether attainment changed.
The two evidence streams should therefore remain separate. The University of Tokyo study supplies a short-term, randomised comparison but only through a company summary with incomplete methods. The UK pilot supplies classroom context but no controlled effect estimate. Together they justify further independent evaluation; they do not establish that the feature improves lasting learning across ages, subjects or school systems. A future report should state the denominator, comparator and measured outcome for each claim rather than allowing research and product testimony to blend into one impression.[2]
What administrators, teachers and families need to decide
Administrators face at least three decisions. First is access: which age groups may use Search, Gemini, Notebook or Classroom AI, and whether teachers can override a default. Second is data governance: what information students may enter, how activity is logged, which contractual terms apply and how safeguarding incidents are handled. Third is pedagogy: when guided prompts add value and when students need direct instruction, discussion, practical work or deliberate practice without an assistant.
Teachers need visibility at the point of use. A student may experience an AI-generated overview inside Search even if the school has not introduced a lesson branded as AI. Instructions should name the service and the relevant risks—incorrect sources, overconfident explanations, inappropriate content and reduced independent effort—rather than rely on a single generic AI policy. Staff also need an alternative for pupils who cannot or should not use a particular service, so access controls do not become an accidental barrier to research or homework.
Families should be told what a school has enabled, not just that it uses Google Workspace. Google says its education products include age-appropriate protections and administrator controls, but those are provider claims and settings can change. Clear local communication should explain whether Search is enabled, which AI features may appear, what student data is permitted and how parents or pupils can question an output. A product default is a starting configuration; accountability stays with the school and relevant public authorities.[1][2][3]
Why the product design raises a competition and dependency question
Google says Classroom serves 150 million users worldwide. At that scale, combining familiar Search access with embedded AI can influence how schools adopt AI even when administrators receive more switches. A school that depends on Search for everyday research may find that refusing integrated AI imposes a larger practical cost than declining a standalone chatbot. That does not prove unlawful tying or coercion, and the reviewed sources do not establish a competition-law case. It does create a dependency question that procurement teams should examine rather than treating each switch as an isolated choice.
Schools can reduce lock-in by documenting which learning task each service performs and whether an alternative exists. They can export lesson materials, avoid making one provider account the only route to core resources and review settings after major product changes. Public purchasers can ask providers to separate essential search or productivity functions from optional generative features where technically possible, and to give advance notice when the boundary changes.
Independent regulators and researchers will need to test how these controls behave in practice. Important questions include whether disabling one feature removes its data flows, whether AI-generated Search elements are consistently labelled, how age and region alter the experience and whether accessibility features work without forcing broader AI access. The present announcement explains intended controls; it is not an audit of their operation across the millions of managed accounts that may receive them.[1][3]
Evidence check and what would change the assessment
The control changes are supported by Google's own dated product announcement and independent reporting that tested their practical meaning. The learning result is weaker evidence than a peer-reviewed paper: it is a first-party summary of research conducted with an academic population, with no full methods document linked. The measured outcome follows a half-day statistics course. It cannot show durable knowledge, transfer to other subjects, effects on younger pupils or consequences for students who need additional support.
Confidence in the educational claim would rise with a public protocol and dataset description, exact arm sizes, baseline balance, attrition, confidence intervals and a clearly specified primary outcome. Independent replication should compare Guided Learning not only with generic AI, but with teacher-designed practice, conventional digital tutoring and no-AI resources. Delayed assessments would show whether the extra gain persists. Subgroup analyses would need enough participants to avoid mistaking noise for differences by background or prior attainment.
The governance assessment would change if schools could enable a clearly non-generative Search mode, or if independent testing showed that disabling AI features removed all relevant AI processing while preserving search access. It would worsen if controls proved inconsistent, hard to audit or misleading to users. Until those questions are answered, the reasonable conclusion is narrow: Google has added useful administrative separation around several products, but it has not restored a simple boundary between Search and AI, and its early learning evidence is not yet sufficient for a broad effectiveness claim.[1][2][3]
Watch the background, then check the evidence
Illustrative photo · not a video frame
UNESCO · 8 min 30 sec · Background explainer
Teacher to Teacher: Is AI reshaping education for the better?
YouTube loads only when you choose; playback starts when you press play.
A conversation about how teachers can guide AI use while preserving pupils’ judgement. Watch for the difference between helping a learner think and supplying an answer.
Evidence check: Teachers’ perspectives provide context. They do not establish that a particular AI tool improves learning outcomes.
Published 2025-01-23. Source checked 14 October 2026. Check captions on the publisher’s video page.
Read the education evidencePhoto: Vitaly Gariev / Unsplash · Licence
Continue the story
Read next
A second perspective on AI in education, with its own sources and qualifications.
Do learning agents improve results? A 32-person study cannot isolate the agent effect
An unreviewed China–Singapore study found larger short-term assessment gains with an agent-organised learning programme than a comparison tutor. Different materials, practice and assessment designs mean the result does not show that multi-agent coordination caused the difference.
Read the next reportExplore the full source and evidence record
What this means for people
- Teachers may encounter AI inside Search even when they have not chosen a separate chatbot, so lesson instructions and safeguarding guidance need to reflect the service pupils actually see.
- Students can receive more staged questions instead of immediate answers, but the published evidence does not show that a short-term gain lasts or applies across ages and subjects.
- Families and school leaders need a plain account of which services are enabled, what data use is permitted and what alternative exists for pupils who do not use them.
Global context
The reported research spans Japan and the United Kingdom, while the Classroom feature is described as globally available and the Search pilot begins in selected US schools. Education systems differ in age rules, procurement, curriculum, privacy law and teacher capacity, so a control or outcome in one country should not be assumed elsewhere. Google’s scale makes transparent defaults important, but each school remains responsible for local safeguarding and instructional choices.
What the evidence does not yet show
- The product controls are described by Google and have not been independently audited across regions, account types and age settings.
- The Japanese study is available only as a first-party summary; no full paper, protocol or preregistration was linked when checked.
- The exact randomised arm sizes, baseline balance, attrition, assessment instrument and uncertainty intervals were not reported publicly.
- A half-day university statistics course cannot establish durable effects for school-age learners or other subjects.
- The UK pilot covered three schools and four weeks but disclosed no participant denominator, control group or measured attainment effect.
- The sources do not establish a competition-law breach, a privacy violation or that all schools receive identical controls.
What to watch next
- A full paper or protocol for the University of Tokyo randomised study, including exact denominators and uncertainty
- Independent replications with school-age pupils, active non-AI comparators and delayed assessments
- Whether schools can preserve conventional Search without accepting integrated generative features
- Regional differences in controls, student data handling and AI labelling
- Measured outcomes from the UK and US school pilots rather than selected feedback alone
Living evidence record
Impact record IAI-1C7RCWJ
Evidence stage
Observed
Confidence
Corroborated
Reporting basis
Multi-source analysis
Independent or research support
Present
Record status
Monitoring
Last checked
4 November 2026
Source trail
3 direct sources across 3 source types.
People impact
Documented in this record.
Uncertainty
Limits and next checks are explicit.
Stages describe the evidence available—not whether a technology is good or bad. See the public method.
Evidence trail
Sources used for this report
Links checked 4 November 2026
Research question, topics and regions
Whether a guided generative-AI learning mode improves short-term statistics learning compared with standard AI tools, and how schools can control access
Topics: School AI controls · Google Search · Guided Learning · Education technology · Student safeguards
Regions: United States · United Kingdom · Japan · Global
This report is labelled multi-source analysis. We summarise and analyse source material in our own words; company statements remain attributed claims until independently supported. Translated summaries preserve the meaning of the original source and link back to it. Read our editorial standards.
Reader commentary
What does this mean where you are?
Tell us what you have seen at work, in your community or in everyday life. A question is welcome too. No account is needed; we review notes before publishing them.
Explore commentary across the portal →Published reader notes
0No published reader notes yet. You can start the evidence-led discussion above.
Prefer a private correction or response? Contact the newsroom.