Artificial Intelligence

Can Schools Turn Off AI in Google Search? Why Choosing a Product Is Not Controlling It

Schools can keep Google Search without Gemini, but no documented setting removes AI Overviews from Search. BBGK explains why choosing a product is not the same as controlling its AI features, and offers a five-question Control Granularity Check.

AI & Judgment | Institutional Governance | November 2026

Product choice is not feature controlConceptual diagram, not a Google admin screen. On the left, product-level switches: Google Search on, Gemini apps off, confirmed as separate controls. On the right, features inside Search: AI Overviews, AI Mode and Guided Learning, each marked with a question mark because no documented administrator switch removes them while keeping Search, as of 15 November 2026.BBGK ANALYSIS | AI & JUDGMENTWhen AI is built into Search,who gets to decide?PRODUCT LEVEL: WHICH SERVICE?Google SearchONGemini appsOFFSeparate switches: confirmed by Google≠FEATURE LEVEL: INSIDE SEARCHAI Overviews?AI Mode?Guided Learning?No documented admin switch, 15 Nov 2026Choosing a product is not the same as controlling what it does.SEE DEEPER | KNOW WIDERahsshohelahmed.com
BBGK conceptual diagram, not a depiction of Google’s admin console. A switch for a whole service does not establish a switch for the AI features inside it. Status as of 15 November 2026.

Key findings

  • Schools can enable Google Search without enabling Google’s standalone Gemini products. Google confirmed they are separate controls. That is real product-level choice.
  • Inside Search, the picture is different. Google’s own Search Help calls AI Overviews a core feature that cannot be turned off, and Google has said AI Mode and AI Overviews remain integrated into Search. As of 15 November 2026, no documented administrator setting keeps Search while removing them.
  • The lesson reaches beyond schools: an institution’s authority becomes operational only when its tools can implement the boundary it has chosen. Where a vendor does not supply the switch, control moves to workarounds that cost money and technical capacity, which not every institution has.

Can schools turn off AI in Google Search? Partly. As of 15 November 2026, a school administrator can keep Google Search while switching off Google’s standalone Gemini products. What no published Google control does is keep conventional Search while removing the generative features built into it, such as AI Overviews. Google says the configuration details of its new K-12 controls are still to come.

That gap sits at the center of a dispute that began with Google’s 3 November announcement, and it matters more than the dispute itself. A school can hold the formal authority to approve a service and still lack the settings it needs to apply its own rules inside that service. The same problem will surface wherever AI is added to software that organizations already depend on.

What Google announced, and what it clarified

In its 3 November update on AI for education, Google said Gemini, Gemini Notebook, Gemini in Classroom and Google Search would be off by default for K-12 student accounts until school administrators explicitly turn them on. The post gives no effective date. AI Business reported the change for early 2027, and Google says it will share specifics with schools in the coming months. The same announcement brought Guided Learning, a step-by-step coaching mode, into Search through a pilot in selected U.S. schools.

Early coverage read the plan as an ultimatum, and Google’s own words invited that reading. Chris Phillips, Google’s vice president and general manager for education, said in a briefing that a school wanting AI off has “the choice to not have Search on at all,” as EdSource reported.

Google then clarified. A company spokesperson, Davis Thompson, told Government Technology that Search and the Gemini products are “separate controls.” In a post on X, Google said schools can turn off Gemini and Gemini Notebook and still use Search, while AI Mode and AI Overviews remain integrated into Search, according to the same EdSource report. Ars Technica’s headline summed up the result: schools can keep Search while skipping Gemini, but AI search is non-negotiable.

The clarification matters. Declining Gemini does not mean losing Search, so the dispute cannot fairly be framed as Google forcing every district to adopt its chatbot. A narrower question remains: can an administrator enable Search without the generative features inside it?

Evidence boundary

Google’s 3 November blog post does not itself say that administrators cannot separate AI Overviews or AI Mode from Search. That limitation comes from Google’s statements as reported by news outlets. Separately, Google’s Search Help page on AI Overviews describes them as “a core Google Search feature, like knowledge panels” that cannot be switched off, while noting that the Web filter shows text-based links without them. That page describes ordinary Search for any user. It is not the final specification of the planned K-12 controls.

Choosing Search is not configuring Search

Consider a school with a defensible instructional rule: students may use a search engine to find primary sources, but some assignments must be completed without AI-generated summaries or worked answers. The rule does not reject AI. It separates information retrieval from answer generation and defines when students must do the thinking themselves.

If an administrator can turn Search on and the Gemini apps off, that school has a meaningful product choice. But if enabling Search necessarily brings generated answers with it, the school still cannot implement its narrower rule through the controls Google provides.

BBGK evidence summary, 15 November 2026. Each row separates what the record supports from what it does not establish.
Question What the evidence supports What it does not establish
Can Search stay on while Gemini is off? Yes. Google calls them separate controls. That Search is free of generative AI.
Can AI Overviews be switched off inside Search? Google’s Search Help says no for ordinary Search; the Web filter avoids them one search at a time. The final design of the K-12 administrator controls.
Can AI Mode be restricted? Los Angeles Unified and Albany report having it off in their current setup. That AI Overviews disappear with it.
Does Guided Learning sit outside these features? No. Google said it runs through AI Mode and AI Overviews. That a school could keep Guided Learning while removing AI Overviews.
Has the new default taken effect? No date in Google’s post; reporting points to early 2027. That any school already runs under the new controls.

The practical difference shows up in LAist’s reporting on Los Angeles Unified. A district spokesperson said students can still use Google Search while AI Mode is turned off, but some searches may still show an AI-generated overview, and the district has asked Google for a modification so the summaries can be turned off. In Albany, a district spokesman told the Times Union that the district had disabled Gemini and AI Mode but could not disable AI Overview summaries, as FingerLakes1 reported.

The Guided Learning detail is easy to miss. PPC Land reports that Google’s own post on X said the feature runs through AI Mode and AI Overviews. Google’s main safeguard against shortcut answers is delivered through the same features some schools want to switch off. For an administrator, the real choice is not AI or no AI. It is which AI behavior, for which students, on which tasks.

The working rule. Institutional authority becomes operational only when the institution can implement the boundary it has chosen.

What districts are already doing

Schools are not making one uniform choice about AI, which is exactly why one switch per product is not enough. Under New York City Public Schools’ 2026-27 guidance, student-facing generative AI is not permitted in grades 2-K through 8, while high schools have limited, supervised use. News 12 reported that companion chatbots are barred at every grade level. Los Angeles Unified has a moratorium on student use of generative AI on district-issued devices, according to the same LAist report.

Seattle went further at the feature level. The district announced that Gemini, other unapproved generative AI tools and Google Search AI summaries would be blocked for K-12 students from 9 November, Axios reported. KUOW reported that the district was also closing a loophole that let students reach Gemini through search results.

Set side by side, the districts diverge. Seattle says it is blocking AI summaries in Search; Los Angeles and Albany say they cannot. The district statements do not explain how Seattle does it. One route is a third-party filtering layer: Techaeris reports that Lightspeed Systems released Google Search AI controls on 2 November that default searches to web results and block AI Mode, at additional cost.

BBGK inference: when a vendor does not supply a switch, control does not disappear. It moves to a workaround layer that each institution must buy, configure and maintain. That makes governance capacity uneven. A district with an IT budget can enforce a boundary that a smaller district can only write into policy.

What the AI Delegation Boundary Framework adds

BBGK’s AI Delegation Boundary Framework sorts decisions into three tiers: Automate, Assist and Own Directly. It asks which tasks AI may perform, where human oversight is needed and where a named human must keep responsibility. Applied to schools, it looks like this. The framework was not developed for, or tested on, this Google case.

Tier 1Automate

Bounded, auditable routine operations where errors are manageable, such as spam filtering or scheduling. Not every search or summary task belongs here.

Tier 2Assist

Guided practice, explanation or feedback, where learners and educators stay responsible for understanding and decisions.

Tier 3Own Directly

Educational policy, accountable judgments and deliberately independent student work, kept under human control.

One limit needs stating plainly. Own Directly does not mean prohibit all AI. It means the appropriate human authority makes the consequential decision and remains responsible for it. A school may choose supervised AI help for one exercise and independent reasoning for the next. Its problem begins when the software it has adopted cannot execute that distinction.

A default-on feature inside a trusted tool also changes who decides. BBGK’s analysis of the judgment calls nobody made showed how decisions can pass to an AI system without anyone choosing to delegate them. An AI summary appearing above every search result is the institutional version of the same drift: a delegation that happens by configuration rather than by decision. A boundary drawn in policy must map onto an enforceable setting in the tools that carry it out. Otherwise the institution has an intention, not a control.

Why the learning evidence demands precision

There is no adequate research basis for saying that all AI support improves learning, or that all of it undermines learning. In its November update, Stanford’s SCALE AI Hub for Education reviewed 52 randomized and quasi-experimental studies of generative AI in K-12 and higher education. Only 10 involved U.S. K-12 students or educators, and the authors caution that the findings should not be treated as broadly generalizable, Education Week reported.

SCALE’s 2026 evidence review adds two findings that bear directly on this dispute. Tools built to give step-by-step reasoning rather than direct answers show more promise than general-purpose AI tools. And gains students show while using AI can weaken or disappear when they are assessed without it.

That second finding is the one that matters for the Own Directly tier. If an assignment exists to build independent capability, AI help during it can hide the very gap the assignment is meant to reveal. BBGK’s analysis of cognitive offloading in young people examines how reliance on assistance differs from developing the capability itself.

Safety assessments add a second consideration. In July, Common Sense Media rated both AI Overviews and AI Mode an unacceptable risk for young users, PBS NewsHour reported. Google has challenged that assessment, according to eWeek. Neither the research review nor the safety rating evaluates Google’s planned controls. Together they support a cautious instructional distinction: a tool that helps explain a concept is not equivalent to a tool that supplies the answer before the student engages. Google’s Guided Learning design is a serious attempt to sit on the right side of that line, but a vendor’s design objective is not an outcome finding.

The BBGK Control Granularity Check

The BBGK Control Granularity Check is a five-question test for any AI-integrated service an institution is about to enable. It asks whether the controls the vendor provides match the boundary the institution’s own policy draws, and what it will cost to close the gap if they do not.

It is a BBGK working tool that operates inside the AI Delegation Boundary Framework. It has not been experimentally validated.

  1. Name the capabilityList the AI features inside the service, not just the product name
  2. Assign a tierAutomate, Assist or Own Directly, by task and by age group
  3. Find the switchIs there a documented control for each feature, at the level you govern?
  4. Price the workaroundIf not, what does a filter, managed browser or alternative cost, and who pays?
  5. Govern the next changeWhat happens when the vendor adds or merges a feature?

Questions 1 and 3 are answerable from vendor documentation. If the documentation cannot answer them, that is itself the finding.

1. Name the capability, not the product

“Google Search” is a product. AI Overviews, AI Mode and Guided Learning are capabilities with different learning effects. Governance that stops at product names will miss the decision that matters.

2. Assign each capability a delegation tier

For each capability, decide by task and age group whether it belongs in Automate, Assist or Own Directly. The same feature can be Assist for a high-school research project and out of bounds for a closed-book exercise.

3. Find the switch, at the level you govern

Is there a documented vendor control for each capability, and does it work at the level where the policy applies: district, school, grade or class? A control the vendor describes in a press briefing is not a control until it appears in the admin console and its documentation.

4. Price the workaround

If the switch does not exist, name the alternative: a third-party filter, a managed-browser setting, a different search tool, or teaching practice alone. Record its cost, who maintains it and which students it leaves uncovered.

5. Govern the next change

Ask how much notice the vendor gives before adding or merging features, what the default state of a new feature will be, and who re-runs this check when it changes. A boundary that holds today can fail at the next product update.

The check sits beside BBGK’s other tools. The BBGK Evidence Meter grades how much confidence a claim deserves; the Control Granularity Check asks whether a stated policy can be enforced at all.

The strongest defense of Google’s approach

Google has a legitimate answer to part of the criticism. Its new approach makes named products opt-in for K-12 accounts, reversing a default-on rollout, and gives administrators independent switches for Search and the Gemini apps. That is real control. A school can decline a chatbot without declining every Google service, and it can still withhold Search entirely.

There are practical reasons to integrate. One product is easier to maintain and secure than many separately configured versions. Guided Learning may reduce some risks of direct answer generation, and a school that switches AI Overviews off may lose that coaching too. Google’s Web filter already offers a links-only view for any user who selects it. And schools can choose other search tools, managed browsers or local restrictions, although migrating or enforcing alternatives carries costs.

These points rule out treating Google as offering no choice. They do not resolve the narrower concern. Product-level choice is sufficient only if it matches the institution’s actual policy. If the rule is to allow source discovery but restrict generated answers, administrators need evidence that their tools can enforce that distinction, or another workable way to do it.

What happens next, and the principle that outlasts it

The record is still moving. Four dates will test this analysis:

  • 19 November 2026: in a letter announced on 13 November, New York City Council Speaker Julie Menin and Council Member Shekar Krishnan asked Google to say in writing by this date whether city schools can keep Search without AI Mode, AI Overviews or other student-facing generative AI, and if not, why not.
  • 30 November 2026: Senator Edward Markey’s 9 November letter to Alphabet CEO Sundar Pichai asked Google to commit to letting schools disable AI Overviews, Gemini and Guided Learning while keeping Search, with answers due by this date, WebProNews reported. The letter also asked whether Google is able and willing to disable the AI features without disabling Search, according to Boston.com.
  • Late November 2026: the Guided Learning pilot in Search begins in selected U.S. schools.
  • Coming months: Google’s promised configuration details, ahead of the reported early-2027 change in defaults.

As of this article’s 15 November research cutoff, neither response deadline had arrived, and nothing here infers an outcome from them. Whatever Google answers, the governance question will outlast it. Providers will keep merging features, revising controls and, under pressure, making settings more granular. The question is how institutions keep accountable choice when foundational tools change faster than their rules, training and procurement.

That is a question of transition governance, the concept BBGK set out in its analysis of AI acceleration and the capacity of institutions to adapt: what administrators are told before a change, which capabilities are on by default, what can be disabled independently, how new behavior is tested, and whether institutions have the time and means to keep their own safeguards. For providers, a serious choice architecture separates enabling a platform from enabling the high-consequence capabilities inside it.

If Google ends up offering conventional Search without generative summaries, this case will show that institutional pressure can produce finer controls. If it does not, the case will keep showing the limits of choosing between bundled services. Either way, the proposition holds: decision authority needs an implementable boundary.

Frequently asked questions

Can schools disable Gemini and keep Google Search?

Yes. Google has said Search and its standalone Gemini products are separate controls, so administrators can enable Search without the Gemini apps. That is product-level control. It does not make Search free of generative AI.

Can schools turn off AI Overviews in Google Search?

Not through any documented Google setting as of 15 November 2026. Google’s Search Help calls AI Overviews a core Search feature that cannot be turned off, and Google has said AI Mode and AI Overviews remain integrated into Search for schools. The Web filter shows links without AI Overviews one search at a time, and some districts use third-party filters. Google has not yet published the final configuration of its K-12 controls.

Is turning off AI Mode the same as turning off AI Overviews?

No. AI Mode is a conversational search experience; AI Overviews are generated summaries that appear within ordinary results. Los Angeles Unified and Albany both reported having AI Mode off while AI-generated summaries could still appear.

Why does feature-level AI control matter for schools?

A school policy may allow students to find sources while restricting generated answers during particular learning tasks. Research reviewed by Stanford’s SCALE Initiative found that gains students show while using AI can weaken when they are assessed without it. A product-level switch cannot enforce that finer distinction when both experiences sit inside the same service.

What is the BBGK Control Granularity Check?

It is a five-question test for any AI-integrated service: name the capability, assign it a delegation tier, find the switch at the level you govern, price the workaround, and govern the next change. It asks whether vendor controls match the institution’s own policy. It is a BBGK working tool, not a validated instrument.

Research note. Sources were checked against official pages and published reporting on 15 November 2026. Several statements by Google and by school districts are known only through news reporting and are attributed as such. The Lightspeed Systems product is reported by a single outlet. The 19 and 30 November response dates had not arrived at the research cutoff. The Control Granularity Check is a proposed tool, not a tested intervention.

Sources and further reading

  1. Google (2026). AI for education: updates for the learning era. Google Blog, 3 November 2026. Read the source. Company announcement; no effective date
  2. Government Technology (2026). Google reverses decision to turn off search for schools not using AI. Read the source. Reporting; spokesperson statement
  3. EdSource (2026). Google’s new AI ultimatum sparks concern over tech giant’s power in California schools. 5 November 2026. Read the source. Reporting; executive briefing and company post
  4. Ars Technica (2026). Google: Schools can keep search while skipping Gemini, but AI search is non-negotiable. Read the source. Reporting
  5. AI Business (2026). Google’s AI-first Search creates challenges for schools and businesses. 6 November 2026. Read the source. Reporting; rollout timing
  6. PPC Land (2026). Google will cut Search for K-12 students in schools that keep AI off. Read the source. Reporting; Guided Learning delivery
  7. Google Search Help. AI Overviews in Google Search. Read the source. Official documentation; ordinary Search, not K-12 controls
  8. LAist (2026). How will LAUSD respond to Google’s AI mandate? What families should know. Read the source. Reporting; district statement
  9. FingerLakes1 (2026). Google clarifies school AI policy, but dispute over student searches remains. 8 November 2026. Read the source. Secondary reporting of Times Union
  10. New York City Public Schools. Guidance on Artificial Intelligence, 2026-27. Read the source. Official district policy
  11. News 12 (2026). New York City schools ban student use of generative AI through 8th grade. 2 September 2026. Read the source. Reporting
  12. Axios Seattle (2026). Seattle schools dial back tech. 9 November 2026. Read the source. Reporting; district announcement
  13. KUOW Soundside (2026). Seattle schools Superintendent Shuldiner on classroom tech policy, school mergers. 9 November 2026. Read the source. Reporting; superintendent interview
  14. Techaeris (2026). Guided Learning comes to Google Search as school AI goes off by default. 3 November 2026. Read the source. Single secondary source; vendor product not independently checked
  15. New York City Council (2026). Speaker Julie Menin and Council Member Shekar Krishnan demand clarity from Google on mandated AI integration in NYC schools. 13 November 2026. Read the source. Official statement; response due 19 November
  16. WebProNews (2026). Senator Markey demands Google decouple AI from Search for schools. Read the source. Reporting; letter shared with the outlet; response due 30 November
  17. Boston.com (2026). Sen. Markey urges Google CEO to let schools opt out of AI without losing search. 9 November 2026. Read the source. Reporting
  18. Education Week (2026). Is AI good or bad for learning? The answer isn’t so simple. 10 November 2026. Read the source. Reporting on Stanford SCALE November review
  19. Fesler, L., Martinez Claeys, J., Agnew, C. and Loeb, S. (2026). The Evidence Base on AI in K-12: A 2026 Review. Stanford SCALE Initiative. Read the source. Evidence review; causal studies only
  20. PBS NewsHour (2026). Google’s AI search features pose unacceptable risk to children, new report finds. Read the source. Reporting on Common Sense Media assessment
  21. eWeek (2026). Google disputes child-safety report on AI Search as control gap persists. Read the source. Reporting; Google’s response

AHS Shohel Ahmed is Founder and Principal Analyst of BBGK (Beyond Boundaries Global Knowledge), an independent editorial platform on AI and human judgment, search and discoverability, and human consequence. He has more than ten years of client strategy experience across AI visibility, SEO and digital authority, and develops BBGK’s framework canon, including the AI Delegation Boundary Framework.

Author profile | LinkedIn

Editorial disclosure: this article applies BBGK’s existing AI Delegation Boundary Framework to a reported platform-governance dispute and proposes a BBGK working tool. The operational interpretation is BBGK’s own, not a claim made by Google, Stanford SCALE or any school district. No outcome, technical impossibility or corporate motive is inferred from the pending response deadlines.

Comments

comments

Get Essays by Email