Google Will Turn Gemini and Search Off for K-12 Student Accounts Until Administrators Turn Them On, and Gave No Effective Date: A Default Changed by Blog Post Is Not Notice Your Contract Recognizes

Framing

Since Edition 32, this brief has sorted AI governance by instrument: a mayor's moratorium, a union and vendor standard, a state rule, a court motion, eight state lawsuits, a bankruptcy notice, a Senate letter, and last week a statute. This week the instrument is a vendor's own default. Edition 37 named the off switch a procurement specification, and Edition 39 recorded two senators asking Google why a switch is sometimes missing. On November 3, Google announced that Gemini, Gemini Notebook, Gemini in Classroom, and Google Search will be off for K-12 student accounts until administrators turn them on. I found no public reply from the senators and no source connecting the two events.

Read the mechanism. A default is the cheapest instrument a platform owns. It needs no statute, vote, or negotiation; it moves a decision to the administrator, and it moves accountability with it. Google's post gives no effective date and says only that specifics will come in the coming months. K-12 Dive reports that critics describe the choice as Search with its AI feature or no Search, and that it's unclear whether settings can be applied by grade or school. Whether that description is accurate depends on details Google hasn't published. Here's how I read it: the structural point stands either way. When a vendor bundles a basic utility with an AI service, the district is making a purchasing decision each time it flips a setting, and the contract is silent on who decides.

The research counterpoint is decision-making capacity. Memon and colleagues coded 36 state guidance documents: vendor evaluation appears in 83 percent, a review cycle in 47 percent, funding in 47 percent, and 72 percent carry no version history. Twenty-eight states cite one nonprofit's model guidance. Demszky and colleagues found that automated feedback changed what tutors did by 8 to 15 percent and left learner practice scores and attendance statistically unchanged over four weeks. Google's own report, published the same week, supports its efficacy claims with a mix of vendor-run studies, a preprint, and a 2020 consulting assessment. Edition 34 counted seven of sixteen widely used products with independent achievement studies. Administrators are being handed decision rights faster than they're handed evidence.

The timing argument is thirty days. Google's Search pilot starts in select U.S. schools later this month, the Character.AI court calendar runs from November 14 to December 4, and the Delaware form agreement allows closing as late as December 17. Before any of that settles, I'd have your district export its current organizational unit settings for each student-facing AI service, send its vendors a written request for effective dates and notice terms, and add a notice of material change clause to the next renewal. Waiting costs you the baseline, and I'd rather you have it before the change than after. If the default changes on Google's schedule, your administrators will learn what changed from a blog post, and a board member will ask what your agreement says.

Top Research and Policy Signals

1. Google Said Gemini and Search Will Be Entirely Off for K-12 Student Accounts Until Administrators Explicitly Turn Them On, and Gave No Date for the Change

Source type. Vendor announcement, not a law or a contract term; Google blog post dated November 3, 2026.

Citation (APA). Phillips, C. (2026, November 3). Responsible AI for schools with Guided Learning. Google. blog.google

Chris Phillips, listed as vice president and general manager of education at Google, writes that Google is "now updating our approach to controls for Gemini, Gemini Notebook, Gemini in Classroom, and Google Search." The post states: "Access to these products will be entirely off for K-12 student accounts until school administrators explicitly turn them on." It gives no effective date. Google says it will "share more specifics with schools in the coming months."

The same post says Guided Learning became globally available in Google Classroom on November 3 and that Google is "starting with a pilot in select U.S. schools later this month" for Guided Learning in Search, before expanding to other districts in the coming months.

K-12 Dive, reporting on November 5, says critics describe the choice as students using Search with its AI feature or not using Search at all. It reports that administrators can enable Search, Gemini, Gemini in Classroom, or Gemini Notebook for students at any time; it's unclear when the default-off setting takes effect, and it's unclear whether administrators can apply settings to individual students, grades, or schools. Superintendent Rich Calkins of Pocantico Hills in New York says he is "very suspect of any sort of corporation, entity, or group," and his district is waiting for Google's official rollout.

A delegated sub-agent reports that other outlets quote a later Google clarification that schools can turn services on individually; I didn't open them.

Study context. This is a vendor announcement. It isn't a contract term, a law, or a technical specification, and it says nothing about whether Search for schools will carry AI features by default. The critics' description of a forced choice comes from K-12 Dive's reporting and isn't a statement Google makes.

Leadership implication. Before Google publishes its specifics, I'd have your chief technology officer export the current organizational unit settings for Gemini, Gemini in Classroom, Gemini Notebook, and Search for every student unit, and file the export in the contract record as your baseline. Send Google or your reseller a written request for the effective date, whether existing settings will be kept or reset, whether Search can be enabled without its AI features, and how notice of the next default change will reach your administrators. Name who may turn each service on, and add a notice of material change clause to your next renewal.

2. A Delaware Court Order Set November 24 as the Bid Deadline for Character.AI, and the Form Sale Agreement for the Non-Disney Assets Never Uses the Words User, Minor, or Chat

Source type. Company notice of a court order in a pending Chapter 11 case, dated November 5, 2026.

Citation (APA). Character.AI. (2026, November 5). Chapter 11 update and transaction notice. Character.AI Blog. blog.character.ai and Rijo, L. (2026, November 6). Character.AI sale contract omits 40 attorneys general's user data conditions. PPC Land. ppc.land

The company's notice, in Case No. 26-11603 (BLS) in the U.S. Bankruptcy Court for the District of Delaware, states that the court entered an order dated November 4, 2026, at Docket No. 106, approving bidding procedures. The notice sets the sale objection deadline for November 18, 2026, at 4:00 p.m. Eastern; the bid deadline for November 24 at 12:00 p.m.; an auction on December 1 at 10:00 a.m.; and a sale hearing on December 3 and 4 at 9:30 a.m. The stalking horse is Disney Worldwide Services, Inc., under an agreement dated October 13, 2026. The notice doesn't mention user data or minors.

PPC Land reports that a 48-page form asset purchase agreement for the non-Disney assets was filed as Docket No. 136 on November 5, with a draft stamp of November 4. According to that report, the words user, minor, and chat don't appear in it, and its only consumer privacy reference sits in the definition of books and records. The report gives November 14 as the deadline to name an additional stalking horse and December 17 as the date after which either side may terminate if closing hasn't occurred, citing section 10.1(h).

Edition 38 recorded a notice of 40 jurisdictions asking for nine conditions on any sale, including limits on training with chat content and deletion before any data sale. Edition 40 gave November 20 as the bid deadline from earlier reporting; the company's notice gives November 24.

Study context. I didn't open the order, the docket, or the form agreement, and a form agreement can change before the sale hearing. The notice says nothing about user data, so the absence of user data terms in the form agreement is a draft issue, not a ruling on what a buyer may do.

Leadership implication. I'd ask your counsel to add a transfer clause to every AI vendor agreement at renewal. It should bar transfer of student data in a sale, merger, assignment, or insolvency without written notice, give the district a right to terminate and a deadline for deletion, and require the vendor to certify deletion. For each conversational tool your students reach, ask the vendor in writing what happens to stored student chats if the company is sold. File the answers beside the contract. Your board attorney shouldn't wait for a court calendar to learn what your agreement says.

3. A First Comparative Analysis of 36 State AI Guidance Documents Finds Vendor Evaluation in 83 Percent and a Review Cycle in 47 Percent, and Coverage That Does Not Track Each State's Own Student Populations

Source type. Preprint, not peer-reviewed; arXiv version 1 submitted November 4, 2026, categories cs.CY and cs.DL, comments field "Under Review."

Citation (APA). Memon, S. A., Gueorguieva, A.-M., Kumar, A., Wang, L. L., West, J. D., & Weber, N. (2026). U.S. state guidance on AI in K-12 education: A comparative content and citation analysis (arXiv:2611.06762) [Preprint]. arXiv. arxiv.org

The authors analyzed 36 guidance documents from 35 U.S. states and Puerto Rico, totaling 1,539 pages. Two coders applied 28 yes/no questions with 86.3 percent agreement and a Cohen's kappa of 0.73, and the authors compared coverage with state demographics, capacity, and partisanship using Mann-Whitney U tests, along with a citation and text reuse analysis.

Data privacy and academic integrity each appear in 97 percent of documents. Vendor evaluation appears in 83 percent and procurement or implementation in 78 percent. A roadmap or timeline appears in 56 percent, funding or budget in 47 percent, and a review cycle in 47 percent. Seventy-two percent have no version history. English learners appear in 50 percent of documents, rural or underserved students in 42 percent, and students are addressed directly in 22 percent. No state names parents as contributors, and only five name students.

The null results matter. Coverage of English learners wasn't associated with a state's English learner share (p = .35), nor coverage of disability with its share of students with disabilities (p = .83), nor coverage with free or reduced-price lunch share (p = .87). The abstract reports no relationship between priorities and partisanship. TeachAI is cited by 28 states, 78 percent of the documents, and some states nearly copy its model guidance.

Study context. Yes/no coding records whether a topic appears, not how well it's handled, and the study doesn't observe whether schools follow any guidance. It covers only the jurisdictions that issued guidance, and as a preprint under review, its numbers may change.

Leadership implication. I'd treat your state's guidance as an input, not your policy. Ask your policy committee to build a gap table with five rows: vendor evaluation, timeline, funding, review cycle, and supports for English learners and students with disabilities. In each row, mark what your board policy supplies that the state document lacks. Ask your state agency in writing for the guidance version history and next review date. Add a parent and a student to the committee that reviews your AI policy this year, because the study found no state that did.

4. A Randomized Trial of Automated Feedback to 1,030 Peer Tutors Changed Tutor Behavior by 8 to 15 Percent and Found No Significant Effect on Learner Practice Scores or Attendance

Source type. Preprint, not peer-reviewed; arXiv version 1 submitted November 4, 2026, categories cs.CL, cs.ET, and cs.HC.

Citation (APA). Demszky, D., Hicke, Y., Yun, J., Olson, M., Karpawich, M., Iqbal, A., Wu, M., Huang, J., Ravikumar, A., & Bent, D. (2026). Automated feedback to peer tutors increases their talk moves and learner reasoning but not the balance of talk (arXiv:2611.07072) [Preprint]. arXiv. arxiv.org

Two four-week SAT math bootcamps on Schoolhouse.world, each with eight 75-minute sessions, randomized tutors into a control arm of 399, a tutor feedback arm of 401, and two arms that added feedback to learners, 199 and 200. The analysis covers 1,030 distinct tutors, 7,443 learners, and 6,508 recorded sessions. Classifiers detected talk moves and GPT-4 Turbo wrote the suggestions. The authors describe the tutors as volunteers, mostly high school students.

In the pooled model, tutor feedback raised eliciting of learner ideas by 1.41 per hour over a control mean of 17.16, about 8 percent (p = .005), uptake of learner contributions by 0.52 per hour over 5.914, about 9 percent (p = .004), and learner reasoning by 0.24 per hour over 1.637, about 15 percent (p = .001). Tutor talk time didn't move: the control mean was 91.5 percent, and the effect was negative 0.001 (p = .803).

Learner outcomes were flat. The learner practice test effect was negative 0.007 (p = .171), attendance was positive 0.026 (p = .714), and session ratings and net promoter scores showed no significant effect. Adding feedback to learners didn't change instructional practice. Delivery was incomplete: 77.9 percent of treated learner sessions in the first bootcamp and 56.0 percent in the second. No actual SAT scores were available.

Study context. This is intention to treat under incomplete delivery, with volunteer tutors on one platform, four weeks of outcome data, and classifiers that detect whether a talk move occurred, not its quality. Learner age and grade weren't reported, so it isn't a classroom or district study.

Leadership implication. When a vendor sells feedback or coaching AI for teachers or tutors, I'd require a pilot clause that names a student outcome and a comparison before any renewal. Process measures such as more questions asked or more reasoning prompted don't count as success under that clause. Ask your assessment director to name the student measure and the date it's read, and ask your instructional technology director to refuse a renewal that offers only engagement or usage dashboards. Report the student result to the board beside the adoption numbers.

5. Google's November Report Calls a 2020 Consulting Assessment a Randomized Trial, Cites a UK Tutoring Trial Whose Registry Entry Shows No Results, and Reports Its Own Null Findings Without Effect Sizes

Source type. Vendor report, not peer-reviewed; Google, November 2026, undated beyond the month.

Citation (APA). Google. (2026, November). AI in education: From potential to practice [Report]. services.google.com. Brazão, V. (2026). Testing the efficacy of AI tutoring in secondary mathematics: A scaled RCT in UK classrooms (AEARCTR-0018079) [Trial registration]. AEA RCT Registry. socialscienceregistry.org

On Read Along in India, it reports fluency gains of 114 percent in Hindi, 44 percent in English, and 120 percent in Marathi against a control, and calls the study a randomized controlled trial on page 21. It cites a 2020 impact assessment by Sattva Consulting. On Guided Learning in Sierra Leone, it reports gains "equivalent to between one and two years of typical learning progress" from a randomized trial, citing a 2026 paper by the LearnLM Team at Google and Fab AI, and separately 0.26 standard deviations in a teacher-directed setting it calls early testing.

On the Eedi AI tutor in the UK, the report states "clear learning gains" on in-platform tasks and gives no effect sizes, citing an arXiv preprint and a 2026 source and describing a 10-week trial of nearly 1,200 students. The AEA registry holds an Eedi Labs trial of 1,200 planned students in Years 8 to 10 across 10 UK schools with four arms and a primary outcome on the STAR Renaissance Maths assessment. It was registered April 9, 2026; last updated June 30, 2026; shows status as in development; restricts its analysis plan; and posts no results. I can't confirm it's the trial the report cites.

The report also states that optional tutor-on-the-side pilots "failed to yield meaningful performance gains," with low uptake of 30 percent on Coursera and 39.3 percent in Study Hall, and gives no effect sizes for those null results.

Study context. This is a promotional synthesis written by the vendor, and most of the studies it cites were run by or with Google. I didn't open the cited studies, so I can't say whether the labels are accurate; I can say what the report calls them and what it cites.

Leadership implication. I'd change what your request for proposals asks for. Require each vendor to submit the study, not the summary: design, comparison condition, outcome measure, effect size, who ran it, and whether it was peer-reviewed. Score vendor-run evidence separately from independent evidence, treat a registry entry without results as a plan and not a finding, and ask for the null results the vendor holds. Assign your procurement lead and your evaluation lead to read the file together before the next renewal and record what they couldn't verify.

Emerging Strategic Themes

Theme 1. A default changed by blog post is not notice. Edition 37 called the off switch a procurement specification, and Edition 39 asked why it's sometimes missing. Google has now moved the default for student accounts in a post with no effective date. A contract that says nothing about notice of material change leaves your administrators to learn about it the way the public does. I'd add a notice clause, name the person who may flip each service, and keep a dated export of your settings.

Theme 2. Change of control is a contract event. Edition 38 recorded 40 jurisdictions asking a court for conditions on any Character.AI sale, and this week the court calendar set the dates. Whatever a vendor promises today passes to its buyer only if your agreement says so. I'd write assignment, insolvency, and sale triggers into every AI agreement, with a termination right and a deletion deadline, before the vendor you chose is a different company.

Theme 3. Copied guidance is not adopted governance. Twenty-eight states cite one nonprofit's model guidance, and 72 percent of guidance documents carry no version history. Edition 40 found state purchasing guidance adding evidence standards faster than anyone can check them. A copied document names the right topics and supplies no funding, owner, or review date. I wouldn't adopt state language as policy until your board has assigned the owner and the date.

Theme 4. Read the footnote before the number. A report's randomized trial can be a 2020 consulting assessment, a trial can be a registry entry without results, and a tool can change what tutors do without changing what students score. Edition 34 counted seven of sixteen products with independent achievement studies. I'd ask for the study behind every headline figure, and don't accept a process measure as a learning result.

What Was Not Found

No source states when Google's default-off change takes effect, whether a district's existing settings will carry over, or whether Search can be enabled without its AI features. I read Google's November 3 post and K-12 Dive's November 5 report and searched for Google's follow-up. The post says "coming months," K-12 Dive says the timing is unclear, and K-12 Dive reports it's unclear whether settings can be applied by grade or school. I'd call that a planning problem: districts that run Google Workspace for Education are being asked to plan without a date. New this week.

No source says what happens to children's stored chat records in the Character.AI sale. The company's notice doesn't mention user data or minors, and PPC Land reports that the form agreement for the non-Disney assets doesn't use the words user, minor, or chat. A court encouraged, but didn't order, a privacy ombudsman at the October 30 hearing, as recorded in Edition 40. I didn't open the order or the docket. This gap persists, with new dates.

No peer-reviewed K-12 randomized or quasi-experimental study of a student outcome carried a verified in-window date. The nearest designs are preprints. Demszky and colleagues found no significant learner effect from tutor feedback over four weeks. Favero and colleagues (arXiv:2611.07792, not peer-reviewed) randomized 65 Swiss International Baccalaureate students to a Socratic chatbot or a control chatbot: the composite essay score improved for 12 of 30 students in the Socratic arm, and 13 of 32 in the control, and the Socratic arm trailed on the cogency trait, 5 of 30 improving against 11 of 32. The authors call it exploratory. I'd flag that districts are buying purpose-built tutors on evidence that hasn't cleared peer review. Persistent gap.

No in-window outcome study measured effects on English learners or students with disabilities. The state guidance analysis reports English learners in 50 percent of documents, and finds coverage unrelated to a state's English learner share (p = .35) and disability share (p = .83). I searched for outcome studies by these populations and found none dated in the window. I'd note that accommodation policy is being set for these students without outcome data. This persistent gap carried over from Edition 40.

No independent K-12 evaluation of classroom surveillance, engagement detection, or proctoring appeared in the window. I ran the counter-evidence sweep for emotion recognition, attentiveness scoring, and proctoring and found no in-window study with learning or fairness outcomes. Vendors report detection accuracy, but what a flag does to the flagged student is unmeasured. Persistent gap.

No Georgia district AI instrument could be verified inside the window. The Clayton County work session of October 26 placed Policy Series A, B, C, and IHE for first read; the summary lists no policy codes and doesn't mention AI, and its November 2 agenda carried no policy, AI, or technology items. The Atlanta Public Schools November 2 agenda was behind a bot check I couldn't pass, so I can't say whether it took up its instructional AI policy; the district calendar lists the Policy Committee on November 18 and the board on December 7. I found no record of a Fulton Policy IFBI in its bulletins through October 23. A Forsyth County work session item reported on October 13 rests only on a meeting tracker. I'd treat it as open until a board record says otherwise. Carried forward.

The pattern is that vendor decisions and sale calendars are moving faster than the evidence and the contract language needed to meet them. A default changes without a date, a draft sale agreement is silent on children's chats, and the one tutoring tool Google's report most wants districts to notice has a registry entry and no results. The correct response isn't delay. It's monitoring, documentation, and exit ramps: a dated export of every student-facing AI setting, a transfer clause in every agreement, and an end date for each pilot that doesn't produce its own comparison data.

Novo Executive Summary

On November 3, Google said Gemini, Gemini Notebook, Gemini in Classroom, and Google Search will be off for K-12 student accounts until administrators turn them on, without giving a date, while a Delaware court order set November 24 as the bid deadline for a company whose draft sale agreement is silent on children's chats. In the same week, an analysis of 36 state guidance documents found vendor evaluation in 83 percent and a review cycle in 47 percent; a randomized trial found that automated feedback changed tutor behavior, not learner practice scores; and a vendor report labeled a 2020 consulting assessment a randomized trial. None of this is solved by choosing a safer product. It's solved through architecture: decision rights over each AI service and the person who may change it, a procurement specification with notice, transfer, and evidence clauses, an evaluation plan with a student measure, and role-based literacy pathways that teach staff to read a footnote. Novo Innovative Pathways works with district leaders to build that governance architecture, role-based AI literacy, and implementation strategy so that a district can answer a regulator, board member, or parent about its own settings and contracts first.

Watch This Week

• Later in November 2026, Google. The post says a Guided Learning pilot in Search starts in select U.S. schools later this month and that specifics on the default-off controls will follow in the coming months. I found no public Google reply to the senators' November 2 request, and no source that connects that letter to this announcement. Carried forward from Editions 39 and 40.
• Saturday, November 14 through Thursday, December 17, 2026, U.S. Bankruptcy Court, District of Delaware, Case No. 26-11603 (BLS). Per PPC Land, November 14 is the deadline to name an additional stalking horse; per the company's notice, objections are due November 18 at 4:00 p.m. Eastern, bids are due November 24 at noon, the auction is December 1, and the sale hearing is December 3 and 4; the form agreement allows termination if closing hasn't occurred by December 17. Correction: Edition 40 gave November 20 as the bid deadline. Carried forward from Editions 38 and 40.
• Monday and Tuesday, November 9 and 10, Kansas State Board of Education; Tuesday and Wednesday, November 10 and 11, Paris Peace Forum; Los Angeles Unified Generative AI Ad Hoc Committee. The Kansas page lists regular business meetings in Topeka on both days, and I couldn't confirm an AI agenda item. The Paris forum homepage lists a Child Summit focus, and I didn't find a session for the Coalition for Children's Rights and Protection in the Age of AI. Edition 40 gave November 13 for the Los Angeles committee; a published schedule reported by LAist lists December 9 as its next meeting, and I couldn't confirm November 13. Carried forward from Editions 35 through 40.
• Wednesday, November 18, Atlanta Public Schools Policy Committee, 12:15 p.m., and Monday, December 7, the monthly board meeting. The November 2 agenda wasn't readable, so it's unclear whether the instructional AI policy was taken up. Also dated: Forsyth County work session November 10, Cobb November 12, Gwinnett November 17 and 19, Cherokee November 19. No AI item is confirmed on any of them. Carried forward from Editions 32 through 40.
• Wednesday, December 2 and Wednesday, December 16, 2026, Anne Arundel County Public Schools. WMAR reported an October 21 introduction of the draft AI policy with a second reading on December 2 and a final adoption vote on December 16. The Howard County adoption reported by GovTech in Edition 40 remains unresolved at the board record. Carried forward from Editions 35 through 40.
• Before Friday, January 1, 2027, Colorado Attorney General and New York. The Colorado page, last modified October 27, shows no adopted rules after the October 26 hearing, and the rules must be adopted before the statutes take effect January 1. A Holland & Knight alert dated November 5 says New York's S9051-B takes effect January 1 and that the Attorney General may name further unsafe features by regulation, with no timeline. Microsoft's AI Safety and Privacy Standard has been available to schools since November 1. Carried forward from Editions 33, 34, and 38 through 40.

Questions District Leaders Are Asking

What did Google announce about Gemini and Search for K-12 students?

On November 3, 2026, Google's vice president of education, Chris Phillips, wrote that access to Gemini, Gemini Notebook, Gemini in Classroom, and Google Search will be entirely off for K-12 student accounts until school administrators explicitly turn them on. The post gives no effective date and says Google will share more specifics with schools in the coming months. It is a vendor announcement, not a law or a contract term.

When does Google's default-off change for student accounts take effect?

No public source gives a date. Google's post says only that specifics will come in the coming months, and K-12 Dive reported on November 5 that the timing is unclear and that it is unclear whether settings can be applied by student, grade, or school. No source found this week says whether existing district settings will carry over or whether Search can be enabled without its AI features.

What should a school district do before Google's default change?

Export the current organizational unit settings for Gemini, Gemini in Classroom, Gemini Notebook, and Search for every student unit and file the export in the contract record as a baseline. Send Google or your reseller a written request for the effective date, whether existing settings will be kept or reset, whether Search can be enabled without its AI features, and how notice of the next default change will reach administrators. Name who may turn each service on, and add a notice of material change clause to the next renewal.

What are the deadlines in the Character.AI bankruptcy sale?

The company's November 5, 2026 notice of a Delaware bankruptcy court order in Case No. 26-11603 (BLS) sets a sale objection deadline of November 18 at 4:00 p.m. Eastern, a bid deadline of November 24 at noon, an auction on December 1, and a sale hearing on December 3 and 4. PPC Land reports November 14 as the deadline to name an additional stalking horse and December 17 as the date after which either side may terminate if closing has not occurred. The notice does not mention user data or minors.

What did the analysis of 36 state AI guidance documents find?

Memon and colleagues (arXiv:2611.06762, a preprint under review) coded 36 guidance documents from 35 states and Puerto Rico, 1,539 pages in total. Data privacy and academic integrity each appear in 97 percent, vendor evaluation in 83 percent, funding in 47 percent, and a review cycle in 47 percent, and 72 percent have no version history. Twenty-eight states cite TeachAI's model guidance, and coverage of English learners and disability is unrelated to each state's enrollment share of those students.

Did automated feedback to tutors improve student scores?

Not measurably. Demszky and colleagues (arXiv:2611.07072, a preprint) randomized 1,030 volunteer tutors in two four-week SAT math bootcamps on Schoolhouse.world. Automated feedback raised tutors' eliciting of learner ideas by about 8 percent, uptake by about 9 percent, and learner reasoning by about 15 percent, while learner practice test scores, attendance, and session ratings showed no significant effect.

Sources

Governance and Policy

Phillips, C. (2026, November 3). Responsible AI for schools with Guided Learning. Google. blog.google

Character.AI. (2026, November 5). Chapter 11 update and transaction notice. Character.AI Blog. blog.character.ai

Rijo, L. (2026, November 6). Character.AI sale contract omits 40 attorneys general's user data conditions. PPC Land. ppc.land

Merod, A. (2026, November 5). Google will turn off its search engine for schools by default. Is it an AI? K-12 Dive. k12dive.com

Colorado Department of Law. (2026). Artificial intelligence. coag.gov

Holland and Knight. (2026, November 5). New York enacts AI chatbot safeguards for minors. hklaw.com

Atlanta Public Schools Board of Education calendar, atlantapublicschools.us, and Clayton County Public Schools board agendas and summaries for October 26 and November 2, 2026, clayton.k12.ga.us

Kansas State Department of Education, state board meeting page, and Paris Peace Forum 2026 homepage.

Research, Peer-Reviewed

None this week. No peer-reviewed K-12 study with a verified publication or online date inside the research window passed verification as a signal. Two Frontiers in Education articles with Greek, Portuguese, Spanish, and Kazakh samples carried conflicting online dates and measured perceived learning or teacher self-report, and a third paper's publisher page didn't load; none is cited here.

Research, Preprint (Not Peer-Reviewed)

Memon, S. A., Gueorguieva, A.-M., Kumar, A., Wang, L. L., West, J. D., & Weber, N. (2026). U.S. state guidance on AI in K-12 education: A comparative content and citation analysis (arXiv:2611.06762) [Preprint, not peer-reviewed; under review]. arXiv. arxiv.org

Demszky, D., Hicke, Y., Yun, J., Olson, M., Karpawich, M., Iqbal, A., Wu, M., Huang, J., Ravikumar, A., & Bent, D. (2026). Automated feedback to peer tutors increases their talk moves and learner reasoning but not the balance of talk (arXiv:2611.07072) [Preprint, not peer-reviewed]. arXiv. arxiv.org

Favero, L., Käser, T., Pérez-Ortiz, J. A., & Oliver, N. (2026). Thinking is not writing: Designing AI chatbots to connect reasoning and revision (arXiv:2611.07792) [Preprint, not peer-reviewed]. arXiv. arxiv.org

Institutional Report (Not Peer-Reviewed)

Google. (2026, November). AI in education: From potential to practice [Report; vendor report, not peer-reviewed]. services.google.com

Brazão, V. (2026). Testing the efficacy of AI tutoring in secondary mathematics: A scaled RCT in UK classrooms (AEARCTR-0018079) [Trial registration, not a result and not peer-reviewed]. AEA RCT Registry. socialscienceregistry.org