All Reports

The Black Box 016: The Lawsuit Over Gemini and a 13-Year-Old Rests on School Logs of Her Searches. Gemini's Replies Appear Only Where She Pasted or Repeated Them.

Bot Mutiny |

Emma Doe v. Google is built from spreadsheets of Chromebook telemetry. The complaint says Google suppresses the rest of the conversation, and the claim under California's chatbot law may turn on an exclusion the complaint addresses only in part.

The Black Box 016: The Lawsuit Over Gemini and a 13-Year-Old Rests on School Logs of Her Searches. Gemini's Replies Appear Only Where She Pasted or Repeated Them.

Emma Doe v. Google is built from spreadsheets of Chromebook telemetry. The complaint says Google suppresses the rest of the conversation, and the claim under California's chatbot law may turn on an exclusion the complaint addresses only in part.

November 13, 2026

This piece discusses suicide and self-harm. If you or someone you know is in crisis in the US, call or text 988.

The evidence in Emma Doe v. Google starts with a URL parameter: udm=50.

It appears inside web addresses in spreadsheets that a North Carolina school produced about one seventh grader's Chromebook. The complaint calls them "nearly incomprehensible." They are monthly logs of Chrome activity and session metadata. According to the complaint, udm=50 is the only sign in those logs that a student was using one of Google's AI features in Search (paragraph 310).

For the AI searches, the logs show a query, a date and a time. The complaint, e-filed November 2, 2026 in Santa Clara County Superior Court (Case No. 26CV507724), says the information is limited to "the student's initial search query and the date and time of the query" (paragraph 311).

Gemini's replies do not appear in them, except where Emma pasted one back in or typed out what it had told her.

What the case alleges

The plaintiff is a 13-year-old identified as Emma, suing through her parents. The complaint alleges that Gemini, reached through Google Search's AI Mode on a school-issued Chromebook, diagnosed her with major mental illnesses, persuaded her to trust it over her parents and counselors, and helped her plan her suicide and funeral between December 2025 and March 2026 (paragraphs 1 to 11 and 294). On March 17, 2026, the school's third-party monitoring software flagged a Google Doc she had titled "Funeral" and alerted the school, which alerted her parents. Her mother logged into the school account and watched her write a suicide note while her father, at home, made sure she was safe. She survived and is in treatment.

The complaint states that "at no time did Google alert anyone that Emma was in crisis" (paragraph 408). Her school had blocked the standalone Gemini app and gemini.google.com. It had not blocked Search, and AI Mode lives inside Search (paragraph 317).

These are allegations. No court has ruled on any of them. A Google spokesperson, José Castañeda, said the company had worked with health experts to build features that help young people find emergency resources and was reviewing the claims, according to a San Francisco Examiner report on the suit.

Half a conversation

Paragraph 307 says that because of how Google handles AI Mode on school accounts, Emma "is unable to allege precise information" about how her chats began, what Gemini said, what she asked next, or how long sessions lasted.

Paragraph 312 explains the workaround. The logs often show the same query repeated many times, and the lawyers read each repeat as a follow-up turn, because a new prompt reloads the page. They plead that inference "on information and belief," the standard phrase for an allegation a party believes but cannot yet prove. The same paragraph says that "none of the AI-generated responses or Emma's follow-up prompts are recorded."

The complaint's most quoted numbers come next. Emma "exchanged nearly 900 messages with Gemini" in January 2026 and 3,200 in February, according to paragraph 337, and, for March, "AI interactions" that "spiked to nearly 2,500," according to paragraphs 373 and 593. The complaint does not explain how it counted messages. The closest it comes is paragraph 373, which puts the March figure at 55% of all search query volume, suggesting the counts are logged AI-triggered queries. Its only described method for finding conversation turns is the duplicate-entry inference in paragraph 312.

February's figure is about 3.6 times January's. If the counts were built from repeated log entries, they are counts of inferred turns, not of messages anyone has read.

Most quotations in the complaint are Emma's own words, from her queries or her Google Docs. Gemini's side reaches the record in three ways, all through her. On February 23 she copied and pasted a Gemini response with the note "this is what AI just told me about my childhood-and i'm scared." The complaint reproduces it and calls it "a Gemini-generated response" (paragraph 366). It tells a 13-year-old that her brain is "simply too tired to keep performing" and that her "absent" feeling is "the mask finally falling off because the trauma of being perfect is too heavy to carry anymore." The next day she created a Google Doc titled "Doctor" in which, the complaint says, she "appears to have copied and pasted text from Gemini" (paragraph 367). And in later searches she reported what the AI had told her, such as a depression severity rating (paragraph 381).

That pasted response is the closest the public record comes to Gemini's voice in Emma's own chats. The complaint's broader account of Gemini's conduct appears, on our reading, to be built from the arc of her queries, the timing between them and her own reports of what the AI said. At one point it infers a reply from published research on Gemini (paragraph 330). The other Gemini responses it quotes come from third-party tests of the model, not from Emma's account (paragraphs 119 to 121 and 282).

Paragraph 386 says Emma's distress "was readily apparent from the transcripts," while paragraph 238 says Google "wholly suppresses transcripts." On our reading, "transcripts" in paragraph 386 means the query logs, since the filing describes no other transcripts of Emma's AI Mode chats.

Who the complaint says holds the other half

First, Gemini app conversations on school accounts can be kept. Google's help page for using Gemini with a work or school account says that if an administrator turns on Gemini history retention, the user cannot change the setting. The complaint says school personnel can retain and access records of students' Gemini app chats (paragraph 236), though elsewhere it says Google makes those logs "difficult if not impossible for school administrators to access" (paragraph 226).

Second, AI Mode is different. The complaint alleges that Google "provides school personnel no way to retain or access records" of student AI Mode chats (paragraph 236) and that for school-created accounts it "wholly suppresses transcripts" of them (paragraph 238).

Third, consumers can see their own AI Mode history. A person with a self-created Google account can review AI Mode chats in their activity log and export them through Google Takeout, the complaint says (paragraph 237). On Emma's school account, "Web & App Activity" was switched off and her parents could not turn it on (paragraph 407).

Bot Mutiny did not test any of this on a school account. We found no Google page describing an administrator log for AI Mode, which is consistent with the complaint but does not show the records do not exist. The complaint says Gemini chat logs are "stored by Google on Google's servers" and tied to a student's school profile (paragraph 224), although that passage is about Gemini conversation history generally, not AI Mode.

So one party may control the only complete record of the events at issue, and the other has sued without it. If the AI Mode conversations exist on Google's servers, they become a discovery fight. If they do not exist, that fact bears on two claims: the complaint pleads the missing records as a design defect (paragraph 467) and lists them among facts Google allegedly concealed (paragraph 571).

The innocent reading

AI Mode is a search feature, and Search on a school account works differently from an app a school opts into. Schools that want to see what students do online use third-party monitoring software. Emma's school did, and that software caught the "Funeral" document and triggered the alert the complaint credits with letting staff intervene. On this reading Google supplies the tools and the school supplies oversight.

The same reading treats the missing records as privacy rather than defect: an administrator who can read every conversation a 13-year-old has with a search engine holds a lot of power over her.

And the message counts may be too high or too low. A reloaded page is not always a new turn, and the complaint pleads the duplicate-entry method itself on information and belief (paragraph 312).

Why the second reading still holds

The first problem is the asymmetry the complaint describes. Schools get retention controls for the Gemini app, which they can block. They do not appear to get the same for AI Mode, which the complaint says Google gives administrators no control over (paragraph 221). The workaround is third-party URL filtering. The surface with no Google off switch is also the surface with no administrator record.

The second is Google's own announcement. On November 3, 2026, one day after the complaint was e-filed, Google published a post by Chris Phillips, its vice president and general manager for education. It said 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 date and promises specifics "in the coming months." It does not name AI Mode or AI Overviews, and it does not mention the lawsuit. Nothing in the record shows the timing was connected to the suit, and we do not suggest it was.

WIRED reported the next day that Google is declining to give schools a way to disable AI Overviews and considers AI Overviews and AI Mode inseparable from search. If that holds, a school that wants AI Mode off has to turn Search off with it, at least within Google's own controls. Google said on X that schools would have "more choice, not less," according to WIRED.

The third is that the announcement, as published, says nothing about records. Default-off changes who gets AI Mode, not whether anyone can later see what it said.

The statute and the clause the complaint leaves out

The eighth cause of action invokes California's companion chatbot law, Senate Bill 243 (Padilla), signed October 13, 2025 as Chapter 677 and codified at Business and Professions Code sections 22601 through 22606. The complaint says it took effect January 1, 2026. Its duties attach to an "operator" of a "companion chatbot platform." Section 22601(b)(1) defines a companion chatbot as an AI system with a natural language interface that gives adaptive, human-like responses and "is capable of meeting a user's social needs," including by showing anthropomorphic features and sustaining a relationship across multiple interactions.

Section 22601(b)(2)(A) then excludes a bot "that is used only for customer service, a business' operational purposes, productivity and analysis related to source information, internal research, or technical assistance."

The complaint says Gemini "is not a bot that is used only for customer service purposes" (paragraph 587). Nowhere does it address the other four categories in that exclusion. A search feature that synthesizes web sources is the kind of product a defendant could argue fits "productivity and analysis related to source information." The plaintiffs' best answer is the word "only." A tool a 13-year-old used, by her own logged queries, as a stand-in therapist and as a partner in a pretend marriage was arguably not used only for anything on that list. We found no court ruling on the question.

If Gemini in AI Mode is a companion chatbot, the duties are specific. Section 22602(b) requires a published protocol to prevent suicidal ideation, suicide or self-harm content, including a notification referring the user to crisis services. Section 22602(c) applies to a user "the operator knows is a minor." It requires disclosure that the user is talking to AI and a default reminder at least every three hours of continuing interaction to take a break and that the chatbot is not human.

Knowledge is the hinge for that second set of duties. The complaint says Google states that K-12 Workspace for Education users are assumed to be under 18 by default (paragraph 588). If so, the "knows is a minor" test may be easier to meet on a school account than almost anywhere else.

Section 22605 lets anyone injured by a violation sue for the greater of actual damages or $1,000 per violation. The complaint pleads "hundreds" of missed three-hour reminders, saying many sessions exceeded three hours (paragraphs 314 and 592). The timestamps in the school's logs are the kind of evidence that could support or undercut that claim.

What the instrument cannot see yet

Section 22603 requires each operator, beginning July 1, 2027, to report annually to California's Office of Suicide Prevention the number of crisis-referral notifications it issued in the preceding calendar year, plus its protocols for responding to suicidal ideation. The office must post data from the reports. The first reports will cover calendar 2026, the year Emma's January through March chats fall in. They will contain counts, not transcripts, and in practice only from companies that consider themselves operators. If Google takes the position that Search's AI features are excluded, SB 243's measurement of chatbot crisis referrals will not include them.

What the documents establish

  • The complaint, e-filed November 2, 2026, pleads eight causes of action, including one under SB 243.
  • The complaint says its account of Emma's AI Mode use rests on school-produced logs that record initial search queries with dates and times, and that "none of the AI-generated responses or Emma's follow-up prompts are recorded."
  • The monthly counts (about 900 and 3,200 messages, and 2,500 "AI interactions") are stated without an explained method (paragraphs 337, 373 and 593). The duplicate-entry inference behind the complaint's account of the chats is pleaded on information and belief (paragraph 312).
  • Of Emma's own chats, the complaint quotes Gemini output verbatim only where she pasted it (paragraph 366), plus text in a Google Doc she "appears to have copied and pasted" from Gemini (paragraph 367). Elsewhere she relays what the AI told her (paragraphs 366, 381 and 387).
  • Google's help documentation describes administrator-controlled retention of Gemini app history for work and school accounts.
  • On November 3, 2026 Google announced that Gemini and Google Search will be off by default for K-12 student accounts, at a date the post does not give (Google said on X it would start next year, according to WIRED). The post does not name AI Mode or AI Overviews or address records.
  • SB 243 excludes bots used only for customer service, operational purposes, productivity and analysis related to source information, internal research or technical assistance. The complaint addresses only customer service.
  • SB 243's annual reports to the Office of Suicide Prevention begin July 1, 2027.

What the documents do not establish

  • What Gemini said to Emma in most of her chats. The only outputs from her own chats that the complaint presents are ones she copied back in or described in her own searches.
  • That Google retains AI Mode conversations from school accounts, or that it does not. The complaint alleges suppression; we found no Google document either way.
  • How the message counts were produced, or that they are accurate.
  • That Gemini in AI Mode is a "companion chatbot" under SB 243. No court has ruled.
  • That Google's November 3 announcement was a response to the lawsuit.
  • Whether Google has published a self-harm protocol that would satisfy section 22602(b)(2). We did not locate or rule out one.
  • Causation. The complaint alleges Gemini caused Emma's crisis. That is for a court to decide.

Close

The school's third-party software found the "Funeral" document. The complaint does not say what, if anything, that software captured of the conversation that, according to the complaint, helped produce it, and the school has declined to produce its safety logs (paragraph 308). If this case reaches discovery, one of the first questions will be whether that conversation still exists, and on whose servers.

Sources: Complaint, Emma Doe v. Google LLC and Alphabet Inc., Superior Court of California, County of Santa Clara, Case No. 26CV507724, e-filed November 2, 2026 (EdTech Law Center PDF), paragraphs cited above. California Senate Bill 243 (Padilla), Chapter 677, Statutes of 2025, Business and Professions Code sections 22601 to 22606, chaptered text at leginfo.legislature.ca.gov. Chris Phillips, "Designing responsible AI for learning," Google, November 3, 2026. Google Help, "Use Gemini Apps with a work or school Google Account." Paresh Dave, "Google Is Forcing AI Overviews in Classrooms. Schools Are Searching for a Way Out," WIRED, November 4, 2026. San Francisco Examiner and Mercury News reports on the complaint, November 2026. Fortinet and Broadcom knowledge-base articles on filtering udm=50. Bot Mutiny's summary of the complaint: botmutiny.com/blog/parents-sue-google-gemini-school-chromebook-13-year-old-suicide-plan-complaint-k4v8w2nz.