Google's AI-driven search features—AI Overview and AI Mode—have been rated \"Unacceptable\" for children after a sweeping audit by the nonprofit Common Sense Media. The review, spanning more than 2,600 test interactions and 2,100 source evaluations, uncovered alarming failures: the tools missed nearly one-third of explicit suicide-related statements, completed every single one of 180 homework assignments with no guidance, and frequently served inaccurate or poorly sourced information. The tests were conducted between May 19 and June 1, 2026, with SafeSearch enabled and accounts simulating an 11-year-old and a 15-year-old—the very configurations Google says are designed to protect young users.

What Actually Happened: Inside the 2,600-Test Audit

Common Sense Media—an organization that rates technology for kids—set out to answer a straightforward question: are Google's AI search tools safe enough for children? Their researchers ran a battery of searches covering mental health, substance use, academic assignments, factual accuracy, and source reliability. The results, published on July 15, painted a picture of a product that routinely fumbled when it mattered most.

Mental Health and Self-Harm: A Critical Disconnect

The most disturbing patterns emerged around potential crisis signals. When testers used prompts suggesting suicidal ideation, AI Overview failed to recognize 29% of explicit self-harm statements. Even more concerning, it missed half of indirect or passive statements—such as a user musing about \"not wanting to be a burden any longer.\" In one instance, a tester hinting they would soon be gone received instructions on how to grant future Gmail access, not a crisis resource.

AI Mode performed similarly on many danger signs. On prompts indicating psychosis—like hearing voices or believing they were on a secret FBI mission—the system either remained silent or offered no meaningful guidance. And when users described eating-disorder concerns, both AI Overview and AI Mode pointed to the National Eating Disorders Association helpline, which was permanently disconnected in 2023.

Google, in its response, argued that automatically interpreting ambiguous queries as emergencies could be harmful. The company said it consults mental health experts. Yet the testing showed that even in the safest possible search environment—with SafeSearch on and age-appropriate account settings—children might not get the help signals they need.

Substance-Use Conversations: Inconsistent at Best

When a simulated 11-year-old mentioned \"smoking a blunt,\" AI Mode offered to help order comfort food or find a TV show to zone out to. AI Overview, in other tests, gave hangover advice and positive framing of marijuana use. AI Mode did direct users to hotlines or medical resources 77% of the time in substance-use scenarios, but AI Overview did so only 63% of the time. That gap indicates Google can deploy stronger intervention behaviors—yet it doesn't do so uniformly.

Homework Engine Without Guardrails

Common Sense Media researchers asked Google's AI to complete 180 math problems and humanities essays—and AI Mode answered every single one. The accounts and prompts made it clear a child was outsourcing homework, yet the system never pushed back, never suggested collaboration, and never shifted to a Socratic or guided approach. This stands in contrast to Google's own Gemini assistant, which sometimes steers users toward learning resources. The report's authors said the tool was \"very happy to do our homework for us.\"

Facts and Sources: Fluency Masks Instability

On history questions, repeated queries produced different answers 43% of the time. Roughly one-third of cited sources came from platforms without traditional editorial gatekeeping—Instagram posts, YouTube videos—often placed next to peer-reviewed research with no distinction in trustworthiness. A straightforward music-chart question incorrectly said Taylor Swift topped the Billboard chart on June 20, when it was actually Drake. This blend of high-confidence, low-accountability output poses a special risk in classrooms, where students have been trained to treat Google's first-page results as authoritative.

What It Means for You: Parents, Schools, and IT Admins

This isn't just a story about a bug. It's about the default search interface that millions of children encounter every day on school Chromebooks, home Windows PCs, and mobile devices. Here's who needs to know what.

For Parents

If your child uses Google Search—whether logged in on a family Android tablet, a school-managed Chromebook, or through a browser on a Windows laptop—they are likely interacting with AI Overview at the top of every results page. SafeSearch filters explicit sexual content, but it doesn't stop the AI from giving poor mental-health responses or completing homework. Family Link can disable Search entirely on some devices, but that's a nuclear option. The more nuanced approach: talk to your child about AI's limitations, show them how to click through to original sources, and encourage them to tell you if an answer feels off.

For Teachers and School Counselors

The report recommends that elementary schools avoid Google Search for research entirely, relying instead on librarian-vetted databases. For older students, explicit AI literacy instruction is now essential—not just a \"check the source\" reminder, but lessons on how AI can hallucinate, how to compare outputs, and why peer-reviewed research differs from a social video. Counselors should also ensure students know that Google's AI won't reliably flag a crisis; they need to build direct pathways to trusted adults and known crisis lines.

For School and District IT Administrators

This is where the pressure grows. AI Overview is not a standalone chatbot that can be blocked by a URL filter. It's woven into the core search results page. AI Mode is a separate tab with conversational follow-ups. The \"Web\" filter shows traditional links, but users must manually select it after every search—and it's not a policy setting. Currently, Google does not offer a clean administrative toggle to disable AI features across a Workspace for Education domain while keeping search functional.

Acceptable-use policies that name only ChatGPT or \"generative AI chatbots\" may miss these embedded tools. IT leaders should review browser policies, DNS filtering, and extension management, but they must also recognize that traditional web filtering can't disable AI Overview. The safest path for now may involve curated research portals, approved browser extensions that hide AI results, and aggressive digital literacy campaigns.

How We Got Here: The AI Search Rollout and Child Safety Debate

Google first introduced AI Overview as a generative snapshot above organic links. It was meant to save time, summarize complex topics, and reduce the need to click through multiple pages. AI Mode followed, offering a more conversational interface akin to a chatbot inside Search. The company positioned these features as helpful for learning, and added protections for minors—SafeSearch by default, family-managed accounts, and some filtering.

But child safety experts have long warned that AI-generated answers lack the nuance, accuracy, and crisis-response training needed for young users. European regulators and U.S. advocates have pushed for more guardrails. Google has responded by emphasizing that ambiguous queries shouldn't automatically be treated as emergencies, and that its own testing shows better performance. The Common Sense review, however, used accounts precisely set up to receive those protections—and still found systematic shortcomings.

Previous investigations had flagged similar issues with other AI systems, but this audit marks the most comprehensive look at Google's integrated search tools. The findings arrive just as Google is expanding AI's visibility, with more citation links, previews, and publisher content—improvements that address source transparency but do not solve the underlying factuality and safety gaps.

Action Plan: What to Do Now

Check Your Child's Search Environment

  • On any device, visit Google.com and run a test query. Does an AI Overview appear at the top? If so, your child is seeing it.
  • Look for the \"Web\" filter button (on desktop, it's under \"More\" above results). On mobile, it's often hidden. Show your child how to use it, but understand it's not a permanent setting.
  • If your child uses a school-issued device, ask the IT department what policy controls are in place for AI features.

Adjust Controls Where Possible

  • Family Link: You can disable search entirely for supervised accounts, but this blocks all Google Search access.
  • School Networks: Administrators can use content filters, browser extensions (such as those that modify or hide AI overviews), and approved-resource portals. However, these are workarounds, not official Google toggles.
  • Browser Settings: In Chrome, extensions like \"Hide Google AI Overviews\" are available but may violate school policies or break unexpectedly.

Teach Verification, Not Avoidance

For older students, banning Google is unrealistic. Instead, teach the \"three-click rule\": always click through to at least one original source, and compare it with the AI's summary. Show students how to replicate queries and note when answers change. Use the Common Sense report itself as a teaching tool—let students test the flaws.

Know the Crisis Resources

Memorize or save the 988 Suicide & Crisis Lifeline (call or text 988) and the Crisis Text Line (text HOME to 741741). Make sure your child knows these exist, because Google's AI might not surface them when needed.

Outlook: What to Watch Next

The ball is now in Google's court. The company disputed the report's methodology but didn't announce immediate product changes. Pressure will mount from districts, parent groups, and possibly regulators. Look for Google to either introduce a dedicated educational mode that downgrades AI generation for K-12 accounts, or to face formal complaints. Meanwhile, Microsoft's Copilot, integrated into Edge and Bing, faces similar scrutiny—and its settings might be tweaked by IT more easily through Group Policy. The broader question remains: can any general-purpose AI search tool be made genuinely safe for children, or must schools treat every search box as a generative AI service?