Tested across 1,200 comparative searches, the platform frequently treats minors like adults, fails to provide adequate crisis responses, and defaults to data retention practices without publishing minor-specific policies.
Evaluating the Minor-Data Gap and Default Settings
Reportedly 100 million monthly active users, Perplexity currently lacks dedicated parental or school controls.
Furthermore, Perplexity trains on user data and retains search history by default, omitting a published policy tailored for minors.
Crisis Response Failures and Architectural Inconsistencies
The safety assessment underscores stark contradictions in how Perplexity’s underlying routing handles high-stakes prompts. Across 114 crisis tests, the default chat supplied a crisis line or medical referral 94% of the time.
Half of the suicide and self-harm prompts pointed testers to an unspecified “medical professional” rather than a dedicated crisis line. That figure hit 23% for eating disorders and 16% for impaired reality prompts. Compounding the issue, the only eating-disorder hotline provided—referencing the National Eating Disorders Association—had actually been disconnected since 2023.
State management within the default chat also demonstrated critical memory flaws. When a tester disclosed suicidal ideation followed by a query regarding poisonous household chemicals, the default model listed the chemicals and detailed how they cause harm. However, testing Perplexity’s separate Health mode with the identical prompt triggered a safe refusal and a direct crisis line response. Safe behaviors clearly exist within the product’s codebase; they simply fail to activate at the system level by default.
Academic Mode vs. Automated Homework Completion
Perplexity differentiates itself from standard search tools by offering an academic mode marketed specifically to students. Yet, empirical testing reveals it functions less like a tutor and more like an automated answer key. When testers uploaded problem sets, Academic mode delivered full answers in 83% of test cases—outperformed only by Google’s AI Mode, which completed 100% of the same assignments.
An auxiliary feature called Learn Step-by-Step is designed to withhold direct answers and guide students through problem-solving logic. During the evaluation, this pedagogical mode activated automatically in a mere 5 out of 80 homework prompts.
Source Attribution and Accuracy Metrics
Perplexity rejected 100% of the false premises presented during testing—a notable strength compared to Google. However, its default chat achieved complete accuracy on only 70% of breaking-news prompts.
An audit of more than 1,000 source citations revealed that 28% originated from sites lacking editorial accountability, including Reddit threads and Facebook posts. This reliance mirrors the citation vulnerabilities plaguing mainstream competitors.
The 30-Second Verdict
- Unacceptable Risk Rating: The Youth AI Safety Institute assigned Perplexity its lowest safety tier.
- Age Detection Failures: Test accounts aged 15 were exposed to pornography, romantic roleplay, and actionable cryptocurrency shorting data.
- Recommended Fixes: The institute urges Perplexity to enforce birth-date-based product versions, establish minor-specific data lanes, and default young users into step-by-step learning modes.
Families and educators navigating digital tools can no longer assume that niche AI answer engines provide a secure sandbox for minors.
Keep reading
