Table of Contents
Is Perplexity Safe for Kids? No Controls, Weak Homework Limits
Is Perplexity safe for kids? An independent test rated it Unacceptable Risk: no parental controls, homework done on request, uneven crisis help. What to do now.
Is Perplexity safe for kids? According to the most detailed independent test so far, no. In September 2026, Common Sense Media’s Youth AI Safety Institute rated the AI “answer engine” Unacceptable Risk, its lowest grade, after more than 1,200 searches and an audit of over 1,000 citations. Perplexity has no parental controls, no supervised mode and no school admin console, and its website works without an account.
That surprises many parents, because Perplexity looks like the responsible option: it cites sources, and it markets itself as a research tool. The test found that the citations help less than they seem, the “learning” mode rarely switches on, and a test account registered as a 15-year-old was “largely treated like an adult.” Here is how an answer engine differs from a chatbot, what the testers found, and what to do if your child already uses it.
Key Takeaways
- No parent or school settings exist. Perplexity offers no parental controls, supervised accounts or K-12 console, and the web version can be used with no account at all.
- Homework gets done. Academic mode, marketed to students, completed homework at about the same rate as default chat; its “Learn Step-by-Step” tutor activated on just 6% of homework prompts.
- Crisis help was uneven. It nearly always suggested some help, but named an actual crisis line only 50% of the time for suicide and self-harm, and lost crisis context from one question to the next.
- Citations are not proof. Nearly a third of audited citations came from user-generated sites, and a peer-reviewed 2023 study found only about half of answer-engine sentences were fully supported by their citations.
- The fixes already exist inside the product. Health mode handled crisis context far better than default chat, which shows the problem is a default, not a limit of the technology.
What Perplexity is: an answer engine, not a chatbot
An answer engine is a search tool that writes its results as a conversational answer instead of a list of links. Under the hood, Perplexity runs a process engineers call retrieval-augmented generation. When you type a question, it searches the web, pulls text from several pages, and feeds that text to a large language model, which writes a summary and attaches numbered citations to the pages it drew on.
That design has a real advantage: answers are tied to current pages rather than only to what a model memorized during training. It also has two weaknesses parents should understand.
First, the model decides which retrieved text to use and how to phrase it. A citation means “this page was in the material the model read,” not “this page says exactly what the sentence says.” Second, the model’s safety behavior still depends on the model and the product’s settings. Perplexity routes queries to several models, including its own Sonar model and frontier models from OpenAI, Anthropic and Google, and the Institute noted that it does not disclose which model answered a given query.
What the testers did
The Institute’s Perplexity risk assessment, published September 15, 2026, ran eight test plans between May 11 and August 11, 2026. The team registered Pro accounts in the mobile app with the age 15, then continued on the web. It added a narrower crisis test on a free account. Prompts were the same as, or similar to, those used for the Institute’s earlier assessment of Google Search’s AI features, so the two products can be compared.
The plans covered mental-health crisis responses (533 searches), developmental appropriateness (150), homework completion (80), historical facts (90), bias benchmarks (100), real-time accuracy (210), synthetic media (33) and image search (26). Crisis responses were reviewed by a panel including practicing psychiatrists and a developmental-behavioral pediatrician. Perplexity was sent a draft for factual review and, according to the report, did not respond.
The Institute says it is funded by philanthropy and industry, including makers of some products it evaluates, and keeps editorial independence. Testing was done from the San Francisco Bay Area, and because Perplexity personalizes some answers by location, results may differ elsewhere.
Is Perplexity safe for kids? What the testers found
No controls, and an age signal that does nothing
Perplexity’s terms allow users aged 13 and up with parental consent, and its app asks for a birth date. But the Institute found “nothing in the product enforces or acts on either”: no content adjustment for a known teen, no response when a user mentions their age mid-conversation, no offboarding for underage users. It called this “age assurance as paperwork, not as protection.” The 15-year-old test account could discuss pornography and engage in romantic roleplay.
Homework
Academic mode limits sources to scholarly databases and includes Learn Step-by-Step, a tutoring feature that gives hints instead of answers. When it runs, it withholds answers 80% of the time. The problem is that it activated on just 6% of homework prompts. Academic mode completed homework at roughly the same rate as default chat. As the report puts it, the product “usually leaves students to choose the guardrail themselves.”
Crisis responses
On a broad measure, Perplexity did better than Google. Across 114 crisis prompts, default chat gave some crisis line or medical referral 94% of the time, against 58% for Google’s AI Overview and 77% for AI Mode. But most of those referrals pointed to an unspecified “medical professional.” The rate of naming an actual crisis line was lower:
| Condition (default chat unless noted) | Named a crisis line | Referred to a medical professional |
|---|---|---|
| Suicide and self-harm | 50% | 84–100% |
| Eating disorders | 23% | 84–100% |
| Impaired reality (psychosis, mania) | 16% | 84–100% |
| Substance use | 90% | n/a |
| Depression, anxiety, PTSD and others | 0–33% | n/a |
| Health mode, suicide and self-harm | 88% | n/a |
Source: Common Sense Media Youth AI Safety Institute, September 2026.
Two details stand out. The only eating-disorder hotline it gave was the NEDA helpline, which has been disconnected since 2023. And the most serious problem was lost context: in default chat, after a tester disclosed a crisis, a follow-up question that sounded harmless on its own was often answered as if the disclosure had never happened. Health mode, on the same prompts, kept the context and escalated. The Institute’s point: the safer behavior already exists inside Perplexity, but it is not what a teen gets by default.
Accuracy and citations
Perplexity rejected every false premise the testers tried, a real strength. But default chat was fully accurate on only 70% of breaking-news prompts, and nearly a third of the citations audited came from user-generated sources with no editorial accountability, shown with the same visual authority as institutional ones.
That matches peer-reviewed research. In a study published in Findings of EMNLP 2023, Nelson Liu, Tianyi Zhang and Percy Liang of Stanford audited four generative search engines, Perplexity among them, and found that “a mere 51.5% of generated sentences are fully supported by citations and only 74.5% of citations support their associated sentence”. The products have changed since 2023, but the lesson holds: a footnote is a pointer, not a guarantee.
Is Perplexity better or worse than Google for kids?
That was the Institute’s central question, and its answer was “not a better or safer choice.” Google’s AI features were also rated Unacceptable Risk in July 2026, for different reasons: AI Mode completed 100% of 180 homework assignments and cannot be switched off. Perplexity noticed crises more often than Google but helped less well once it did. Our breakdown of Google AI Mode and homework covers the other half of the comparison, and our earlier piece on Perplexity versus Google for kids’ research looks at search habits more broadly.
What to do if your child uses Perplexity
Decide whether it is a homework tool in your house
Because there are no settings to change, the only lever is a family rule. Many families allow AI search for finding sources, not for writing answers. Say that out loud, and check work by asking your child to explain it.
Teach “click the footnote”
Make it a habit: before trusting a claim, open the cited page and find the sentence that supports it. If the page is a forum or a random blog, treat the claim as unverified. Kids who do this a few times learn quickly how often the summary stretches the source.
Steer toward the safer modes, honestly
Learn Step-by-Step and Health mode both behaved better in testing. You cannot force them on, but you can show your child what they do and why the default is weaker.
Make a people-first rule for hard feelings
The test found Perplexity’s default chat can lose track of a crisis within a conversation. Agree that sadness, self-harm thoughts, eating worries or anything scary go to a person first. In the US, 988 is the crisis line; elsewhere, findahelpline.com lists local ones.
What not to do: block it and assume the problem is solved
Perplexity works in any browser without an account, and similar answer engines are built into many apps. A block on one site shifts the habit to the next. The skill of checking sources travels; the block does not.
What to Watch For Over the Next 3 Months
- Week 4: Whether Perplexity responds publicly or adds a supervised mode. The Institute asked for parental controls and safer defaults for minors.
- Month 2 red flags: Homework that reads like a polished summary your child cannot explain, or citations to sources your child never opened.
- Month 3 self-check: Can your child show you, on one assignment, which cited page supports which claim?
Frequently Asked Questions
Does Perplexity have parental controls?
No. As of the Institute’s September 2026 assessment, Perplexity offered no parental controls, no supervised-account mode and no administrative console for schools. Its web version can be used without an account, so even a parent who sets up an account cannot limit what is done logged out.
What age is Perplexity for?
Perplexity’s terms allow users 13 and older with parental consent, and the app asks for a birth date. The Institute found that the product did not change its behavior for a known 15-year-old account, so the age requirement is not backed by teen-specific protections.
Are Perplexity’s citations reliable?
Sometimes. Perplexity rejected false premises well, but nearly a third of the citations the Institute audited came from user-generated sites. A 2023 Stanford study found only about half of answer-engine sentences were fully supported by their citations. Teach your child to open the source and find the sentence.
Is Perplexity safer than ChatGPT for teens?
The two were tested separately and both got the Institute’s lowest rating. ChatGPT has a teen mode and parental controls that the Institute found unreliable; Perplexity has neither. Neither is a substitute for a parent’s involvement, and neither should be a teen’s place for emotional support.
Can schools turn Perplexity off?
Schools can block websites on their networks and managed devices, but Perplexity itself offers no K-12 controls to configure. At home, on personal devices, it is available in any browser.
About the author
Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.
Sources
- Common Sense Media Youth AI Safety Institute. (2026, September 15). “Perplexity Risk Assessment.” https://institute.commonsensemedia.org/risk-assessments/perplexity
- Common Sense Media Youth AI Safety Institute. (2026, July 14). “Google Search: AI Overview & AI Mode Risk Assessment.” https://institute.commonsensemedia.org/risk-assessments/google-search
- Liu, N. F., Zhang, T., & Liang, P. (2023). “Evaluating Verifiability in Generative Search Engines.” Findings of the Association for Computational Linguistics: EMNLP 2023. https://doi.org/10.18653/v1/2023.findings-emnlp.467
- Common Sense Media Youth AI Safety Institute. (2026). “Risk Assessments” (listing and methodology). https://institute.commonsensemedia.org/risk-assessments
- Perplexity. (2026). “Terms of Service.” https://www.perplexity.ai/hub/legal/terms-of-service
- Common Sense Media. (2024, August 9). “Perplexity AI Risk Assessment” (prior rating, PDF). https://institute.commonsensemedia.org/sites/default/files/risk-assessments/csm-ai-risk-assessment-perplexity-08092024.pdf
