Perplexity
An intriguing twist on search, Perplexity's "answer engine" falls short on accuracy and protections for kids & teens.
The rating
Our assessment of how this product aligns with each of The Institute’s eight AI Principles. Full detail in the Evaluation section below.
AI Principles
Perplexity is seeking to redefine online search with their conversational "answer engine." The company wants you to think of Perplexity as a search tool that provides answers, not just search results.
Perplexity's conversational interface adapts to your inputs and can host a continuous conversation, called a Thread, allowing you to expand on the conversation with additional searches while retaining context. When starting a new thread, you can opt to use the Focus feature to narrow Perplexity's search to types of sources. Threads can be organized into shareable Collections and converted into Perplexity Pages.
What Perplexity gets right
The combination of real-time web crawling, AI summary in responses, and ongoing "conversation" on a particular topic can create engagement and curiosity.
Perplexity's AI responses include both AI-generated summaries and numbered footnotes that link to original sources, which can enhance transparency and credibility for the answers generated by the system.
When we prompted Perplexity with questions that were clearly based on harmful stereotypes, responses would often begin with statements like, "While this is a common stereotype, its validity is debated." In many cases, Perplexity's responses would then include "points to consider" about the stereotype. (That said, many of these responses did not include cited sources.) See some examples in our AI Principles assessment below for Prioritize Fairness.
Where it falls short
Because Perplexity's responses sound correct and they are often summarizing source material, it can be far too easy to assume that source material is trustworthy and accurate. Our testing repeatedly found this was not the case. In addition, the cited sources are not always appropriate for the intended search, and Perplexity's generated answers can contain information that is not reflective of the cited sources. It is up to the user to recognize this risk and check all of the responses accordingly. This is especially worrisome for impressionable young minds. See some examples in our AI Principles assessment below for Be Trustworthy.
Perplexity does not display any warnings to users about the risks of generative AI, and unfortunately the company reinforces this by falsely stating that the system offers "Answers that are accurate and always cited." As with all generative AI chatbots, however, Perplexity can generate inaccurate information, use questionable sources, and reinforce stereotypes. The company does acknowledge inaccuracy risks in other documentation, but the system does not make this readily apparent to users.
When we tested Perplexity with prompts that included obvious stereotypes and clear misinformation, the system was generally clear and unequivocal in disputing them. On the other hand, for a range of prompts designed to assess both factual accuracy and how strongly responses reflect social biases, the chatbot's responses reinforced harmful stereotypes and generated false and irrelevant information, often in more subtle ways that are both difficult for Perplexity to combat and dangerous for impressionable minds. See some examples in our AI Principles assessments below for Prioritize Fairness and Be Trustworthy.
While many generative AI chatbots allow users over age 13, Perplexity more easily engages across a range of topics that might not be suitable for younger users and that could be problematic in any classroom. For example, Perplexity provided detailed responses and links to source material for a series of questions about sexual positions, fetishes, and how-to's for creating adult content, as well as in-depth answers about torture and war crimes.
Perplexity shares extremely minimal information about its AI systems, and we could find no information about any guardrails, content moderation, harms analysis, or safety testing. As such, given the limited information about how Perplexity handles sensitive data, the absence of safeguards noted above, and the complete lack of age gates, we urge more caution when using Perplexity than other generative AI chatbots with similar capabilities.
See more about this in our AI Principles assessment below for Keep Kids & Teens Safe.
More risk assessments
