Calvin Wankhede / Android Authority
summary
- The study found that Perplexity frequently cites AI-generated content as a source of answers.
- Perplexity is an AI search engine that generates answers after searching for information online, similar to Google Search.
- Another recent study claims that Perplexity is breaking the limitations of web scraping to improve its chatbots.
AI-powered search engine Perplexity is facing scrutiny and criticism after a study found that the tool frequently cites AI-generated blog posts as sources for answers. The report, compiled by AI detection service GPTZero, claims that on average, users encounter “indirect hallucinations” — information that is backed by AI-generated sources — after just three prompts.
Our investigation found that while Perplexity cited sources, it often did not consider the veracity of the content; that is, the chatbot would repeat articles written by ChatGPT or AI hallucinations as fact. Even worse, it presented content from social media websites as fact. In one example, a prompt for “Cultural Festival in Kyoto, Japan” was given a plausible answer. However, Perplexity cited a single AI-generated article from LinkedIn as the only source for this prompt.
GPTZero claims to be able to identify AI-generated content with 97% accuracy. Forbes The company also ran the problematic content through a different AI detection algorithm, which came to the same conclusion. In a statement provided to the magazine, Perplexity's chief business officer Dmitri Shevelenko acknowledged that the search tool assigns trust scores to different domains, which is believed to be similar to how Google's PageRank works, meaning the algorithm down-ranks and weeds out low-quality websites.
Shevelenko also acknowledged that Perplexity uses an internal AI detection algorithm to flag problematic content, but he did not explain why the tool relies so heavily on AI-generated content and maintained that the study does not represent a “comprehensive assessment” of Perplexity's sources.
Founded by Aravind Srinivas, a former AI researcher at OpenAI, Perplexity is currently valued at over $2 billion and has received significant investment from Nvidia and Amazon founder Jeff Bezos.
The aforementioned study is not the first controversy to hit AI startups in recent weeks. Just last week, Wired Perplexity claims it scraped the magazine's website despite attempts by the site's engineers to block it, and after testing the chatbot with multiple prompts: Wired They also found that the chatbot sometimes made inferences based on article URLs rather than summarizing them directly — a conclusion that Perplexity's founder and CEO said reflects “a deep, fundamental misunderstanding of Perplexity and how the Internet works.”
