If you purchase something through the links in our articles, we may receive a commission. This aids in supporting our journalism. Learn more. We also encourage you to subscribe to WIRED.
Perplexity, an AI search engine that has sparked debate due to borrowing extensively from news articles and evading web-scraping regulations, has recently committed to being a dependable source for real-time information on the highly competitive US presidential election.
Perplexity asserted that its Election Information Hub would act as “an entry point for understanding key issues, voting intelligently, and tracking election results.”
“There is only one AI that can do this,” declared Perplexity’s CEO, Aravind Srinivas on X. Srinivas seemed to mock The New York Times’ publisher by sharing a post on X, offering assistance while Times Tech Guild employees were on strike during contract discussions; he further clarified that the assistance being offered was for infrastructure, not AI-generated content.
Perplexity’s tool performed admirably last night, delivering mostly reliable voting information and accurately tracking results as they unfolded. This success can be attributed to a significant reduction in its reliance on AI. Currently, Perplexity is in the final stages of securing a $500 million funding round, which would bolster its valuation to $9 billion, according to a source familiar with the matter who spoke with WIRED yesterday.
Utilizing a large language model to summarize web information can be problematic, as these models often fabricate facts when they’re uncertain. To mitigate this, Perplexity implemented stricter controls, ensuring information regarding candidates, ballot measures, and polling sites was summarized from sources provided by Democracy Works, a nonprofit created to furnish accurate voter information. Their live results were sourced through a collaboration with the Associated Press, which supplies this data via an API. Additional insights were gleaned from a meticulously curated list of trusted outlets.
Alon Yamin, cofounder and CEO of Copyleaks, a company specializing in plagiarism detection, expressed appreciation for Perplexity’s enhanced safety measures regarding the election information it disseminates, along with its clear citations. Last week, Copyleaks published an analysis indicating that Perplexity seems to summarize content from certain publishers that are behind paywalls.
However, Yamin cautions that errors can still occur, and users are encouraged to verify the sources of their information. “Nothing generated by AI is entirely authentic,” he noted. “The fundamental issues remain—hallucinations can still happen, and information might be partially accurate, among other concerns.”
Perplexity’s Election Information Hub may create some confusion between confirmed information and unfiltered AI-generated content. While certain results are sourced from reliable channels, seeking additional details may lead to open-ended AI-generated responses from the broader internet.
Other AI firms seem to be adopting a more prudent stance regarding the elections. In tests conducted by WIRED, ChatGPT Search, a newly launched feature from OpenAI, frequently refrained from delivering information about voting. “We’ve instructed ChatGPT to not express preferences, offer opinions, or make specific recommendations about political candidates or issues even when explicitly asked,” stated Mattie Zazueta, a spokesperson for OpenAI, to WIRED.
However, the outcomes were often inconsistent. For example, the tool at times declined to share persuasive points for supporting a particular candidate, but on other occasions, it readily provided some.
Similarly, Google’s search engine also steered away from generating AI-related results concerning the election. The company announced in August that it would restrict AI usage in relation to the election across search and other applications. “This new technology can make mistakes as it learns or as news breaks,” the company noted in a blog entry.
However, even regular search results can present challenges. On voting day Tuesday, certain Google users observed that a search for “Where do I vote for Harris” yielded location information, while a query for “Where do I vote for Trump” did not produce similar results. Google clarified that this discrepancy arose because the search algorithm interpreted the request as being related to Harris County in Texas.
Meanwhile, some newer AI search platforms, such as Perplexity, are pursuing a more daring strategy. You.com, another emerging player that merges language models with traditional web searches, unveiled its own election tool on Tuesday, developed in partnership with TollBit, a firm that offers AI companies controlled access to content, as well as Decision Desk HQ, which provides polling data access.
Perplexity seems to have taken particularly audacious steps towards revolutionizing web search. A June report from WIRED uncovered evidence suggesting that a bot connected to Perplexity was disregarding directives not to scrape content from WIRED.com and other sites owned by WIRED’s parent company, Condé Nast. This investigation corroborated an earlier claim made by developer Robb Knight regarding the behavior of bots managed by Perplexity.
The AI search platform is also facing allegations of excessively borrowing from news websites. For example, in June, a Forbes editor pointed out that Perplexity had extensively summarized details of an investigation published by Forbes, complete with footnote citations. In response, Forbes reportedly issued a warning letter threatening legal action against Perplexity for this practice.
In October, News Corp took legal action against Perplexity, claiming that the platform had copied material from The Wall Street Journal and the New York Post. The lawsuit contends that Perplexity is violating copyright regulations by occasionally creating sections of news articles that were not accurate and misattributing quotes to its publications.
