On Tuesday, two AI startups tried convincing the world their AI chatbots were good enough to be an accurate, real-time source of information during a high-stakes Presidential election: xAI and Perplexity.
Elon Musk’s Grok failed almost instantly, offering wrong answers about races’ outcomes before the polls had even closed.
On the other hand, Perplexity offered helpful, real-time election insights and maps throughout the night, linking to reliable resources and offering historical context where appropriate.
Perplexity took a risky bet, and it paid off.
Late last week, the startup announced the launch of its election information hub, featuring real-time maps populated with voting data from Democracy Works and the Associated Press, the same information sources that power Google’s election map. This approach was different from most other AI chatbots, such as OpenAI’s ChatGPT or Google’s Gemini, which simply refused to answer questions related to the election.
It’s understandable why most AI labs sat this election out. It was the safe and responsible choice for many of them, as they’ve been plagued by embarrassing hallucinations at some point or another in the last year.
In particular, OpenAI recently released its Google competitor, ChatGPT Search. But the Sam Altman-led startup didn’t trust the feature to answer questions about this election, directing users to Vote.org instead. ChatGPT Search is still an early product, which is too unreliable for people to use in everyday scenarios, and OpenAI seems to have recognized that.
In contrast, Perplexity has been testing its Google competitor out in the real world since December 2022, and clearly felt it had enough data to give this election a shot.
Perplexity’s election night success could set it back in its ongoing fight with media companies; specifically, Dow Jones’ recent lawsuit claiming the startup competes with media companies for the same audiences. Despite the many outbound links within the AI chatbot’s answers, Perplexity’s election hub was itself a destination on election night, and it didn’t require users to venture off the app to get all their information. It certainly seems like Perplexity was competing with media companies, who were also vying for eyeballs on election night, even though Perplexity collects its information from those outlets. Perplexity CEO Aravind Srinivas even claimed “record traffic” the day before, and clearly hoped to maintain that momentum.
While Perplexity had deals with Democracy Works, the AP, and a few other media companies to power election features, the startup also indiscriminately used live election coverage from other media outlets such as CBS, CNN, and the BBC. Sure, Perplexity offered attribution, but the company hasn’t announced any revenue-sharing partnerships with these outlets and it’s unclear if any money changed hands.
First, let’s start with Perplexity’s election features that had nothing to do with generative AI: the charts.
People typically love visual election charts, clicking into them, and seeing granular data on a state-by-state level. It was a smart move for Perplexity to build these out, and ensured their AI systems were not the sole source of information in the app.
When visiting Perplexity’s election hub, users were met with a familiar looking electoral map of the United States, with some states blue for Kamala Harris and others red for Donald Trump. Obviously, Perplexity didn’t reinvent the wheel with this feature – copying the display Google and every TV network shows – but they didn’t need to. Throughout the night, this map appeared to update every minute or so, reflecting the information on the Associated Press’ website. It was a good way to follow the election.
There were bugs in Perplexity’s map periodically throughout the night. CEO Aravind Srinivas responded to users on X that were reporting flaws, such as Perplexity not reporting the percentage of votes that had been counted, and quickly addressed them.
Perplexity offered another familiar feature, a state-by-state tracker, giving real-time information about swing states.
Now for the AI part. When asked questions about the current state of the Presidential race, Perplexity answered with hedged responses that still gave mostly accurate information. These answers weren’t as insightful as a commentator on CNN, nor as entertaining as The New York Times’ election needle (which made a comeback this year). However, Perplexity only showed a few small hallucinations, and largely produced relevant facts in a timely manner. That’s more than I can say for any other AI chatbot on the market.
When trying to answer follow-up questions about Harris’s lead in so-called “Blue Wall” states, Perplexity did hallucinate slightly. It was referencing polling data, when it should have been referencing real-time votes at this point in the night. However, the general information here was in the right ballpark, and other AI chatbots just wouldn’t answer this question.
Here’s another followup question we tried: What ballots have yet to be counted in swing states? It was hard to find an answer to this question elsewhere. Granted, the only useful answers here were for Pennsylvania and North Carolina, but at least Perplexity didn’t hallucinate for the other states.
2024 was the first election to feature AI chatbots as a source of information around our democratic processes. However, this will not be the last election where that’s the case. Well-funded AI startups are fighting to deliver information to people in new, faster, more concise ways. Accuracy will be key to success here moving forward. So far, Perplexity has the early lead.
