Stefani Reynolds/Bloomberg via Getty Images
I've been working with artificial intelligence (AI) since the 1970s, when Lisp was state-of-the-art. Lately, like everyone else, I've been looking at AI a lot more closely. While it's nifty and getting genuinely useful, it can still blunder, like the time Google's AI Overviews recommended users to eat "at least one small rock per day." AI can also lose its mind, such as when Grok went totally MechnaHitler. Even when I tell J. Random Chatbot to summarize an Otter.ai transcription of a meeting, it gets things wrong.
Also: 60% of managers use AI to make decisions now, including who to promote and fire - does yours?
Therefore, I treat the results of my AI queries with great caution. However, I found an AI chatbot whose work I can easily double-check to ensure it's in the ballpark of being accurate: Perplexity.
That may not sound like much, but it's a critical feature. Because AI chatbots are trained on information scraped from the web, they are all capable of hallucinating or spouting errant nonsense. For instance, a recently published AI-generated list of the 15 best novels to read this summer included 10 novels that didn't exist. Whoops.
Sure, Perplexity also hallucinates, but it values data from trustworthy sources. Here's what it can do, and why it's my favorite AI chatbot.
1. Reliable for research I'm very picky about my research programs. I have reason. My first business, which I founded in 1978, was named Researchers at Large. In those days, I used the online databases of the day -- DIALOG, NASA RECON, and OCLC -- to do paid academic research. My research skills remain a major reason why I'm still a successful journalist. Today, Perplexity is good enough that it has become the first AI chatbot I trust as a genuine research tool. Also: The best AI chatbots: ChatGPT isn't the only one worth trying For example, chances are you've never heard of him, but I did some of my doctoral work on Dr. Norman Leys, an early 20th-century East African anti-imperialist. Perplexity provided a decent summary of his career, and, importantly, pointed me to the academic studies that cover his life in more detail. Other AI chatbots provided me with much broader and sometimes inaccurate information, but they didn't disclose their sources. Because I studied his life, I could tell the difference. Most people would never realize they'd been led astray. By the way, before you commit to using any AI chatbot, I recommend that you do the same. Ask it detailed questions about a topic you know very well, whether it's Pittsburgh Steelers football in the '70s, Star Wars movies, or clog dancing. If it provides accurate answers to obscure questions and clearly sources the information, you can be more certain that it's worth using. I found that Perplexity does this with all the topics it covers. Whether you're doing academic, investment, or sports research, it can handle complex queries and provide accurate, detailed, and context-aware answers. Show more
2. Source and citation quality Another element that makes Perplexity great for research is its sources. Most notably, it gathers its data from academic studies and other reliable sources. It doesn't consider Joe Random's comments to be just as valid as Dr. Josephine Expert's refined answer. The free Focus feature also lets you fine-tune your searches and avoid some sources. When I was looking up Norman Leys, for example, I told Perplexity to use only academic sources, so I could avoid Wikipedia and other less reliable sites. Or, if you need helpful content on how to fix your washing machine, you can set it to focus only on YouTube videos. Also: The best AI search engines of 2025 Since I first looked at Perplexity, other major chatbots such as ChatGPT and Gemini have significantly improved their citation and source transparency features. While they're closing the gap, Perplexity's attribution is still better at telling you precisely where it got its answers. For example, while it might still pull part of an answer from Reddit, it will tell you where to find the specific post it's quoting from. You can then look for yourself to see if you think the information is trustworthy. Let me add, though, that the accuracy of all chatbots is on the decline. I find that their answers now are less trustworthy than when I first checked in with Perplexity in May 2024. That's because AI's large language models (LLMs) are now "learning" from AI-written results. The result is garbage-in/garbage out, or -- as it's more formally known in AI circles -- we're already going into Model Collapse. The moral of the story? When Perplexity gives you an answer, dig deep into those citations. You may find, as I do sometimes now, that the citations themselves are based on AI slop. Finally, Perplexity has one good feature that, as far as I know, no other AI has. It tells me when it can't find an answer. The others all too often make up answers from whole cloth. While prone to hallucinations like all the LLMs, Perplexity appears to try to stick to the facts. Show more
3. Depth and organization Here's another subject close to my heart that most people don't know that well: I feel Shoeless Joe Jackson should be in the Baseball Hall of Fame. Unless you're a hard-core baseball fan, you'll most likely know of him from the movie Field of Dreams. While Perplexity doesn't mention the movie or the book it was based on, W.P. Kinsella's Shoeless Joe, the chatbot did a good job summarizing the arguments, pro and con, for letting him back into baseball. Now that Major League Baseball has reinstated Shoeless Joe, along with a guy you probably have heard of, Pete Rose, Joe should finally make it. Also: I found 5 AI content detectors that can correctly identify AI text 100% of the time Where Perplexity stands out from the rest is that it also provided additional questions to continue my search, including: What is the history of Shoeless Joe Jackson's ban from baseball? What other players have been banned from baseball? And "What is the Black Sox Scandal?" Whatever topic you're interested in, Perplexity will help you dig deeper into the subject. As you examine an issue, the Collections feature helps you organize your findings and thoughts in a structured manner. You can keep your research notes to yourself or share them with up to five co-workers or friends. Show more
Free users default to Sonar Large with limited access to advanced models for a few queries per day. Perplexity AI Pro ($20/month) users can manually select their models or use "Best Mode" to automatically choose the model that Perplexity thinks is best for their query. Pro users can also use "Pro search" to dig deeper with its internet search, with up to three times the number of sources for the usual query.
... continue reading