What if AI chatbots like ChatGPT and Claude could become search engines? Perplexity AI answered that question by developing a new AI interface that could search the internet and respond to user inputs.
Perplexity was built by former employees from other major AI companies, including Meta and OpenAI. But what exactly is Perplexity? How does it work? Is it safe to use?
In this guide, we will answer all these questions with a deep dive into Perplexity AI.
What is Perplexity AI?
Perplexity AI is an artificial intelligence chatbot and search engine. It uses natural language processing and machine learning to provide complete and accurate responses to user inputs.
As a search tool utilizing large language models, the Perplexity AI chatbot can understand nuance and instructions for various tasks. Live access to the internet can provide up-to-date and highly relevant information. If the topic of your question is on the web, Perplexity can say something about it.
Using a simple and intuitive interface, Perplexity AI users can:
Find new information or ideas on any topic
Summarize articles or other text content
Get assistance for research and writing tasks
Find real-time links to other online sources
See the citations for all the information provided
Live Research & Citation
The last point about finding the verification for any information Perplexity is one of the most substantial features. Most AI chatbots have a lot of information that they can provide. However, they are limited in terms of the level of accuracy and completeness of that information.
For example, if you use popular chatbots like ChatGPT or Gemini, you have the disclaimer that the information may be incorrect. You are responsible for the factual accuracy of any work you submit—regardless of whether an AI model provides it.
Usually, all the other AI chatbots are great at providing new ideas that send your work in new directions. However, with complex or niche topics, it can be hard to find the source of information, even if it’s completely accurate.
Perplexity AI saves you a lot of time by providing:
A concise summary of the information you’re looking for in the middle of the screen
A list of the sources it’s used at the top of the screen
A list of related questions at the bottom of the screen
Annotations for significant details telling you exactly where that information came from
Image credit: Perplexity AI
This thorough but fast approach means Perplexity AI often offers outstanding support for research. Even if you don’t like its answer or have difficulty steering it, it will point out new opportunities for you.
High-Speed Research
If you use Perplexity, you’ll notice how fast it can be. With a decent device and a high-speed internet connection, it takes about a second to get your result. The speed is similar to that of clicking the search button after typing a query on Google. But instead of just seeing many results to dig through, you get an instant, concise summary with citations and additional reading material.
Quality Citations
Perplexity AI boasts of providing nearly instantaneous answers, including citations for high-quality academic papers.
Using the chatbot tool, you can quickly see that there’s a lot of truth to this. But it’s a bit of a mixed bag.
Chatting with Perplexity about niche history topics, for example, gave us citations to:
General history website articles
Wikipedia
Academic papers
Government pages
With a closer look, the sources of information ranged from low credibility to very high credibility. However, we could see exactly where each significant point in Perplexity’s summary came from.
Different types of sources are used for different topics. Switching from history to much more recent events, Perplexity started citing mostly reputable news organizations.
Of course, this is a lot better than most alternatives. Normally (even with ChatGPT’s free plan), you simply get a summary and must check each point entirely on your own. Perplexity may not always be fully accurate, but it makes it far easier to steer your research and fact-check the summary.
Other Content Types
Perplexity AI also offers an image generator, a video generator, and an image search. These can help with research as well. However, the image search suggestions may contain copyright-protected images. It’s up to you to use these additional features responsibly.
How Does Perplexity AI Work?
Perplexity was launched in 2022 by a team including former Google researchers and OpenAI employees. It was developed to compete with Google’s search engine, offering distinct features.
Perplexity employs several AI models. The default model is free to use and is streamlined for speed. It was meant for light use of all kinds, enabling quick searches.
Perplexity Pro
Perplexity Pro is a paid subscription plan that includes a much more comprehensive service. Pro subscribers get access to more searches using some of the most advanced AI language models:
GPT-4 Omni
GPT-4 Turbo
Claude 3 Sonnet
Claude 3 Opus
Sonar Large 32k
Image credit: Perplexity AI
The list of Perplexity Pro AI models is always expanding. With a Pro account, users can choose from any of them.
The strength of Perplexity AI largely comes from the versatility of the search function alongside your choice(s) of AI model. You can choose whichever is suitable for you after testing each model on various tasks. You can combine the knowledge and performance of GPT-4 models with the research value and web connection of Perplexity.
There is also a Perplexity Enterprise Pro plan that offers a more scalable level of service.
Conversation
While Perplexity’s special features stand out far more when it comes to research, their models also offer “traditional” chatbot uses. You can engage in interesting conversations with Perplexity as you could with ChatGPT or other chatbots.
The only major difference in conversations with Perplexity is that their citations provide a window into their answer. For example, the differences between ChatGPT on its own site and ChatGPT on Perplexity would be:
The live access to the internet with which ChatGPT’s AI would scan for an interesting answer
Giving you a longer list of precise sources of the text it returns in conversations
Memory and Context
Perplexity was built to have complete intellectual conversations leading to meaningful user results. It has a large memory of previous communications, enabling conversations to flow continuously.
You can always ask Perplexity follow-up questions. One great use case is that of a research assistant. You can read the citations it gives you, ask new questions about what you’ve seen there, and find more quality sources to work with.
For example, try using Perplexity to explore controversial topics in health. It will provide you with answers that are supported by a range of sources. Some of them will be unreliable, while others will be highly authoritative. You can challenge Perplexity on the information it provides and explore the new academic information it provides. This way, you could explore the intricacies of health issues with the help of authoritative research from different sources.
Is Perplexity AI Safe?
For the most part, yes.
Like any AI tool, Perplexity can be misused. However, the AI models that power it come with strong guardrails against illegal or inflammatory content. It’s generally considered safe enough for children to use it as well.
Perplexity AI commonly causes trouble only if users copy and paste what it generates and use it thoughtlessly. For example, submitting an essay entirely copied from a chatbot can get a student in a lot of trouble. Keep in mind that perplexity can lead to mistakes, so act accordingly.
User Data
Perplexity’s policies regarding user data are fairly typical. They will collect your data, including user credentials and inputs and outputs. They can also collect data from outside sources like Google Analytics. This refers to any third-party service that you attach to your Perplexity account.
According to Perplexity, this data is used to provide and improve services or comply with laws. However, they only share your data with third parties under specific circumstances, including for ads displayed in their apps. Their agreements with Anthropic and OpenAI prohibit training those models with Perplexity user data.
When it comes to the security of your own account, Perplexity uses strong and up-to-date cybersecurity measures. Internal authentication and access controls and user data are aggregated and anonymized.
Is Perplexity AI Factually Accurate?
Compared to other AI chatbots, the question of accuracy is harder to answer for two reasons:
Perplexity enables you to use a wider selection of AI models including those of other AI companies.
Perplexity AI cites its sources, leaving you to make the final analysis.
When you switch the AI model you’re using in the Perplexity AI chat function, you are changing a lot. One of the most significant differences your choice makes is the overall accuracy you can expect.
To ensure you get the most accurate possible answers from Perplexity AI, choose an AI model that performs well in the subject matter about which you’re asking for answers.
You can find information on how well each AI model performs by referring to standard benchmarks. For example, you can quickly find these performance benchmarks:
MMLU: General knowledge of 57 subjects including STEM, humanities, and social science.
GSM8K: Arithmetic reasoning.
MATH: Math word problem-solving.
GPQA: 448 multiple-choice questions in the sciences.
Image credit: paperswithcode.com
There are also a few other standardized benchmarks that test AI models against each other and against human subject matter experts. Be sure to look at a few that are relevant to your use of Perplexity if factual accuracy is important to you. Two of the top performers are included in a pro plan.
While these steps can help you get a better starting point, remember that it’s still your responsibility to fact-check what any AI model produces for you.
Why You Need to Fact-Check Perplexity AI Content
Regardless of which AI model you’re using when you use Perplexity AI, you need to fact-check the information. One great benefit of Perplexity is that it normally gives you a lot of help with this by providing academic research.
That being said, ChatGPT and Claude both provide disclaimers. Any information you use that was generated with their AI models is not their responsibility. If the content they produce is false or offensive, it’s your responsibility if you use it and suffer the consequences.
The good news is that the GPT and Claude models that Perplexity Pro offers score high on AI testing benchmarks. They are, in fact, accurate most of the time. But the potential to make mistakes is still high. The longer the content you produce with them, the more errors the text will contain.
How to Use Perplexity AI
There are plenty of fully acceptable use cases for Perplexity AI:
Creative content creation
Research assistance
Writing assistance
Summarization
Text analysis
The key to using Perplexity successfully is following its terms and your local laws. Beyond that, don’t accept the text it generates as 100% accurate. Perplexity enables you to see the source of most of the impactful statements they make. It’s suggested that you use that ability to scrutinize the source of information. If it’s a great source, you may even end up citing it in your own work, and that’s completely fine.
However, generating academic work, reports for your job, or other serious publications can get you into trouble. If you simply copy and paste Perplexity AI content, you may accidentally be guilty of anti-plagiarism or anti-AI rules. This can get you in serious trouble. Students have been expelled for passing off AI-generated work as their own.
AI chatbots are great assistants that can make life easier if you use them properly. Perplexity stands out in research assistance roles. But the consequences of abusing them can be severe. It’s also getting easier to differentiate AI-produced content from human writing.
How Do You Know If Something Was Written With AI?
AI-generated content is now something that many people are familiar with. The tell-tale signs of ChatGPT, in particular, are now common sights in academia. Studies have demonstrated that professionals like professors and HR are good at catching AI uses.
Catching signs of AI use in resumes, essays, and other documents is now a skill that professionals are trained on. In addition to using their own eyes and intuition, new AI detection tools are making it easier to single out text for further investigation.
AI detectors can scan large bodies of text in just a few moments. They can then:
Report on the probability that the content was AI-generated
Highlight specific sections where AI use is more likely
Report on how likely it is that specific sections were AI-generated
Using a tool like AI Detector makes it even easier for professionals to scan through documents quickly. These tools also detect plagiarism, a common byproduct of copy-pasting from AI chatbots.
If you’re curious whether an AI chatbot generated any text you’re reading, all you need to do is copy-paste it into our AI detector. In just a few moments, you’ll get a report on the likelihood that AI produced it.
A Last Word on Perplexity AI
Perplexity AI offers a unique value proposition. While the default AI model isn’t the best, all models can comb the internet instantly with an AI brush. With a Pro account, you can leverage the intelligence of GPT-4 and Claude 3 as well.
Using Perplexity as a writing and research assistant can be enormously beneficial. However, you need to keep in mind that Perplexity doesn’t guarantee the accuracy of the information it generates. So, be sure to use it responsibly and not accidentally submit false or plagiarized information.
