The most expensive mistake a student can make with AI isn't a clumsy sentence. It's a reference to a paper that doesn't exist.
Chat models write from patterns. Ask one for "five sources on this topic" without giving it any, and it can produce references that look exactly right: plausible authors, a believable journal, a year, even a page range. Some will be real. Some won't. They all look the same.
This post compares the two tools students most often use to find sources, Perplexity and ChatGPT, and then sets out a way to check any reference, whichever tool it came from.
How they work differently
Perplexity searches first and writes second. Every answer arrives with numbered citations, and each number links to the page it came from. On LazySusan it comes as Sonar (fast search), Sonar Pro (deeper search, more sources), Sonar Reasoning Pro (search plus step-by-step reasoning, for comparisons) and Sonar Deep Research (many searches and a long, cited report).
ChatGPT answers from what it learned in training unless you turn on Search. With Search on, it grounds the answer in current web results. ChatGPT on LazySusan also has a Research mode that produces a slower, cited report built from many sources.
So the real comparison is between Perplexity and ChatGPT with Search or Research on. Plain ChatGPT, without either, is the wrong tool for finding sources.
The short version
| Job | Better pick | Why |
|---|---|---|
| Finding sources on a topic | Perplexity | Search is the whole point; every claim has a numbered link |
| Checking whether a specific claim is supported | Perplexity | Shows the pages behind the answer |
| A long, cited overview of a topic | Either: Sonar Deep Research or ChatGPT Research | Both run many searches; both are slower |
| Explaining what a source says | ChatGPT (or Claude) | Better at explaining and writing once you have the source |
| Formatting a reference you've verified | ChatGPT | Good at structured output; check it against your style guide |
Perplexity: start here when you need sources
Perplexity is the right first stop whenever the honest answer to "how do you know?" needs to be a link. Its sources also double as a reading list, which makes it a quick way into an unfamiliar topic.
What does recent research say about the effect of sleep deprivation on working memory in university students? Cite each claim, and prefer peer-reviewed studies and review articles over news coverage.
Is it true that [claim from a lecture or reading]? Show sources on both sides.
Two tips. Ask the question you actually have, with the context that changes the answer: field, country, time period. And open two or three of the citations before you rely on the answer. Perplexity is good at finding sources, but the source is what you're trusting.
Where it falls short: a citation shows where an answer came from, not that the answer is right. A link can point to a weak source, or to a page that says something a little different from the summary. The links can also be to web pages rather than the original paper, so you may need to follow them through.
ChatGPT: better at explaining than finding
Once you have a real source in hand, ChatGPT is good at helping you understand it, summarise it and fit it into your argument. GPT-6.1 Sol writes clear, organised explanations and handles formatting tasks well.
Here is the abstract and methods section of a paper I'm citing. Explain the method in plain language, and tell me what the paper can and can't claim based on it.
Where it falls short: asking for references without Search on. Don't do it. Even with Search on, treat each reference as a lead to check, not a fact.
The five-minute citation check
Run this on every reference, whichever tool gave it to you:
- Find the paper itself. Search the exact title in your library's search or in Google Scholar. If you can't find it, don't use it.
- Check the DOI. If the reference has a DOI, paste it into doi.org, which resolves a DOI to the work it identifies. If it doesn't resolve, or resolves to a different paper, the reference is wrong.
- Check the details. Crossref Metadata Search lets you look up journal articles, books and other scholarly works. Confirm the authors, year and journal match.
- Read the part you're citing. Make sure the paper says what the AI says it says. Summaries drift.
- Cite what you read. If you only read a review article that mentions a study, cite the review, or go and read the study.
It takes a few minutes per source. That's much less time than explaining a fake reference.
A workflow that uses both
- Perplexity (Sonar Pro) to find candidate sources on your question.
- You open each one and run the check above.
- ChatGPT or Claude to help you understand the ones you keep. Paste in the actual text rather than asking from memory.
- You write, citing only what you read.
On LazySusan Perplexity, ChatGPT and Claude sit in the same workspace, so you can move from finding to understanding without copying between apps. Multi-Chat also lets you put the same question to Perplexity and ChatGPT side by side, which makes it obvious when one is citing and the other is guessing. Plans start at $8 a month, with a Student plan for .edu emails.
For long PDFs once you have them, see Gemini vs ChatGPT for research students. More on Perplexity itself: Perplexity on LazySusan.
The verdict
Use Perplexity to find sources and ChatGPT to understand them. Never use a chat model's memory as a bibliography. And whichever tool you use, a reference isn't real to you until you've opened it.
Use it honestly
- Check your course's AI policy. Some courses allow AI for finding sources but not for writing; some ask you to note which tools you used.
- AI is not a source. Cite the paper, not the chatbot, unless your style guide tells you to cite the tool itself for something it generated.
- Verify every reference. A made-up citation can be treated as an academic integrity issue even if you didn't mean it.
- Don't submit AI-written work as your own where it isn't allowed.