Does Perplexity make up sources?
Sometimes, yes. Perplexity shows a link for each source, but published tests still catch it citing wrongly. In a 2025 Tow Center test on news articles, it answered 37% of queries incorrectly. A 2025 study of academic references put it among the chatbots with the highest hallucination rates. Open every source and read the cited passage before you use it.
Two different ways a citation goes wrong
An invented source points to a page or paper that does not exist. A real but unsupported source exists, but does not say what the answer claims. Perplexity's links make the first kind easier to spot; the second kind only shows up when you read the page.
A third pattern from the Tow Center test: the right article, but a copy. Perplexity Pro cited syndicated versions of Texas Tribune articles for three of ten queries. For a thesis, cite the original publisher, not a republished copy.
What the two studies actually tested
Neither study tested Perplexity on the kind of literature search a thesis needs. The Tow Center asked for the headline, publisher, date and URL of news excerpts. The arXiv study asked free chatbots to generate academic references from a prompt. Read the scope column before you repeat a number.
| Study | What it found | Scope limit |
|---|---|---|
| Tow Center / CJR (Jaźwińska & Chandrasekar, 6 Mar 2025): 8 AI search tools, 1,600 queriesDossierSource 1 | “Perplexity answering 37 percent of the queries incorrectly, while Grok 3 had a much higher error rate, answering 94 percent”. Premium versions answered more prompts correctly but also gave more confidently wrong answers. | News articles, not academic papers. The authors say the findings “are not intended to be extrapolated to all models or news organizations”. |
| Cabezas-Clavijo & Sidorenko-Bautista (arXiv 2505.18059, 2025): 8 free chatbots, 400 academic referencesDossierSource 1 | Abstract: “Copilot, Perplexity, and Claude exhibited the highest hallucination rates”. Across all chatbots, “39.8% were either erroneous or entirely fabricated”. | Free versions generating references from a prompt. We read the abstract only, so we give no Perplexity-specific rate. |
Check a Perplexity source in two minutes
Open the link. Find the sentence that supports the claim; if you can't, the citation fails. For a paper, search the exact title in a scholarly index and resolve the DOI. Cite the original publisher, not an aggregator or copy.
Before you paste in your own draft
Perplexity can use content from Free, Pro and Max accounts for model training: “AI Data Retention” is on by default. Turn it off under Account settings > Preferences > AI data retention. The opt-out only covers data collected after that date; previously collected training data cannot be removed.
Questions this guide answers
Every answer below is assembled from the dated fields on this page. Nothing is written separately for search.
Does Perplexity make up sources?
It can. Perplexity shows a link for each source, but tests found wrong articles, copied versions and, in an academic-reference study, high hallucination rates. Treat every citation as unchecked until you have opened it.
Is Perplexity accurate for academic research?
There is no published test of Perplexity on thesis-style literature searches that we could verify. The 37% figure is from news articles. Use it to find leads, then confirm each paper in a scholarly index.
Is Perplexity Pro more reliable?
Not clearly. In the Tow Center test, premium tools answered more prompts correctly than their free versions but also gave more confidently wrong answers instead of declining.
How do I check a Perplexity citation?
Open the link, find the sentence that supports the claim, and cite the original publisher. For papers, match the exact title in a scholarly index and resolve the DOI.
Does Perplexity train on my searches?
It can on Free, Pro and Max accounts: “AI Data Retention” is on by default. Turn it off under Account settings > Preferences > AI data retention. The opt-out only covers data collected after that date.