Perplexity launches Deep Research

Background

By early 2025 Perplexity's answer engine could answer a single question with cited sources. Many questions, however, need more than one search, such as comparing options or summarising a field. Deep Research was built for that kind of task, extending the product from a single answer to a longer piece of research.

What happened

On 14 February 2025 Perplexity launched Deep Research^[1]. The feature runs dozens of searches, reads hundreds of sources and reasons over them to write a full report^[1]. The company said most tasks finished in under 3 minutes^[1].

Key details

The report is the main output. Users can export it to PDF or turn it into a Perplexity Page, a shareable page built from the report^[1].

Perplexity reported two results for the feature. It scored 21.1% on Humanity's Last Exam, a difficult test of expert knowledge, and 93.9% on SimpleQA, a test of short factual questions^[1]. These figures are the company's own.

Access was broad at launch. Deep Research was free for all users on the web, with daily limits. Pro subscribers received many more queries^[1].

Impact

Deep Research showed that a question-answering product could take on longer research jobs, not only short queries.

For students and readers, the feature is a clear example of how AI tools can be used to gather and organise information. It also shows why the source trail matters, because the value of the report depends on whether readers can check its claims.

Readers comparing AI research tools can use the same test: whether a report points to its sources clearly enough for the reader to check each claim.

Limits

The time to finish a task is described as the company's general claim, not a guarantee for every query. Readers should check the report's sources before relying on any AI-written summary.