Perplexity Deep Research: How It Works, Limits and Best Prompts
Perplexity Deep Research turns a complex question into a multi-stage investigation. It searches widely, reads sources, analyses documents and data, then produces an editable report with citationsโusually within a few minutes.
Free accounts currently receive one Research query per month. Paid plans provide higher plan-dependent access, but only some Enterprise tiers publish fixed numbers.
What is Perplexity Deep Research?
It is Perplexity's report-building mode for questions that require broader searching, iterative reasoning, evidence comparison and a more substantial output than a normal answer.
A research workflow, not one long search
Deep Research plans its approach, conducts multiple searches, reads relevant web pages and uploaded material, identifies gaps, refines the investigation and synthesises the evidence into a structured report. Perplexity says the process can involve dozens of searches and hundreds of sources.
The result is designed for market analysis, academic exploration, product comparisons, due diligence and other knowledge work. It can save hours of initial discovery, but it does not remove the need to inspect important citations, challenge weak assumptions and verify high-stakes claims.
To begin, open Perplexity on the web, a supported mobile app or the Mac app, select Research from the mode selector beside the search box, then enter your brief and attach any relevant files.
How Perplexity Deep Research works
The exact path changes with the question, but the system follows a recognisable sequence from scope-setting to an editable final report.
Understand the task
Research interprets your objective, scope and output requirements. Broad prompts may trigger clarifying questions before work begins.
Build a research plan
It breaks the request into subtopics and decides which searches, documents, calculations or comparisons are required.
Search and read
The system performs iterative searches, reads source material and can incorporate uploaded documents directly into the investigation.
Analyse evidence
It cross-references findings, reasons about contradictions and can use an improved code sandbox for calculations and data analysis.
Show live progress
You can see sources being read and key findings appearing. Follow-up instructions may be added while the research is still running.
Write the report
The findings stream into an editable file with citations. You can refine, share or export the finished research.
You cannot manually choose the Research model
Unlike Pro Search, Deep Research automatically selects the model combination it considers suitable. Plan tier influences capability and access, but the normal Search model selector does not control a Research run.
What Advanced Deep Research can do
The 2026 update expanded the feature beyond web summarisation into a more interactive research and report-editing workflow.
Ask clarifying questions
For an unclear or overly broad brief, Research can ask about audience, geography, timeframe or deliverable before the investigation begins.
Search more of the web
It browses widely and attempts to reach sources that a quick search may miss, then changes direction when new evidence reveals a gap.
Work with uploaded files
Attach documents or data and ask Research to compare them with each other, cross-reference them against the web or extract relevant evidence.
Run calculations
An improved code sandbox supports analysis, calculations and data-led reasoning where a report needs more than textual synthesis.
Accept live follow-ups
Add a useful constraint or new question while work is underway, rather than waiting for the completed report and starting again.
Create editable reports
The report streams into a file that can be edited, refined and shared. Finished work can also be exported or turned into a Perplexity Page.
Is Perplexity Deep Research free?
Free users receive limited access. Paid plans increase the allowance, with the highest published limits reserved for Enterprise tiers.
$200 annually
$2,000 annually
$400 annually
$3,250 annually
Deep Research vs Search, Pro Search and Computer
Using the heaviest mode for every question wastes time and allowances. Match the tool to the depth and output the task requires.
Standard Search
Use it for simple facts, definitions and everyday questions. It returns a concise cited response without a long autonomous investigation.
Pro Search
Use it for a nuanced question needing broader sources and follow-ups. Paid users can normally select an eligible AI model before searching.
Deep Research
Use it for multi-part analysis, due diligence and substantial reports. It searches iteratively, chooses models automatically and supports live progress.
Perplexity Computer
Use Computer when the job requires tools, connectors, document creation or actions beyond delivering a research report. Computer consumes separate credits.
How to write better Deep Research prompts
A good prompt defines the research decision, boundaries, evidence standard and deliverable. More words do not help unless they reduce ambiguity.
Guest-posting market research
OutreachResearch the UK guest-posting and digital PR market in 2025โ2026. Identify buyer priorities, common service models, pricing signals and trust concerns. Prioritise agency websites, publisher policies, surveys and recent industry sources. Separate verified findings from inference, include a comparison table and cite every material claim.
Competitor content-gap report
SEOCompare the public content strategies of [Brand A], [Brand B] and [Brand C] for [topic] in the UK. Map recurring themes, formats, audience stages and evidence quality. Identify defensible gaps for a specialist site. Do not invent traffic or keyword-volume data; flag where a dedicated SEO platform is required.
Evidence-led content brief
ContentCreate a research brief for an expert article about [topic] aimed at [audience]. Cover search intent, questions readers need answered, current primary evidence, competing viewpoints and practical examples. Use UK sources where possible, exclude low-quality aggregators and finish with a claim-to-source table for editorial checking.
Digital PR opportunity research
PRFind five evidence-led digital PR angles connecting [brand expertise] with current UK trends from the past six months. For each angle, explain the news hook, useful public datasets, likely journalist audiences, competing coverage and the original contribution our brand would need to make.
Tool comparison and buying decision
CommercialCompare [Tool A] and [Tool B] for a UK SEO agency with [team size] and [workflow]. Verify current pricing and plan limits from official sources. Evaluate research quality, collaboration, exports, privacy and total annual cost. Show the best use case for each and finish with a conditional recommendation, not a universal winner.
Using Perplexity Deep Research for SEO and guest posting
Deep Research works best as an evidence and planning assistant. It does not replace specialist SEO databases, direct publisher checks or editorial judgement.
Content research briefs
Collect recent studies, expert viewpoints, statistics and source links before writing. Ask for a claim-to-source table so an editor can audit the evidence efficiently.
Topic and entity mapping
Explore the subtopics, terminology, stakeholders and questions surrounding an entity. Convert the findings into a draft hub-and-supporting-page architecture.
Competitor content gaps
Compare what competing sites publicly cover and how well they support their claims. Use an SEO suite separately for search volumes, rankings and backlink metrics.
Publisher prospect context
Research a site's audience, editorial themes, authors and public contribution policies before outreach. Always confirm live contact and submission information directly.
Digital PR ideation
Find timely data, unanswered questions and contrasting trends that could support an original campaign rather than simply repeating an existing news story.
Client strategy support
Cross-reference a client brief or audit with current market evidence, then produce options, risks and assumptions for a strategist to review.
How to check a Deep Research report
A cited report can still contain weak sources, misread evidence or overconfident synthesis. Review the report before publishing or acting on it.
Open the citations that support major claims
Confirm that each linked page says what the report claims, that the source is credible and that a summary has not removed an important qualification.
Prefer primary and recent evidence
Official documentation, original research, public datasets and direct statements normally deserve more weight than listicles repeating another source.
Look for missing viewpoints
Ask what geography, timeframe, user group or contrary evidence is absent. A polished synthesis can still reflect an incomplete search frame.
Recalculate important numbers
Check totals, percentages, currency conversions and comparisons against the underlying data. Small calculation errors can change a commercial recommendation.
Separate evidence from inference
Mark conclusions that Perplexity derived rather than found directly. Inference can be valuable, but readers should not mistake it for a documented fact.
Edit, export and share Research reports
The completed output is more flexible than a normal answer, but sharing should follow your organisation's confidentiality and review policies.
Edit the report
Refine the wording, add context, remove weak claims and restructure the report before it becomes an editorial or business deliverable.
Export the work
Perplexity's Research guide lists PDF and document exports, helping you move a reviewed report into an established workflow.
Share as a Page
Convert suitable research into a Perplexity Page for convenient sharing, after checking citations, confidential material and access expectations.
Related Perplexity guides
Understand the wider platform, compare subscription options and choose the right environment for your research.
Perplexity Deep Research questions answered
Concise answers to the access, model, timing and report questions users ask before starting a research run.
Is Perplexity Deep Research free?
Yes, with a strict limit. Perplexity currently lists one Research query per month for Free users. Pro, Max and Enterprise plans provide higher allowances.
How long does Perplexity Deep Research take?
Many reports finish within a few minutes, but timing varies with the scope, source availability, uploaded files, calculations and current demand. A narrowly defined brief is usually more efficient than an unrestricted topic.
Can I choose which AI model Deep Research uses?
No. Research automatically selects its model combination. This differs from Pro Search, where eligible paid users can choose a model from the Search selector.
What is the difference between Pro Search and Deep Research?
Pro Search produces a deeper cited answer within seconds and supports interactive follow-ups. Deep Research performs a longer, iterative investigation and creates a substantial editable report, making it better for multi-part analysis.
Can Deep Research analyse uploaded documents?
Yes. The Advanced Deep Research update added direct work with uploaded documents, allowing the system to combine file evidence with web research and analysis.
Can I add instructions while the research is running?
Yes. Perplexity says users can submit follow-up questions during an active Research run. Use this to narrow a region, request a missing comparison or correct an interpretation before the final report is complete.
Can Perplexity Deep Research reports be exported?
Yes. Perplexity lists PDF and document export options, and a report can be converted into a shareable Perplexity Page. Review citations and remove confidential information before sharing.
Is Deep Research accurate enough to publish directly?
No AI research report should be published without review. Open material citations, confirm dates and numbers, check primary sources, look for missing viewpoints and clearly label uncertainty or inference.
Start with a decision-ready prompt
Define the question, scope, evidence standard and output before starting. Deep Research can accelerate discovery and synthesis, but the strongest report is one that a human has challenged, checked and refined.
