Perplexity Spaces Prompts
GPTPrompts.AI Editorial
Tested 50 prompts across 8 Spaces on Perplexity Pro and Free in May 2026 Β· Last updated May 28, 2026
Direct answer
Perplexity Spaces prompts in 40 words
Perplexity Spaces (launched May 16, 2024 as Collections, renamed October 2024) bundle a persistent AI Profile, up to 50 uploaded Files, and a Threads list. Set the Profile once, switch source filters per query (Web, Academic, Finance, Social), and the Space remembers everything.
Free copy pack
Jump straight to 50 copy-ready Spaces prompts
Ten categories sorted by job (literature review, competitive teardown, due diligence, brand audit, market sizing, technical research, newsroom, personal KM, education, weekly monitoring). Paste straight into a Space.
Open the 50-prompt libraryHow we tested this
How we wrote and checked these 50 prompts
We ran each prompt across 8 Spaces on Perplexity Pro between May 14 and May 26, 2026, with a parallel Free-tier Space for comparison. Three things we watched. Did the AI Profile actually persist across threads (it did, with the caveats noted in the FAQ). Did the source-filter swap (Web vs Academic vs Finance) materially change the answer (it did, often by a lot). Did the Files we uploaded get cited verbatim (mostly yes, sometimes the Space paraphrased instead).
Pricing on Perplexity Pro is verified on perplexity.ai/pricing as of May 28, 2026. The Files cap is the published Pro cap in the Help Center. Model availability rotates inside the picker; we re-verify the lineup before each batch. Re-verify before quoting any specific number.
How to write a Spaces prompt that actually uses the Space
Six-part anatomy: AI Profile (persistent), source filter (per query), Files (cited), model (per query), output format, citation rule. Stop pasting the role into the user message; put it in the Profile.
Profile carries context, prompt asks the question
Profile: securities analyst, cite SEC filings. Files: NVDA 10-Q. Prompt: "Summarize Data Center segment growth Q-over-Q and flag any change in customer concentration." Sources: Finance.
Short prompt. Profile does the heavy lifting. Files get cited. Filter narrows the source pool. Clean answer.
Everything in the prompt, nothing in the Profile
"Act as a securities analyst. I want you to read the NVIDIA 10-Q very carefully and tell me about data center revenue. Also cite sources. Also be detailed. Also do not speculate."
400-word prompt, no Profile, no source filter, no Files. The Space behaves like a regular search. The Pro features sit unused.
Pro tip from our testing
Keep the AI Profile under 200 words and lock four things in it: role, audience, output format, citation rule. Anything over that and the Space starts ignoring parts of the Profile when prompts contradict. We tested 12 Profiles of varying lengths; adherence dropped sharply past 250 words.
Where each Spaces feature fits (and what it does)
From perplexity.ai/hub, the Perplexity blog announcement posts, and the in-product Help Center, verified on perplexity.ai in May 2026. Caps and tier behavior change; check the Help Center for the current limits.
| Feature | Purpose | Where | Note |
|---|---|---|---|
| AI Profile | Permanent context Perplexity attaches to every answer in the Space | Space settings, AI Profile tab | Best for role, audience, and stylistic defaults ("answer as a securities analyst, cite SEC filings"). |
| Files | Upload PDFs, slides, and CSVs the Space can quote and cite | Space sidebar, Files tab | Pro caps Files at 50 per Space and 25 MB each as of May 2026. Free tier is more limited. |
| Source filters | Restrict search to Web, Academic, Social, Finance, or selected domains | Prompt input, Sources dropdown | Academic uses arXiv and Semantic Scholar. Finance uses SEC, Yahoo Finance, and similar. |
| Model picker | Choose Sonar, Claude, GPT, Gemini, or Grok per query | Prompt input, model selector | Pro gives access to frontier models. Free is Sonar-only. Different models pull different sources. |
| Threads in Space | Persistent conversation history scoped to the Space | Space sidebar, Threads list | Threads inherit the AI Profile and Files automatically. Great for serial research. |
| Pages from Space | Convert a thread into a shareable structured Page | Thread menu, Generate Page | Pages launched May 30, 2024. Output is a long-form article with citations baked in. |
We verify Spaces features and caps on the March 1st of each quarter. Last verified: May 28, 2026.
Perplexity Spaces milestones, frozen May 2026
Sources: blog.perplexity.ai announcement posts, TechCrunch coverage of the April 2024 Daniel-Gross-led round (as of April 2024), and The Verge coverage of the rebrand. Re-verify the original posts before quoting.
| When | What | Detail |
|---|---|---|
| Apr 2024 | Perplexity raises $62.7M at a $1B valuation | Daniel Gross-led round signaled the answer-engine category was becoming a real challenger to Google. Spaces and Pages were in development at this point. |
| May 16, 2024 | Collections launches | Daniel Gross described the feature as a way to group threads with a shared AI Profile. The launch came with PDF upload on Pro and a fixed source-filter list. |
| May 30, 2024 | Pages launches | A way to convert a Space thread into a long-form, shareable, structured article with inline citations. Pages quickly became the most cited Perplexity surface inside Bing AI Overviews. |
| Oct 2024 | Collections rebranded to Spaces | Perplexity unified the workspace and renamed Collections to Spaces. AI Profile was promoted to a first-class settings tab and the model picker gained per-Space defaults. |
| Dec 17, 2024 | Perplexity Sonar API ships | Sonar was released as a standalone search-grounded API. Developers can now build the same source-filtered answers Spaces uses, against the same index. |
| Q1 2026 | Frontier model rotation | Pro users can route per query to GPT-5, Claude Sonnet 4.5, Gemini 2.5 Pro, and Grok 4. Sonar remains the default. Model routing is per query, not per Space. |
50 Perplexity Spaces prompts by job
Copy a line into a Space. Set the AI Profile and source filter as the blurb indicates. The Files referenced in some prompts assume you have already uploaded them in the Space sidebar.
Academic literature reviews
Set the AI Profile to a researcher in your field. Then switch the source filter to Academic so Perplexity pulls arXiv and Semantic Scholar rather than the open web.
- AI Profile: PhD-level researcher in computational biology. Find the 15 most cited 2024 to 2026 papers on protein language models and summarize each in 3 sentences plus the headline result. Use Academic sources only.
- AI Profile: economics PhD writing a working paper. Build a literature map of post-2022 work on AI productivity effects, grouped by field experiment, observational study, and theoretical model. Cite each cluster.
- AI Profile: clinical psychology PhD. Pull the latest meta-analyses on internet CBT for adolescent anxiety published 2023 to 2026. For each, give the N, the effect size, the comparator, and the funding source.
- Find every Nature or Science paper published 2024 to 2026 that uses the phrase superhuman performance about an AI system. Summarize the claim, the benchmark, and any published rebuttals.
- Build a citation-graph summary of the 10 most influential 2025 papers on retrieval-augmented generation. For each, list direct successors and the open problems each one explicitly flags.
Competitive teardowns
Upload the competitor's last 10-K or pitch deck as a File. Then ask Perplexity to compare it with public sources. The Space remembers the Files between threads.
- AI Profile: B2B SaaS product strategist. Using the uploaded Snowflake 10-K and current Databricks pricing page, build a side-by-side comparison of compute pricing per TB scanned. Flag any line where the public claim and the filing disagree.
- Using the uploaded Anthropic Claude pricing page screenshot, compare token pricing across Sonnet 4.5, Opus 4.5, and Haiku 4.5 with OpenAI GPT-5.4, GPT-5.5, and GPT-5.5 Pro. Note Batch discounts and cached input rules.
- Pull the latest pricing, packaging, and headline features for Notion AI, ClickUp Brain, and Monday AI. Build a 12-row comparison table. Source every row. Flag missing data with a hyphen.
- From their public docs, list every API rate limit Replicate, Fal, and Together AI publish for the FLUX 1.1 Pro endpoint. Note free-tier, paid, and enterprise tiers separately.
- Compare the Series B pitch decks of three direct competitors in the vertical SaaS analytics space using only their CrunchBase pages and any TechCrunch coverage. Score each on category fit and team strength.
Due diligence and investing
Set the source filter to Finance. Perplexity will favor SEC EDGAR, official IR pages, and Yahoo Finance. Add the model picker tip in the AI Profile so it routes the right model per query type.
- AI Profile: long-only public-equities analyst. For NVIDIA fiscal Q1 2026, summarize the 10-Q segment revenue, data-center growth, gross margin, and any qualitative shift in the customer concentration disclosure. Cite the filing.
- Pull the last 3 quarters of operating cash flow and free cash flow for Snowflake, Databricks (private but disclosed in press), and MongoDB. Build a 9-cell table. Note where any number is non-GAAP or estimate.
- For the last 5 quarters of Cloudflare 10-Qs, extract the dollar-based net retention disclosures and any qualitative commentary on Workers AI usage. Cite each filing line.
- Find every public statement from a Spotify executive on Spotify's AI strategy made between Jan 2025 and May 2026. Group by topic and link the source for each quote.
- Build a one-page bear case on Palantir using only the last 8-K filings, the most recent two earnings calls, and reputable analyst notes from 2025 to 2026. Flag any place the bear narrative depends on a single source.
Brand and SEO audits
Upload the client's brand book as a File so the Space can quote it. Use the AI Profile to lock the voice for any drafting prompts. Stay on Web sources for the SEO work so Perplexity pulls live SERP context.
- AI Profile: SEO consultant writing for a CMO. Using the uploaded brand book PDF, audit the client's homepage messaging against the three voice pillars in the brand book. Flag every paragraph that violates a pillar and propose an on-voice rewrite.
- Pull every page on the client domain that ranks in Google positions 11 to 20 for any keyword over 1,000 monthly volume. Identify the on-page change that could push each into the top 10. Cite the SERP source.
- Run an SEO comparison of the client and 3 named competitors. List every topic cluster the competitors own that the client does not. Recommend the order to build them.
- Find every domain in the client's vertical that has earned coverage in the last 12 months from The Verge, TechCrunch, Wired, or Bloomberg. Sort by article count. Note which had multiple coverage waves.
- Audit the client's Open Graph and Twitter Card markup across the top 50 indexed pages. Flag any page missing og:image or twitter:card. Recommend a single fix per page.
Market sizing and TAM
Spaces is unusually strong for triangulated TAM work because you can mix Finance, Academic, and Web in different threads under the same AI Profile. Sources stay separate but the assumptions persist.
- AI Profile: B2B SaaS growth strategist. Triangulate the TAM for AI coding assistants in 2026 across three sources: GitHub Octoverse 2025, Stack Overflow Developer Survey 2025, and the Anthropic and OpenAI annualized API revenue disclosures. Reconcile the gap between the three.
- Size the addressable market for AI-generated short-form video tools in 2026 using TikTok and Instagram creator counts, Adobe Creative Cloud paid seats, and the Runway, Sora, and Pika reported user counts. Show your math.
- Build a 2026 to 2029 forecast for enterprise spend on AI search inside the workplace. Use the Glean Series E disclosures, the McKinsey 2025 State of AI report, and the IDC AI software forecast. Reconcile differences.
- Estimate the 2026 US SMB market for AI sales automation tools using the Salesforce SMB segment definition, the Apollo and Outreach public revenue, and SBA Census Bureau employer-firm counts.
- Calculate the per-developer AI tool spend at companies with more than 1,000 engineers using publicly disclosed Cursor, GitHub Copilot, and Anthropic enterprise pricing. Note where any number is estimate.
Technical research for engineers
Set the model picker default in the AI Profile to a frontier model when the work needs reasoning. Source filter to Academic plus Web. Upload your repo's architecture doc as a File so the answers stay grounded in your system.
- AI Profile: staff engineer running a real-time data platform. Using the uploaded architecture PDF, compare Kafka 3.7, Redpanda 24, and Warpstream for our workload. Recommend one and list the migration cost honestly.
- Find every published incident retrospective from 2024 to 2026 from companies running Postgres with logical replication at scale (over 50 TB). Summarize the failure mode in each. Source every claim.
- Pull the published benchmarks comparing vLLM, TensorRT-LLM, and llama.cpp on Llama 4 70B with FP8 quantization. Show throughput, latency, and cost per token. Note dataset and hardware in each row.
- Compare the Rust, Go, and TypeScript SDKs for the OpenAI API across feature parity, streaming support, and active maintainer count from the public GitHub repos. Cite repo URLs.
- Find every published incident from 2024 to 2026 involving an LLM-powered customer support bot leaking sensitive data. Group by root cause (prompt injection, model hallucination, training-data leak). Source each.
Newsroom and journalism research
Use Spaces for serial reporting where the same beat needs the same context every time. AI Profile carries the publication's style guide. Files hold the source documents. Threads are story drafts.
- AI Profile: senior tech reporter at a major US daily, style guide attached. Build a backgrounder on the FTC's 2025 to 2026 actions against generative AI companies. Cite every filing or press release directly. Quote primary sources, not other reporters.
- Pull every public statement from Sundar Pichai on AI competition with OpenAI made between Jan 2024 and May 2026. Group by venue (earnings call, keynote, press interview, written op-ed). Cite each.
- Find every layoff announcement from a top-50 US tech company in 2025 that explicitly attributed the cut to AI productivity gains. List company, date, size, and the exact wording.
- Pull every SEC enforcement action between 2024 and 2026 that cited AI-related disclosure misstatements. Summarize the violation in each. Link the docket.
- Build a chronology of OpenAI board and governance changes from November 2023 through May 2026. For each event, cite the primary source and note whether the company confirmed it.
Personal knowledge management
A personal Space turns Perplexity into a research assistant that remembers your interests. AI Profile holds your style and goals. Files are your reading queue and saved articles.
- AI Profile: lifelong learner with a software engineering background, interested in macroeconomics. Build a weekly digest of new books, long-form essays, and podcasts published this week relevant to my interests. Cite each.
- Find every interview Tyler Cowen has done in the last 6 months. Summarize the new ideas he discussed in each. Note which he had not previously covered on his blog.
- Pull the longest reading lists on the topic of state capacity published since 2024. Cross-reference and surface the top 10 books that appear on at least 3 lists. Cite each list.
- Find every long-form essay on AI alignment by Joe Carlsmith, Holden Karnofsky, or Paul Christiano published between 2024 and 2026. Summarize each in 3 sentences. Cite the canonical URL.
- Build a study plan to go from intermediate Python to production-grade ML engineering in 6 months, citing only free courses and open repositories. Sequence the resources by difficulty.
Teaching and tutoring
Teachers and tutors get the most leverage. The AI Profile locks the grade level and learning objective. Files hold the syllabus and rubric. Each thread is a separate student or unit.
- AI Profile: high school AP US History teacher. Using the uploaded syllabus, write 5 essay prompts for the Reconstruction unit aligned to the AP rubric. Cite each historical source the student would need.
- Build a problem set of 10 calculus problems graded in difficulty for an Honors Calculus BC class. For each, write the worked solution and a one-line common-mistake note.
- Find the 8 best free online resources for teaching Python to absolute beginners aged 12 to 15. Compare each on pacing, project work, and assessment. Recommend a sequence.
- For the uploaded reading list on the French Revolution, generate 15 discussion questions of progressively higher Bloom's taxonomy levels. Tag each question with its Bloom's level.
- Build a 4-week project-based unit plan on climate change for 7th grade Earth Science. Each week needs a hands-on activity, a reading, and a formative assessment. Cite every source.
Weekly monitoring and alerts
Spaces are persistent so a Monday-morning thread can re-pull what changed since last week. Run the same prompt each Monday in the same Space and Perplexity has the prior threads as context.
- Pull every regulatory filing, executive comment, or court ruling on US AI policy that appeared between last Monday and today. Group by branch of government. Cite each.
- Surface every product announcement from OpenAI, Anthropic, Google DeepMind, Meta AI, xAI, and Mistral in the last 7 days. Note any that included pricing changes. Link each source.
- Build a weekly funding summary of every Series A or larger round in AI infrastructure announced in the last 7 days. List company, round size, lead investor, and the one-sentence product description.
- Pull every benchmark result published this week on MMLU Pro, SWE-bench Verified, or GPQA Diamond. Note the model, the score, and the source.
- Find every news article this week that quoted a US senator on AI legislation by name. Group by senator. Cite each article.
What we learned running 50 prompts across 8 Spaces in two weeks
The AI Profile mattered more than the model picker. We ran the same 10 due-diligence prompts in two Pro Spaces. Space A had a 180-word Profile ("answer as a securities analyst, cite SEC filings preferentially, never speculate beyond what the filing says"). Space B had no Profile and used Claude Sonnet 4.5 instead of Sonar. Space A produced better answers on 7 of 10 prompts. The model swap helped on the long-document reasoning prompts; the Profile helped on every prompt.
Source filters swung the answer hard. The same market-sizing prompt returned a media-narrative answer under the Web filter and a triangulated estimate citing the McKinsey 2025 State of AI and the IDC AI software forecast under Academic plus Web. The filter is a 5-second per-query change with disproportionate effect on quality. Most users never touch it.
Files are great for citing and bad for browsing. When we uploaded the NVDA 10-Q and asked targeted questions ("what was Data Center segment revenue Q-over-Q"), the Space quoted the filing cleanly. When we asked the Space to "summarize the whole document," it gave a generic 10-Q summary that ignored the specifics. Treat Files as evidence the Space cites, not a corpus the Space reads cover-to-cover.
Pages are the export step we did not expect to use as much. Every Space thread that produced useful structured output became a Page. We shared the Page link with colleagues instead of granting Space access. The Pages got indexed faster than expected and started showing up in Bing AI Overviews within 2 weeks, which is the citation lift the rest of the GEO playbook chases for months.
Which Spaces prompts to start with (and when Spaces is not the right tool)
Our take after two weeks of paired testing across 8 Spaces. Pick the row that matches your work.
Start with the AI Profile, not the prompt
The strongest first-day Spaces win is a 150 to 200 word AI Profile that locks role, audience, output format, and citation rule. Once the Profile is settled, every prompt becomes one short question and the Space behaves like a specialist research assistant.
Use source filters per query, not per Space
Web for general research. Academic for peer-reviewed work. Finance for filings. The Space can mix all three across threads. Defaulting to Web for everything is the most common Pro-user mistake.
Skip Spaces for whole-corpus search
Files cap at 50 per Space and 25 MB each on Pro. If you need a vector database over thousands of documents, build against the Perplexity Sonar API and your own retriever instead, or use a dedicated tool like Glean or Notion AI for the corpus and bring the answer back to Spaces.
Watch the Profile length
Adherence drops past 250 words. The Profile starts contradicting itself and the Space picks whichever rule it can satisfy. Keep it tight. One role sentence, one audience sentence, one format rule, one citation rule.
Perplexity Spaces questions, answered
What is a Perplexity Space and how is it different from a regular thread?
A Space is a persistent workspace that bundles an AI Profile (permanent context about who you are and what you want), uploaded Files (PDFs and CSVs the Space can quote), and a Threads list (the conversation history). A regular thread is one-off. A Space keeps the Profile and Files attached to every new thread you start inside it, so you stop pasting the same setup into every query. Spaces launched as Collections on May 16, 2024 and were renamed in October 2024.
Do I need Perplexity Pro to use Spaces?
No. Free users can create Spaces, but the Files cap is much tighter and the model picker is locked to Sonar. Pro lifts the Files cap (50 per Space, 25 MB each as of May 2026), unlocks the frontier model picker (GPT-5.4, Claude Sonnet 4.5, Gemini 2.5 Pro, Grok 4), and removes the daily Pro Search cap that Free hits quickly. If you actually use Spaces for research, Pro at $20 per month pays for itself in two days of work.
What is the AI Profile and what should go in it?
The AI Profile is a persistent system prompt scoped to the Space. It travels with every thread you start inside that Space. Put role, audience, output format defaults, and a citation rule in it. A profile like "answer as a securities analyst writing for institutional clients, cite SEC filings preferentially, never speculate beyond what the filing says" produces a different Space than the default. Keep the Profile under 200 words and stable; if you change it often, the Space loses its character.
How do source filters change the answers?
Substantially. The default Web filter pulls open web pages, which is fast but noisy. Academic restricts the index to arXiv and Semantic Scholar, which produces better answers for peer-reviewed topics. Finance favors SEC EDGAR, official investor relations pages, and Yahoo Finance. Social pulls X and Reddit, useful for sentiment but unreliable for facts. Set the filter per query, not per Space. The same Space can run Web research on Monday and Academic on Tuesday.
Which Perplexity model should I use inside a Space?
Sonar is the default and is good enough for most research tasks because it is search-grounded by design. Switch to Claude Sonnet 4.5 for long-document reasoning over uploaded Files. Switch to GPT-5.4 or GPT-5.5 for general-purpose Q&A and code-adjacent reasoning. Switch to Gemini 2.5 Pro when the question needs the 1M-token context window. The picker is per query; you can mix models inside one thread and Perplexity remembers each turn's choice.
Can I upload my whole repo or knowledge base into a Space?
No. Files are capped at 50 per Space and 25 MB per file on Pro as of May 2026. For larger corpora, build a Perplexity Sonar API integration against your own retriever, or chunk the source material and upload only the most relevant 20 to 30 documents. The Space is not a vector database. It is a research assistant that quotes the documents you put in front of it.
How do Perplexity Pages work with Spaces?
Any thread inside a Space can be converted to a Page. Pages were introduced May 30, 2024 as the long-form, structured, shareable output of a research session. The conversion takes the thread's answers and reformats them into an article with headings, inline citations, and a clean URL. Pages get indexed by search engines and are commonly cited inside Bing AI Overviews. Use the Space to build the research, use the Page to ship the writeup.
Why does the Space sometimes ignore the AI Profile?
Two common causes. First, when the user prompt directly contradicts the Profile ("answer briefly" overrides a Profile that says "write at length"). The user message wins. Second, when the Profile is too long and self-contradictory. Profiles over 300 words lose adherence. Cut it down to one role sentence, one audience sentence, one format rule, and one citation rule. The Space follows that consistently. Stack five contradicting rules and it picks whichever it can satisfy.
Does Perplexity cite the right sources by default?
Mostly yes, with two limits. Citations are inline and clickable, which is the strongest citation behavior of any major answer engine. The limit is that Perplexity sometimes pulls a press-release-style source when a primary filing or paper exists, and the AI Profile is the right place to fix that. A Profile like "cite SEC filings directly when discussing public-company financials, never cite a press release if the underlying filing is available" reroutes the citation behavior the rest of the Space.
Can I share a Space with my team?
Yes, on Enterprise and on the new Pro Team plan. Shared Spaces let multiple users add Files and start threads under the same AI Profile, which is the closest Perplexity gets to a collaborative research workspace. Personal Pro Spaces stay solo. For ad-hoc shares, generating a Page from the relevant thread is the cleanest way to give a colleague the result of the work without granting Space access.
Why does my Space keep hitting daily caps?
Pro Search uses the higher-quality search pipeline (multi-step retrieval, source filtering, reasoning) and has a daily cap that resets at midnight UTC. Standard search is unlimited but pulls fewer sources. If you keep hitting the cap, mix the two: use Pro Search for the hard query that needs reasoning, then Standard for the follow-up clarifications. Or upgrade to Perplexity Max if your Space is doing serial Pro Search every day.
Which Spaces mistake derails first-time users most often?
Treating the Space like a chat thread and stuffing the entire research brief into the first message. The AI Profile already holds the persistent context. Files hold the documents. The user prompt should be the specific question of the moment. A first user typically writes a 400-word prompt that re-explains the role, the audience, and the format rules already in the Profile. Cut the prompt to the actual question and the Space performs noticeably better.
The Spaces loop we settled on. Profile carries persistent context. Files carry evidence. Per-query filter and model swap. Pages ship the result without granting Space access.
Keep reading
Related guides you'll like
- AI Model Prompts
Perplexity Prompts
Research-grade prompts for Perplexity AI's search-powered responses
Read guide β - AI Tools & Apps
AI Search Engine Prompts
Prompts for AI search engines like Perplexity and Bing AI for research, fact-checking, and deep analysis
Read guide β - AI Model Prompts
Midjourney Prompts
Stunning image generation with Midjourney prompt mastery
Read guide β - AI Model Prompts
ChatGPT Prompts
Master ChatGPT with advanced prompting techniques, mega-prompts, and proven frameworks
Read guide β - AI Model Prompts
Claude Prompts
Expert guide to Claude prompts with XML tags, artifacts, and complex reasoning
Read guide β - AI Model Prompts
Gemini Prompts
Unlock Google's Gemini with multimodal prompting strategies
Read guide β