These 22 prompts are built for the parts of Qwen that make research fast and checkable: the Deep Research agent, the Web Search toggle, file upload into a 1M-token context window, and Code Interpreter for quick stats. They are grouped by job: kicking off cited reports, current web briefs, comparing sources, analyzing your own documents, fact-checking, and pulling structured data out of files. Run them at chat.qwen.ai (free) or in the iOS and Android apps.
Every prompt is written out in full and ready to copy. They follow the Role + Task + Context + Constraints + Output format shape, and the theme running through all of them is the same: make Qwen show its sources so you can verify the answer instead of trusting it. Turn on Thinking when a step needs real reasoning, leave it on Auto for the rest. New to Qwen? Start with the 40 best Qwen prompts roundup; for code work see the Qwen coding prompts, and how to prompt Qwen for reasoning goes deeper on Thinking mode. Swap anything in [BRACKETS] for your own details.
Deep Research reports
Deep Research is an agent mode: switch it on and Qwen plans a research path, searches many sources on its own, resolves conflicts between them, and returns a structured, cited report — usually 3,000 to 5,000 words. Give it a clear question and scope, then turn the finished report into a webpage or a podcast.
1. Kick off a Deep Research report
Switch to Deep Research. I want a thorough, cited report on: [TOPIC / QUESTION]. Scope: [WHAT TO INCLUDE, WHAT TO IGNORE, TIME PERIOD, REGION]. I'll use it for [PURPOSE / AUDIENCE]. Search a wide range of sources, prefer primary and recent ones, and where sources disagree, lay out each side and say which is better supported. Structure the report with clear sections, cite every claim with a linked source, and end with the key open questions and what I'd need to resolve them.Why it works: Giving Deep Research the scope, purpose, and a rule for handling conflicts up front shapes the whole run, so you get a report aimed at your question instead of a generic overview.
2. Scope the research before it runs
Before you start researching [TOPIC], help me scope it. Ask me 4-6 sharp questions that would change how you research this — audience, depth, time frame, region, which sub-questions matter most, and what I already know. Wait for my answers, then propose a short research plan: the key sub-questions you'll investigate and the kinds of sources you'll prioritize. Once I approve the plan, switch to Deep Research and run it.Best for: Broad or fuzzy topics — a two-minute scoping exchange keeps the agent from spending its run on the wrong angle.
3. Literature-style landscape report
Switch to Deep Research. Produce a landscape report on the current state of [FIELD / QUESTION]. Cover: the main approaches or positions, who the key players or sources are, what's well established versus contested, recent developments, and where the field is heading. For each major claim, cite the source and note its date. Organize by theme, not by source, and finish with a short table of the leading positions and their strongest evidence.Why it works: Organizing by theme and forcing a strongest-evidence table turns a pile of sources into a map you can navigate, which is the point of a landscape review.
4. Turn the report into a webpage or podcast
Take the Deep Research report above and repackage it two ways. First, a shareable webpage version: a tight headline, a two-sentence summary, clear sections with subheads, and a sources list at the bottom with working links. Second, a script for a two-host podcast episode of about [10] minutes: a natural back-and-forth that explains the main findings and the open questions to a general listener, no jargon left unexplained. Keep every fact faithful to the report.Best for: Reusing a finished report — Qwen can generate a hosted webpage and a multi-speaker podcast from a Deep Research run, and this prompt drafts both.
Cited web briefs
When you don't need the full agent, the Web Search toggle is faster. Turn it on for anything current and demand a link and a date next to every fact, so the brief is checkable rather than a confident guess from the training cutoff.
5. Cited brief on a current topic
Turn on Web Search. Write me a brief on [TOPIC / QUESTION] as of this month. Cover the current state, the main viewpoints, and where things are heading. Cite every source with a link, prefer primary and recent sources, and put the publication date next to each fact. End with what's still uncertain or contested, and flag anything you couldn't confirm from a solid source.Why it works: Requiring a link and a date per fact, plus an explicit "couldn't confirm" line, gives you a brief you can spot-check instead of trusting wholesale.
6. What changed since a date
Turn on Web Search. Tell me what has changed with [TOPIC] since [DATE]. Focus only on genuinely new developments — announcements, data, decisions, or reversals — not background I already know. For each item give the date, a one-line summary of what happened, why it matters, and a linked source. Order them newest first, and skip rumors you can't tie to a credible source.Best for: Catching up on a topic you already understand without re-reading the basics.
7. Live market or price snapshot
Turn on Web Search. Give me a current snapshot of [MARKET / PRODUCT / METRIC]: the latest figures, the recent trend, and the main factors moving it right now. Present the numbers in a small table with the value, the date it's from, and a linked source for each. Note how fresh the data is and whether any figure is an estimate rather than a confirmed number. Keep commentary brief and factual.Why it works: Pinning each number to a date and source, and separating estimates from confirmed figures, keeps a fast snapshot honest about how current it really is.
8. Answer one question with dated sources
Turn on Web Search. Answer this specific question: [QUESTION]. Give me the direct answer first in one or two sentences, then the evidence behind it. Cite each supporting source with a link and its date, prefer primary sources over aggregators, and if credible sources disagree, say so and give both. If the honest answer is "it depends" or "not settled," say that plainly rather than picking a side.Best for: A single pointed question where you want the answer and its receipts, not a long essay.
Compare & synthesize sources
The value of research is often in the disagreement. These prompts make Qwen lay sources side by side in a table, name where they agree and conflict, and commit to the best-supported claim instead of blending everything into a vague consensus. Turn on Thinking so the synthesis is reasoned, not hand-waved.
9. Compare sources in a table
Turn on Web Search and Thinking. Find the main sources on [QUESTION] and compare them in a table with these columns: Source | Key claim | Evidence given | Date | Link. Include at least [5] sources spanning different viewpoints. After the table, tell me in a sentence each where the strongest and weakest evidence sits, and flag any source that looks outdated or low-quality. Cite everything with a link.Why it works: A shared-column table forces every source onto the same axes, so gaps and weak evidence stand out that a blended paragraph would hide.
10. Find agreements and conflicts
Turn on Web Search and Thinking. On the question of [QUESTION], map where the sources agree and where they conflict. Give me three lists: points of consensus, points of genuine disagreement, and points where the difference is really just definitions or framing. For each disagreement, name who takes which side and what evidence each relies on. Cite sources with links, and don't smooth over a real conflict to make it look settled.Best for: Contested topics where you need to know exactly what's settled and what isn't before forming a view.
11. Weigh two opposing arguments
Turn on Thinking. Here are two opposing positions on [TOPIC]: [POSITION A] and [POSITION B]. Steelman each one — state the strongest honest case for it and the best evidence it rests on. Then judge them: which assumptions does each depend on, where is each weakest, and which is better supported overall given what we know. Be clear about your reasoning and about what evidence would change the verdict. If you turn on Web Search, cite what you use.Why it works: Steelmanning both sides before judging stops the analysis from quietly favoring whichever argument was stated first or more forcefully.
12. Synthesize the best-supported claim
Turn on Web Search and Thinking. I've gathered conflicting information on [QUESTION]: [PASTE OR ATTACH THE SOURCES / NOTES]. Synthesize it into a single best answer. Weigh the sources by quality and recency, resolve the conflicts explicitly (say why one wins), and give me the conclusion the evidence most supports right now — with a confidence level. Note which sources you discounted and why, and what would move the conclusion.Best for: Landing on a defensible position after you've collected sources that don't fully agree.
Summarize & analyze uploads
Qwen's 1M-token context window means you can attach a long report, a contract, or several documents at once and it reads all of them before you ask. Use the paperclip for a PDF, DOCX, TXT, or image, and tell Qwen to ground its answers in the file only.
13. Summarize a long PDF, grounded
I've attached a document. Read it in full, then give me: a 3-sentence summary, the 5-7 most important points as bullets, any conclusions or recommendations it makes, and anything it leaves unanswered or assumes without evidence. Ground every point in the document only — if something isn't in it, say so rather than filling the gap from general knowledge. Note the page or section for each key point so I can find it.Why it works: "Ground every point in the document only," plus page references, keeps the summary faithful and lets you verify each claim against the source in seconds.
14. Answer questions from a document only
I've attached a document. I'll ask you questions about it. Answer only from what's actually in the document, quoting or citing the relevant part for each answer. If the document doesn't address something, say "not covered in the document" instead of guessing or using outside knowledge. First question: [YOUR QUESTION].Best for: Interrogating a long report or contract where a wrong, outside-sourced answer would be worse than "not covered."
15. Critique a report's argument
Turn on Thinking. I've attached a report/paper. Read it, then critique its argument. Lay out its main claim and the evidence it uses, then assess: are the conclusions supported by the data shown, what assumptions is it making, what's missing or cherry-picked, and where is the reasoning weakest. Separate substantive problems from minor ones. Be specific and quote the passages you're reacting to. Keep it fair — note what it does well too.Why it works: Thinking mode plus quoted passages turns a vague "is this any good?" into a specific, evidence-anchored critique you can act on.
16. Compare several uploaded documents
I've attached [NUMBER] documents. Read all of them, then compare them on [DIMENSIONS — e.g. their main claim, methodology, findings, and recommendations]. Build a table with one column per document and one row per dimension. After the table, tell me where they agree, where they conflict, and which is most rigorous and why. Ground everything in the documents and note the source document for each point.Best for: Reviewing multiple proposals, papers, or vendor docs at once — the 1M context holds them all in one conversation.
Fact-checking
Fact-checking is where cited answers matter most. Turn on Web Search, make Qwen trace a claim to its origin, and require an explicit rating — True, Misleading, False, or Unproven — so you get an auditable verdict, not a shrug.
17. Fact-check a claim with sources
Turn on Web Search. Fact-check this claim: "[CLAIM]". Find the original source, what the primary evidence actually says, and how credible outlets have covered it. Rate it True / Misleading / False / Unproven with a short reason. Cite every source with a link and its date, prefer primary sources over commentary, and clearly separate what's confirmed from what's still uncertain. If the claim is partly true, say which part.Why it works: A required rating plus linked primary sources turns a fuzzy "is this true?" into a verdict you can check and defend.
18. Verify a statistic to its origin
Turn on Web Search. Trace this statistic to its origin: "[STAT AND WHERE YOU SAW IT]". Find who first produced the number, the methodology and sample behind it, the date, and whether it's being quoted accurately or stripped of context. Tell me if it's current or outdated, and whether more recent or better data exists. Give the linked primary source, and flag it if the number can't be traced to one.Best for: The number that keeps getting repeated — this walks it back to the study so you know whether to trust it.
19. Rate a set of claims
Turn on Web Search. Here are several claims from [SOURCE — e.g. an article, ad, or speech]: [PASTE CLAIMS]. Check each one independently. Return a table: Claim | Rating (True / Misleading / False / Unproven) | What the evidence shows | Linked source. Prefer primary sources, note the date of each, and after the table give a one-line overall read on how reliable the source is.Why it works: Rating claims one by one in a table exposes a source that mixes solid facts with misleading ones, which a single overall verdict would blur.
Extract & structure data
Research often ends in a spreadsheet. Upload a file and Qwen will pull the tables, entities, or figures into clean structured data, and Code Interpreter can run Python on it for quick stats and charts without you leaving the chat.
20. Extract a table from a file
I've attached a file. Extract [WHAT YOU NEED — e.g. every date, name, and amount] and return it as a clean table with columns [COLUMN 1, COLUMN 2, ...]. Read the whole document, don't skip sections, and leave a cell blank rather than guessing if a value isn't stated. After the table, list anything ambiguous you had to interpret, and give me the same data as CSV so I can paste it into a spreadsheet.Best for: Pulling structured data out of a PDF, contract, or report without hand-copying — the CSV output drops straight into a sheet.
21. Pull entities into structured data
I've attached a document. Extract every [ENTITY TYPE — e.g. company, person, date, dollar figure, obligation] and organize them into structured records. For each, capture [FIELDS — e.g. name, role, context, source location in the doc]. Return it as a table and as JSON. Don't invent fields that aren't supported by the text, mark anything uncertain, and note any entity you found more than once with conflicting details.Why it works: Asking for both a table and JSON, with a rule against invented fields, gives you data you can load into a tool and still trust against the source.
22. Quick stats with Code Interpreter
Turn on Code Interpreter. I've attached a data file ([CSV / XLSX]). Load it, then: describe the columns and row count, compute summary statistics for the numeric fields, and answer these questions with actual calculations: [YOUR QUESTIONS — e.g. trend over time, top categories, correlations]. Show the code you ran, flag any data-quality issues you hit (missing values, duplicates), and make one clear chart of the most important finding.Best for: A fast, real analysis of a dataset — Code Interpreter runs the Python so the numbers are computed, not estimated.
That's the research toolkit. For everything else Qwen can do, the 40 best Qwen prompts roundup is the place to start, and the Qwen prompt cheat sheet lists every mode and modifier on one page.
Frequently Asked Questions
What is Qwen Deep Research and how long are its reports?
Deep Research is an agent mode in Qwen Chat that autonomously searches many sources, resolves conflicts between them, and returns a structured, cited report — typically 3,000 to 5,000 words. When it finishes, you can turn the report into a hosted webpage or a multi-speaker podcast in one click.
How do I turn on Web Search in Qwen?
Toggle Web Search on before you send the prompt. It gives Qwen real-time access to the web for current events, prices, and anything past its training cutoff. Ask it to cite every source with a link and note the publication date so the answer is checkable.
How large a document can Qwen read?
Qwen3.7-Max has a 1M-token context window, so it can read entire reports, contracts, or several documents at once — roughly 750,000 words. Attach a PDF, DOCX, TXT, or image with the paperclip and Qwen reads the whole thing before you ask a question.
How do I force cited, checkable answers from Qwen?
Turn on Web Search or Deep Research, then require a link and a publication date next to every claim, tell Qwen to prefer primary and recent sources, and ask it to flag anything unconfirmed. Grounding a document summary is the same idea: tell it to answer only from the attached file and to say so when something isn't stated.
Is Qwen free to use?
Yes. Qwen Chat is free at chat.qwen.ai and in the iOS and Android apps, with Deep Research, Web Search, file upload, and Code Interpreter included. The open-weight Qwen3 models are on Hugging Face and GitHub under Apache 2.0, so you can also self-host.
How do the [BRACKET] placeholders work?
Anything in square brackets is a swap-in. Replace [TOPIC], [QUESTION], or [CLAIM] with your own details before sending. The rest of each prompt is written to be copy-paste ready.