These 22 prompts turn Kimi — Moonshot AI's assistant, built on the Kimi K2.7 / K2 family (a 1T-parameter mixture-of-experts model, ~32B active, with a 256K-token context) — into a research assistant: running Deep Research reports, pulling cited web briefs, comparing conflicting sources, summarizing uploaded documents, fact-checking claims, and extracting structured data from files. Kimi is free to start at kimi.com, and every prompt is written out in full and ready to copy.
Two habits make Kimi trustworthy for research. First, turn on Web search for a fast sourced answer or Deep Research for a multi-source cited report, and demand a link and publication date for every claim — without search, Kimi answers from memory and can be confidently wrong. Second, upload the real document; the 256K context holds long reports and several sources at once, so Kimi can cite specific sections or extract data rather than paraphrasing. Verify anything high-stakes against the primary source, since Kimi is not a live database or finance terminal. For the full set, start at the best Kimi prompts roundup, and for automated multi-step runs see the Kimi agent prompts.
Deep Research reports
Turn on Deep Research for these — Kimi runs many searches across many sources and assembles a longer, cited report. You can also route these through Agent Mode (OK Computer) to get the result as a downloadable report artifact. Always demand cited links, dates, and a preference for recent primary sources.
1. Deep Research report on a topic
Turn on Deep Research. Produce a thorough, cited report on [TOPIC] as of [MONTH YEAR]. Structure it as: background, current state, key debates, and what's likely next. Run enough searches to cover the topic from multiple angles, prefer recent primary sources over secondary write-ups, and cite every claim with a link and its publication date. End with a short "confidence and gaps" section noting what you couldn't verify.Best for: A deep, sourced briefing on a topic you need to actually understand, not skim.
2. Cited market landscape report
Turn on Deep Research. Map the current landscape of [MARKET / INDUSTRY] as of [MONTH YEAR]. Cover the main players and their positioning, recent funding or M&A, pricing norms, and where the market is heading. Prefer recent primary sources — company filings, official announcements, and reputable analysts — and cite every fact with a link and date. Separate what's confirmed from what's still speculation.Best for: A grounded market map that reflects this quarter, not last year's marketing pages.
3. Literature review with Deep Research
Turn on Deep Research. Draft a literature review on [RESEARCH QUESTION]. Find the most important recent studies, reports, and credible analyses, group them by theme or school of thought, and trace how the thinking has evolved. Prefer primary studies over commentary, cite every source with a link and date, note methodological disagreements, and end with the open questions where the field needs more work.Best for: A first draft literature review with sources grouped by theme rather than listed one by one.
4. Deep Research answer to a hard question
Turn on Deep Research. I need a well-sourced answer to: [HARD QUESTION]. Don't answer from memory. Search widely, weigh the best evidence on each side, and give me a direct answer with the reasoning shown. Cite every claim with a link and its publication date, prefer recent primary sources, and clearly flag anything where the evidence is thin or conflicting.Best for: Getting a defensible, sourced answer to a question a single search can't settle.
Cited web briefs
Turn on Web search for these — it pulls live results for a single question and answers fast. Always add "cite every claim with a link and date" so the brief is current and checkable instead of a plausible-sounding guess.
5. Cited web brief on a topic
Turn on Web search. Give me an up-to-date brief on [TOPIC]. I want the current facts, not background from memory. Structure it as: what's true right now, what changed most recently, and what's still contested. Cite every claim with a link and its publication date, prefer recent primary sources, and flag anything you couldn't confirm from a dated source.Best for: A fast, sourced status check on a topic that keeps moving.
6. Latest developments this month
Turn on Web search. What are the most significant developments in [FIELD] from the last 30 days? List up to 8, newest first. For each: what happened, why it matters, the date, and a source link. Prefer primary sources, skip anything you can't tie to a dated credible source, and note if a story is still developing.Best for: Catching up on a fast-moving field with every item dated and linked.
7. Source roundup for a research question
Turn on Web search. Find the most important recent sources on [RESEARCH QUESTION]: studies, reports, and credible analyses from the last [TIMEFRAME]. For each, give a two-sentence summary, the key finding, the publication date, and a link. Cite every source with a link, prefer primary sources, then close with where the evidence agrees and where it conflicts.Best for: A sourced reading list plus a quick read on where the evidence lines up.
Compare sources
When sources disagree, ask Kimi to hold them side by side rather than blend them. Turn on Web search to pull the sources and Thinking mode so it reasons across them, and demand a link and date for every source it weighs. The agent prompts can automate gathering the sources first.
8. Compare sources on a contested topic
Turn on Web search and Thinking mode. Find the leading sources arguing different positions on [CONTESTED TOPIC]. Build a table: Source | Position | Strongest evidence they offer | Link and date. Then tell me where they actually disagree, which side has the better-sourced case, and what evidence would settle it. Cite every source with a link, and prefer recent primary sources over opinion pieces.Best for: Understanding a genuinely contested question without picking a side prematurely.
9. Cross-check two conflicting reports
Turn on Web search. I have two sources that seem to disagree about [TOPIC]: [SOURCE A] and [SOURCE B]. Read both, then lay out exactly where they agree, where they conflict, and why the numbers or claims differ (different dates, definitions, methods, or bias). Cite the specific passage and link for each point of conflict, and tell me which one to trust for [MY PURPOSE].Best for: Resolving two write-ups that report different facts about the same thing.
10. Consensus vs outlier view
Turn on Web search and Thinking mode. On [QUESTION], what does the weight of credible recent sources say, and which reputable sources dissent? Give me the mainstream view with citations, then the strongest outlier view with citations, and judge whether the outlier is a fringe take or an early signal worth watching. Cite every claim with a link and date, and prefer primary sources.Best for: Separating the consensus from a contrarian view that might be right early.
Summarize documents
Upload the real file with the paperclip — PDF, DOCX, or slides — and Kimi reads the whole thing. The 256K context means you can upload a long report, or several at once, and ask it to cite the exact sections it drew from.
11. Executive summary of an uploaded document
I've uploaded [DOCUMENT]. Read the full document and write a one-page executive summary for a [ROLE] who hasn't read it. Cover the main argument or purpose, the 5 most important points, any numbers that matter, and the conclusion or recommendation. Cite the section or page for each point so I can trace it back. End with three questions the document leaves unanswered.Best for: Getting the gist of a long report you don't have time to read in full.
12. Section-by-section digest
Read the uploaded [DOCUMENT] and give me a section-by-section digest. For each major section: a two-line summary, the single most important takeaway, and any figure or statistic worth remembering. Keep the original section order and headings so I can map your digest back to the source. End with one paragraph on how the sections fit together.Best for: Navigating a long document by jumping straight to the section you need.
13. Key claims and evidence extractor
From the attached [DOCUMENT], extract every substantive claim the author makes. For each, give a table row: Claim | Evidence they offer | How strong the evidence is (strong / weak / none) | Section. Be honest where a claim is asserted without support, and don't add claims that aren't in the document.Best for: Seeing at a glance which arguments in a report are backed and which are just asserted.
Fact-check & verify
Kimi can be confidently wrong from memory, so verification always means turning on Web search. Demand the original source, cited links, and dates, prefer recent primary sources, and ask it to separate what's confirmed from what's uncertain. Turn on Thinking mode for the final verdict.
14. Fact-check a specific claim
Turn on Web search and Thinking mode. 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 recent primary sources, and separate what's confirmed from what's still uncertain. If you can't verify it, say so plainly.Best for: Checking a claim against the primary evidence before you repeat it.
15. Verify statistics in a document
I've uploaded [DOCUMENT]. Pull out every statistic, figure, and quantitative claim in it. Then turn on Web search and check each one against a primary or credible source. In a table: Stat as stated | What the source actually says | Match / Off / Can't verify | Source link and date. Prefer the original primary source, don't guess, and mark anything you can't confirm as "Can't verify".Best for: Catching stale or misquoted numbers before you cite a report in your own work.
16. Confirmed vs uncertain sort
Turn on Web search. Here's a set of statements I've gathered about [TOPIC]: [PASTE STATEMENTS]. For each, tell me whether it's confirmed by a recent credible source, partly true with caveats, or unsupported. Give a link and date for every "confirmed", prefer primary sources, and put the results in two clear lists — Confirmed and Needs verification — so I know what I can rely on.Best for: Sorting a pile of half-remembered facts into what you can trust and what you can't.
Extract data from files
Kimi is multimodal, so it reads tables inside PDFs, scanned images, and spreadsheets. Upload the file and ask for structured output; when the figures need to be current or verified, turn on Web search and demand a source link for each.
17. Extract a table from a PDF
I've uploaded [PDF]. Find the table(s) about [WHAT] and reproduce them as clean Markdown tables with the exact values and headers from the source. Don't round, reorder, or invent rows. Note the page each table came from. If any value is unclear in the source, mark it [unclear] rather than guessing.Best for: Getting a usable table out of a PDF without retyping it by hand.
18. Structured data from a messy file
I've uploaded [FILE], which is messy and unstructured. Extract the [ENTITIES / RECORDS] into a structured table with these columns: [COLUMN LIST]. One row per record, values pulled verbatim from the source, and a "source location" column pointing to where each row came from. Flag rows where a field was missing rather than filling it in. Output as a Markdown table I can copy into a spreadsheet.Best for: Turning a wall of unstructured text into clean rows and columns.
19. Pull entities and figures from a report
Read the uploaded [REPORT] and extract every named entity and key figure: organizations, people, dates, monetary amounts, and percentages. Group them by type in a table with the exact value and the section it appears in. Then turn on Web search and verify the three most important figures against a primary source, citing a link and date for each.Best for: Building a quick fact sheet from a report, with the key numbers verified.
Market & competitive analysis
Competitive work needs current facts, so turn on Deep Research or Web search and demand primary sources — company sites, filings, and official announcements — with a link and date on each. For heavier multi-step gathering, see how to prompt Kimi for agentic tasks.
20. Competitive landscape scan
Turn on Deep Research. Scan the competitive landscape for [PRODUCT / COMPANY] in [MARKET]. Identify the top 6-8 competitors and build a table: Competitor | Positioning | Pricing | Key strength | Key weakness | Source link and date. Prefer primary sources like company sites and filings, cite each cell you can, and end with where [PRODUCT / COMPANY] has the clearest opening.Best for: A sourced competitor map you can drop into a strategy doc.
21. Competitor deep-dive profile
Turn on Deep Research. Build a profile of [COMPETITOR]: what they sell, their pricing, target customer, recent product moves, funding or financials, and their public strengths and weaknesses. Prefer primary sources — their site, filings, and official announcements — and cite every claim with a link and date. Separate confirmed facts from inferences, and end with how they'd likely respond to [OUR MOVE].Best for: A one-page, sourced dossier on a single competitor.
22. Market sizing with cited sources
Turn on Deep Research and Thinking mode. Estimate the market size (TAM/SAM/SOM) for [PRODUCT / MARKET] in [REGION] as of [YEAR]. Show your method and every input number, and cite each input with a link and publication date, preferring primary sources over recycled estimates. Give a range, not a single false-precision figure, and flag the assumption that most affects the result.Best for: A defensible market-size estimate with the math and sources shown.
That's the full research kit. Bookmark the best Kimi prompts roundup for the whole collection, add the agent prompts when you want Kimi to gather sources on its own, and reach for the cheat sheet when you just need the toggles and formula on one page.
Frequently Asked Questions
Which Kimi model are these research prompts built for?
They target the Kimi K2.7 / K2 family from Moonshot AI — a 1-trillion-parameter mixture-of-experts model with roughly 32B active parameters and a 256K-token context window that holds long documents and many sources in a single chat. Everything runs free to start at kimi.com, and the prompts also work with Web search, Deep Research, and file upload since those are shared features.
What is the difference between Web search and Deep Research in Kimi?
Web search pulls live results for a single question and answers quickly — ask it to cite each claim with a link. Deep Research runs many searches across many sources and produces a longer, cited report, which is better for landscape scans, literature reviews, and market analysis. Use Web search for a fast sourced answer and Deep Research when you want depth.
Does Kimi cite its sources?
With Web search or Deep Research turned on and an explicit instruction, yes. Ask it to cite every claim with a link and a publication date, prefer recent primary sources, and separate what is confirmed from what is uncertain. Without search turned on, Kimi answers from memory and can be confidently wrong, so demand links whenever the facts need to be current.
Can Kimi read my PDF or data file, and how long can it be?
Yes. Upload a PDF, DOCX, spreadsheet, or image and Kimi reads the full file, not just a summary. The 256K context window is large enough to hold long reports or several documents at once, so you can ask it to cite specific sections or extract structured data instead of pasting a shortened version.
What is Agent Mode / OK Computer in Kimi?
Agent Mode (OK Computer) lets Kimi run a multi-step task on its own and hand back a finished artifact — for research that often means a structured, cited report you can download. It is useful when you want Kimi to plan the searches, gather sources, and assemble the report rather than answering in one turn.
When should I turn on Thinking mode for analysis?
Turn on Thinking mode when the task means synthesizing many facts, weighing trade-offs, or reasoning carefully across a long document. It spends extra effort on the reasoning, which pays off on comparisons, fact-check verdicts, and multi-source synthesis. Pair it with Web search or Deep Research when the analysis also needs current, cited facts.
Is Kimi reliable for research?
It is reliable when you turn on Web search or Deep Research, demand cited links and dates, prefer recent primary sources, and upload the real document instead of summarizing it. Kimi is not a live financial terminal or database, so verify anything high-stakes against the primary source before you act on it.
How do the [BRACKET] placeholders work?
Anything in square brackets is a swap-in. Replace [TOPIC], [DOCUMENT], or [TIMEFRAME] with your own details before sending. The rest of each prompt is written to be copy-paste ready.