Cited search and briefs — finish able to spot-check sources yourself
Finish cited research: search → open 2 sources → brief → conflict table if needed
~25 min
Suggested tool: Kimi
Most research fails the same way: a chat model invents numbers and you never open the source. Forced order: crisp question → cited web search → structured brief → spot-check 2–3 links → conflict table → optional one-pager competitor scan. Good for: policy/subsidy wording, verifying product claims, public competitor facts. Bad for: feeding confidential deal data into models; medical/legal answers are leads only — licensed pros + primary text win. Example: “What subsidy wording applies to old-residential renovation for residents in City X over ~2 years?” Scope: that city, past 2 years, prefer government sites. Acceptable: one-sentence core fact + clickable sources; conflicts listed; still true after you open ≥2 links. Tool pick: Perplexity/Metaso for citations; Kimi for long Chinese reading. Chat models may summarize — numbers must come from links you checked.
You are a careful researcher. Question: What resident subsidy wording applies to old-home renovation in City X (≈2 years)? Scope: City in China, past 2 years, prefer government sites Extra constraints: Resident-facing rules only — skip contractor bidding detail Rules: 1) Prefer government/regulator docs and authoritative first reports; don’t treat marketing posts as sole evidence 2) Every key number/date/clause needs a clickable primary link 3) If unknown, say so — never invent URLs 4) If sources conflict, list them; don’t average Output: - Core fact (≤2 sentences) - Source list (title + URL + one-line excerpt) - Uncertainties / gaps
Outcome checklist
Self rating (1–5)
Turn these research notes (excerpts + links OK) into a brief: [notes] Question: [question]; scope: [region] Strip marketing and speculation; keep verifiable items only. Markdown table: Confirmed | TBD | Source (URL or #). Then ≤3 sentences: conclusion, bounds, what to verify next.
Sources disagree. Build a conflict table — no side-picking or averaging. Question: [question] Materials: [notes] Columns: Claim | Source URL | Date (if any) | Likely reason (definition/outdated/etc.) | Recommendation (trust / keep both / wait for official).
I’m spot-checking these links/claims. Make a human checklist: Question: [question] Materials: [notes] Five checks (date in scope [region], units, second-hand reprint, title vs body, repealed?). For each: how to pass/fail.
Public-info competitor one-pager (no hit piece). Subject: [question] Scope: [region] Notes (optional): [notes] Constraints: [constraints] One Markdown page: one-line positioning, 5 capabilities, pricing entry (link or [TBD]), recent public updates, risks/unknowns. Do not invent funding or private customer lists.
Answer using ONLY the materials below. Treat outside knowledge as unavailable. If insufficient, say so and list what’s missing. Question: [question] Materials: [notes] Output: direct answer + quoted lines from materials + remaining uncertainties.
AI search engine combining LLM with real-time web search, providing accurate answers with citations and deep research mode.
AI search engine providing structured answers and mind maps, ideal for deep research and knowledge organization.
Moonshot AI's assistant known for ultra-long context (2M characters), excelling at document analysis, long-form summarization, and deep research.
After you leave with templates,
Continue your learning path