Skip to main content
ZICQ

DeepSeek prompts for meeting notes, requirements and research | ZICQ

ZICQ is a bilingual AI resource site with a public prompt library and Agent Skills library. This DeepSeek guide focuses on Chinese work documents: meeting notes, requirements review, debugging logs and comparing sources. The English version translates the same original examples. Replace the bracketed fields with your material and check every conclusion against the input.

By ZICQ · Updated

Browse the public prompt library →

How to use these templates

  1. Choose an example and replace every bracketed field with your task and input.
  2. Paste the prompt into DeepSeek. Supply the document or code needed for the task.
  3. Check facts, citations and output format; revise the prompt when the result misses a requirement.

Four reusable prompt examples

Meeting notes and action items

Turn the transcript below into meeting notes in [output language]. Separate decisions, open questions and action items. For each action item, include the owner and deadline only if explicitly stated; otherwise write 'not specified'. Preserve disagreements and do not turn a suggestion into a decision.

Transcript:
[transcript]

Requirements review

Review this product requirement for [project]. Make a table of requirement, ambiguity, clarification question and acceptance criterion. Identify missing assumptions and conflicting constraints. Keep proposed changes separate from the original requirements. Do not invent stakeholder decisions.

Requirement:
[document]

Debugging logs

Analyze the supplied [application/runtime] logs. Expected behavior: [expected]. Timeline: [timeline]. List directly observed errors, possible causes and the next check for each cause. Mark hypotheses as hypotheses. Do not claim to have run commands or confirmed a cause without evidence.

Logs:
[logs with secrets removed]

Compare source claims

Compare the supplied sources on [topic]. Return a table of claim, source title and URL, supporting excerpt and agreement or disagreement. Keep each source's publication date when supplied. Distinguish a contradiction from a difference in scope or date. End with questions that require more evidence.

Sources:
[titles, URLs, dates and excerpts]

Empirical testing notes & best practices

Observations and tuning tips gathered from our hands-on testing across DeepSeek (V3/R1) and ChatGPT (GPT-4o/o1):

  • Temperature & Determinism: For structured JSON extraction and code refactoring, keep temperature between 0.1 and 0.2 to avoid schema deviations. For analytical synthesis, 0.4–0.6 balances coherence with insight.
  • Hallucination Mitigation: Always include negative constraints (e.g., "If evidence is missing, state unknown rather than extrapolating"). Testing shows models without explicit unknown-fallbacks hallucinate significantly more on sparse inputs.
  • Reasoning Models (R1 / o1): When prompting reasoning models, do not over-prescribe manual step-by-step chains. Over-constraining internal reasoning degrades problem-solving accuracy compared to setting clear goals and verification constraints.

Frequently asked questions

Where can I find DeepSeek prompt templates?

ZICQ's public English prompt library is at /en/prompts. This guide includes four original examples; the library groups additional prompts by use case. Check each template's stated tool and requirements before using it.

How do I adapt a prompt to my task?

Specify the task, audience, input material, constraints and output format. Replace every bracketed placeholder, try the prompt with a small example, and revise it after reviewing the output.

Will a prompt enable web search?

A prompt alone does not enable web search or other tools. Use the search controls available in your chat product. For a source-based task without search, supply the source text and URLs yourself.

Are these official DeepSeek templates?

No. These examples were written by ZICQ. ZICQ is an independent resource site and is not affiliated with DeepSeek or OpenAI. Output quality depends on the model, available tools and your input.

Can I browse ZICQ without joining the community?

The prompt library, Agent Skills library, AI news, wiki and rankings are public. The discussion community requires an invitation.

More ZICQ resources

ChatGPT prompt guide →