萃靈 Essevin
The most intimidating part of a first AI session is often the empty input box. Many people eventually type, “Write me an introduction,” receive a bland paragraph, and decide that AI is not for them.
The problem is usually not the person, and not necessarily the model. A complete task was compressed into a few context-free words. The model has to guess the audience, length, tone, and which details must stay untouched.
You do not need a catalogue of prompt jargon to get started. Keep ten opening questions nearby instead. Their job is not to make prompts longer; it is to make the decisions that matter visible.
Ask for clarity before asking for output
A useful opener normally states three things: the result you want, what is already known, and what to do when information is missing. Think of ordering at a cha chaan teng: you do not memorise the menu; you specify the main dish, portion and anything you cannot eat.
Copy the lines below and replace the angle brackets with your situation.
Ten copy-ready openers
1. “Explain it in plain language”
Explain <concept> in plain English for a secondary-school reader. Start with one-sentence conclusion, give one everyday analogy, then list three common misunderstandings. Do not assume prior knowledge.Use this for an insurance clause, a medical report or a technical term. “Start with the conclusion” keeps the answer from disappearing into definitions.
2. “Turn messy material into a table”
Turn the material below into a table with columns: Item, Owner, Deadline, Source sentence. Use only information stated explicitly. Write “TBD” when a field is missing; do not infer.This works for meeting notes, shopping lists and household expenses. The source sentence gives you a quick audit trail.
3. “Ask me three questions first”
Before working on <task>, ask up to three questions that would most change the result. List only the questions for now; wait for my answers before proposing a solution.When the task depends on audience, budget or dates, a short clarification round is faster than correcting a long answer built on the wrong assumptions.
4. “List your assumptions”
List the assumptions you will use before completing <task>. Label each one Known or To confirm. If a missing fact would change the conclusion, stop and ask me.This is a basic guardrail for contracts, quotations and itineraries. Once an assumption is visible, you can correct it before it spreads.
5. “Compare two options against my priority”
Compare <Option A> and <Option B> in a table: cost, time, risk and best-fit situation. Recommend one based on <what I value most>, and state the missing fact that could change your recommendation.Do not ask which option is “better” without saying better for what. Your priority should drive the recommendation, not the model’s default preference.
6. “Break the goal into actions”
Break <goal> into five actions I can complete today. For each action give the required input and a definition of done. Do not describe unfinished work as completed.“Definition of done” is more useful than “try harder.” It tells you when to stop and makes the next follow-up concrete.
7. “Give me a reusable template”
Design a reusable prompt for <repetitive task>. Use fill-in fields. Separate required and optional fields, specify a fixed response for missing information, and include one completed example.Turning a one-off answer into a template is when the time saving starts to compound.
8. “Offer controlled tone choices”
Rewrite the text below in three versions: A formal, B warm and direct, C extremely concise. Preserve every fact and add no promises. Label the intended recipient for each version.This is more precise than “make it professional.” You can choose between explicit options instead of guessing what professional means.
9. “Show the evidence for each item”
After completing <document/list>, check every item against the source. Output two columns: Evidence found and Not stated in the source. Never present an inference as a fact.Confident wording is not evidence. Requiring an evidence column exposes uncertainty where it actually occurs.
10. “Give me only the next action”
Based on the information so far, give me only the single most valuable next action, one reason, and one completion test. Do not list ten tasks or do later steps for me.When the answer is becoming a wall of information, this opener restores an executable pace.
The same prompt works in Claude Code and Codex CLI
These are not product-specific commands. If your material lives in a folder, paste the same prompt into either Claude Code or Codex CLI and add:
List the files you plan to read and why. Do not edit anything until I confirm.Flags, permission prompts and version behaviour differ between the tools. For file work, check each tool’s --help output first, and state “read-only” and “ask before editing” in the prompt. A travel folder, an insurance comparison or a family document bundle is just as valid as a code repository.
Three traps to avoid
State the result but not the use. “Organise these files” omits the reader and the destination format. Adding “who will read this” and “where I will paste it” is often more valuable than another paragraph of background.
Treat “unknown” as a failure. A trustworthy answer marks items as TBD. Leaving a date blank is better than accepting a plausible invented date.
Ask the same question repeatedly. When the direction is mostly right, give two or three precise corrections. Re-pasting the entire prompt throws away useful context.
Build a personal starter list
Choose three lines today: one for understanding, one for sorting, and one for checking. Save the versions that worked in your notes and replace the angle brackets next time. After a few repetitions, you are not learning to “talk to AI”; you are learning to hand over everyday work clearly.
The best first sentence is not the longest one. It is the one that leaves the next action with the least room for guessing.
Information in this article is current as of 6 August 2026 and is provided for general reference only; it does not constitute advice of any kind. Third-party product features, pricing and policies are subject to their official announcements. Essevin service details are as shown on essevin.com and in the console.