DeepDojo.AIPublished 2026-11-06Updated 2026-11-09中文
Prompting AI: Four Elements of a Clear Request
"Help me write an event notice." In conversation, that sounds clear enough. But when it's time to draft, critical details are missing: when is the event happening? Who is it for? Which details must be included?
Anthropic uses an apt analogy in its prompting best practices guide: treat AI like a smart, newly hired colleague who doesn't yet know your internal context. If you want a result you can use immediately, spell out what you know instead of expecting the model to read your mind.
An effective, unambiguous prompt is typically assembled from four components: a specific action, distinct source material, an intended audience, and verifiable constraints. Laying these out clearly gives AI the factual grounding it needs and gives you an objective rubric for reviewing the result:
Action: What you want the AI to do
+
Material: Which facts the output may use
+
Audience: Who will read or use the result
+
Constraints: Which specific boundaries must be respected
↓
The AI knows how to draft; you know how to reviewTaking the four parts one at a time shows what each contributes. At the end they come together as a complete prompt for the event notice, so every part has a place where you can check it.
01 Action: Name What to Do
Many users paste meeting notes into a prompt and add a vague instruction, hoping the model will instinctively infer their intent:
Please handle this for me."Handle this" is one of the most common and least helpful phrases in prompt writing. Think about it: summarizing is handling, rewriting is handling, extracting action items is handling, and categorizing points is also handling. How can the AI know which one you need right now? If your immediate goal is following up on team deliverables, work backward from your final deliverable and state the action directly:
Please organize the meeting notes below into an action checklist.By specifying "an action checklist," you eliminate the guesswork. When in doubt, ask yourself: do I need a checklist, or do I need an announcement? If you need a checklist, say "extract action items"; if you need a notice, say "draft an announcement." (If the task feels too vast and you cannot define the deliverable yet, start by scaling it down as discussed in Starting with AI: How to Pick a First Task.)
02 Material: Separate Instructions from Facts
Once the action is set, provide your context. Many people blend their formatting instructions and event details into a single rambling paragraph. The AI often confuses rules with data, and verifying the draft becomes tedious.
Keep instructions and facts separate: instructions dictate what to do, while source material supplies the verified facts. OpenAI's official prompt engineering guide recommends clearly separating instructions from reference text so the model does not mistake background info for commands or miss key guidelines. Splitting them with line breaks or clear headers takes seconds and saves time on both sides. In our community reading session example, the two roles are distinct:
| Content Given to AI | Its Purpose |
|---|---|
| "Please write an event announcement" | Instruction: tells the AI what to do |
| "This Saturday, 3:00 PM to 4:00 PM, Community Activity Room" | Material: provides confirmed facts |
If your notes don't specify fees, attendee limits, or the agenda, don't assume the model knows just because past events had them. Before submitting your material, categorize each detail:
| Material Status | How to Treat It in the Prompt |
|---|---|
| Confirmed | Provide directly as verified facts for the AI to use |
| Temporarily unknown | Mark as "To be confirmed" or instruct AI to leave a placeholder |
| Essential but missing | Gather the information before generating output |
Highlighting unconfirmed details makes it easy to track who needs to verify what later. Once confirmed, you can update the draft with confidence.
03 Audience: Let the Reader Shape Content
The same event details will look very different when written for colleagues coordinating logistics versus a message sent to a neighborhood parents' chat group. Before writing, ask yourself: who will read this, and what action do they need to take next? Answering these questions clarifies what details come first and what tone to use.
Consider what parents look for when scanning a message in a busy group chat:
- Is this activity suitable for my child?
- Are the time and location convenient?
- What do we need to bring, and how do we sign up?
Keeping the parents' perspective in mind helps prioritize key details: age range, time, location, preparation, and sign-up steps belong front and center, without unnecessary pleasantries. Stating "for a neighborhood parents' group chat" guides the AI to choose an appropriate greeting, strike the right tone, and order the information logically. However, one boundary remains firm: if parents need information that isn't in your notes, confirm it with organizers first, and never let the AI invent details.
04 Constraints: Set Verifiable Boundaries
When worried that AI might generate superficial output, people often append a well-meaning plea:
Please write carefully and make it as high-quality as possible.In practice, this kind of instruction does little. It sounds strict, but you cannot objectively check the result: what does "high quality" actually mean? You get far better results by stating concrete, verifiable boundaries:
Keep the announcement under 120 words. Include the time, location, target audience, preparation checklist, and sign-up method. Do not add event arrangements not mentioned in the source material.When you receive the draft, count the words, verify that all five details are present, and check for fabricated facts. "High quality" now has an objective checklist.
Constraints can govern required inclusions, length, structure, tone, and information sources. Focus on the few requirements that directly impact your use case. Overloading a prompt with contradictory constraints often causes the model to compromise on core points. For example, an overly tight word count might force it to drop vital scheduling details. When in doubt, prioritize factual completeness over arbitrary brevity.

Breaking instructions into a clear task action, separate source material, an intended audience, and hard constraints eliminates AI guesswork and simplifies human review.
05 Assembling the Complete Prompt
Returning to the parents' group notice, the workflow becomes straightforward: define the task and audience, list the facts, and set the constraints. Putting the pieces together yields a production-ready prompt:
Please write an announcement for a neighborhood parents' group chat based on the information below.
Event: Family Reading Session
Time: This Saturday, 3:00 PM to 4:00 PM
Location: Community Activity Room
Audience: Children aged 6 to 10 and their parents
Preparation: Bring a favorite book; arrive 10 minutes early
Registration: Sign up in the group chat by Friday, 6:00 PM
Requirements:
Keep the tone warm and direct, under 120 words.
Include the time, location, target audience, preparation checklist, and sign-up method.
Do not add unmentioned details regarding fees, capacity limits, or schedules.Breaking down each component shows how it contributes to the final result:
| Prompt Component | Corresponding Content | Effect on Output |
|---|---|---|
| Action | "Write an announcement" | Determines the deliverable format |
| Material | Six verified event facts | Provides grounded facts |
| Audience | "Neighborhood parents' group chat" | Shapes tone, greeting, and sequence |
| Constraints | Under 120 words, five required items, strictly grounded | Provides objective verification criteria |
If an AI output feels off, review your original prompt first: did you state the action clearly? Did you provide the necessary facts? Did you define the audience? Are your constraints verifiable? In most cases, the issue traces back to an unstated assumption. Once you have a first draft, learn how to spot missing details and empty filler in Reviewing AI Drafts: Catching Omissions and Fluff.