Giving Clear Instructions

Why Clear Instructions Matter

An AI assistant can only work with the information and instructions provided in the prompt. If the instruction is vague, the AI may have to guess what the user means.

Compare these two requests:

Vague: Make this report better.

Clear: Rewrite the report summary so that it is easier for a non-technical manager to understand. Keep the main facts and limit the summary to 200 words.

The second instruction gives the AI a much clearer task.

Use Specific Action Words

Tell the AI exactly what action you want it to perform.

Useful action words include:

• Summarize.
• Compare.
• Rewrite.
• Extract.
• Classify.
• Translate.
• Analyze.
• Explain.
• Organize.
• Generate.
• Identify.

For example, instead of saying Look at this document, say Summarize the three main findings in this document.

Define the Exact Task

Tell the AI what it should do with the information you provide.

For example:

Read the following customer comments and classify each one as Positive, Neutral, or Negative.

This is clearer than:

Analyze these customer comments.

Define the Scope

Scope tells the AI which part of a subject or document it should focus on.

For example:

Review only the financial risks mentioned in the report.

This prevents the task from becoming unnecessarily broad.

Specify What to Include

If certain information must appear in the result, state it explicitly.

For example:

Include the problem, its likely cause, and the recommended next step.

This gives the AI clear output requirements.

Specify What to Exclude

Sometimes it is equally important to tell the AI what not to include.

For example:

Do not include background information that is unrelated to the current issue.

Another example:

Do not make assumptions about information that is not present in the source document.

Avoid Vague Words

Words such as good, better, appropriate, interesting, and short can have different meanings.

When precision matters, explain what the word means in the context of the task.

Instead of:

Make the email short.

try:

Keep the email under 150 words.

Define the Audience

The intended audience can affect the instruction.

For example:

Explain the concept for a beginner with no technical background.

This gives the AI a clear indication of the required level.

Specify the Desired Level of Detail

If you want a particular level of detail, say so.

For example:

Give a brief explanation with two practical examples.

or:

Provide a detailed explanation covering the main concepts and important exceptions.

Break Complex Tasks Into Steps

Complex instructions can be easier to follow when they are divided into clear steps.

For example:

1. Summarize the report.
2. Identify the three main risks.
3. Explain the potential impact of each risk.
4. Recommend one action for each risk.

This is clearer than asking the AI to analyze the report and tell me what to do.

Use Numbered Requirements

Numbered instructions are useful when a task contains several requirements.

For example:

1. Summarize the document.
2. Identify five important findings.
3. Explain why each finding matters.
4. Present the result in a table.

This structure makes it easier to check whether all requirements were addressed.

Separate the Task From the Source Material

When a prompt contains a large amount of source material, clearly identify what the AI should do with it.

For example:

Task: Identify the three main causes of the problem.

Source material: [document text]

This separation can make the prompt easier to interpret.

State Important Constraints

If the task has important boundaries, include them in the instruction.

Examples include:

• Word limit.
• Number of items.
• Time period.
• Required sources.
• Required audience.
• Topics to avoid.

For example:

Identify exactly five recommendations based only on the information provided.

Be Precise About Numbers

If you need a specific number of results, say so.

Compare:

Give me some ideas.

with:

Give me exactly five ideas.

The second instruction is much easier to evaluate.

Define Time Periods

When dates or time periods matter, specify them.

For example:

Analyze sales from January through June 2026.

This is more precise than:

Analyze recent sales.

Clarify Ambiguous Terms

If a term could have several meanings, define what you mean.

For example:

By recent customers, I mean customers who made a purchase within the last 90 days.

This prevents the AI from choosing its own interpretation.

State the Priority

Sometimes several requirements compete with one another.

Tell the AI which requirement is most important.

For example:

Prioritize factual accuracy over brevity. If necessary, exceed the suggested word count to preserve important information.

This gives the AI guidance when it has to balance requirements.

Avoid Contradictory Instructions

Conflicting instructions can create confusing prompts.

For example:

Give a detailed explanation in exactly 50 words.

The requirements may conflict depending on the subject.

If two requirements can conflict, clarify which one has priority.

Use Clear Output Requirements

Tell the AI what the final result should contain.

For example:

Return a table with four columns: issue, cause, impact, and recommended action.

This is clearer than simply asking the AI to organize the findings.

Give Instructions in a Logical Order

A complex prompt is easier to understand when related instructions are grouped together.

A practical order is:

1. Task or goal.
2. Relevant context.
3. Specific actions.
4. Constraints.
5. Output requirements.

This is not a strict rule, but it is a useful structure.

One Instruction Can Contain Several Actions

A single prompt can request several related actions.

For example:

Summarize the report, identify the three biggest risks, and recommend one action for each risk.

The actions are related and form one coherent task.

Separate Unrelated Tasks

If a prompt contains several unrelated tasks, consider separating them into distinct sections.

For example:

Task 1: Summarize the report.

Task 2: Draft an email based on the summary.

This can be clearer than combining unrelated instructions into one sentence.

Tell the AI What Not to Assume

When assumptions could create problems, explicitly restrict them.

For example:

Use only the information provided. If information is missing, identify the missing information instead of guessing.

This can be useful for analysis and document-based tasks.

Instructions Should Match the Goal

Make sure the instructions actually help accomplish the stated goal.

For example, if the goal is to create a beginner friendly explanation, instructions that require highly technical language would work against the goal.

Use Before and After Comparison

One useful way to improve instructions is to compare a vague prompt with a clearer version.

Before: Make this presentation better.

After: Rewrite the presentation outline so that the main argument is clear to a non-technical business audience. Keep the existing facts, remove repetitive points, and organize the content into five sections.

The second version specifies the action, audience, boundaries, and structure.

Another Example: Data Analysis

Vague: Analyze these sales numbers.

Clear: Analyze the monthly sales data from January through June. Identify the month with the largest decline, calculate the percentage change from the previous month, and explain two possible factors using only the information provided.

The second instruction clearly defines the scope and required actions.

Another Example: Writing

Vague: Write a customer email.

Clear: Write a professional email to a customer whose order is delayed by two days. Explain the reason for the delay, apologize, provide the revised delivery date, and keep the message under 150 words.

Another Example: Learning

Vague: Teach me about neural networks.

Clear: Explain the basic idea of neural networks to a complete beginner. Use simple language, avoid mathematical formulas, and give two everyday examples.

Review the Response

After receiving an AI response, check whether the instructions were followed.

Ask:

• Did the AI perform the requested action?
• Did it stay within the requested scope?
• Did it include the required information?
• Did it avoid excluded information?
• Did it follow the requested length?
• Did it use the required format?

Refine the Instruction

If the result is not what you wanted, identify the missing or unclear instruction.

For example, if the response is too long, add a word limit.

If it discusses irrelevant topics, narrow the scope.

If it misses an important point, explicitly require that point.

If the structure is inconvenient, specify the desired format.

Clear Instructions Are Iterative

You do not need to write the perfect instruction on the first attempt.

Prompting is often an iterative process:

1. Write the initial instruction.
2. Review the response.
3. Identify what was missing or unclear.
4. Improve the instruction.
5. Run the task again.

A Practical Instruction Framework

A useful structure is:

Action + Scope + Requirements + Constraints + Output

For example:

Summarize the attached report. Focus on financial risks. Include the three most important risks and their potential impact. Keep the summary under 300 words. Present the result using headings and bullet points.

This framework can help you construct clear instructions for many different tasks.

Do Not Over-Instruct Simple Tasks

Clear instructions do not mean adding unnecessary instructions.

For a simple task, keep the request simple.

For example:

Translate this sentence into French.

There is no need to add many unrelated requirements.

What You Should Remember

• Clear instructions tell the AI exactly what action to perform.
• Specific action words reduce ambiguity.
• Define the scope when only part of a subject matters.
• State what should be included and excluded.
• Use measurable requirements when precision matters.
• Break complex tasks into clear steps.
• Numbered requirements can make multi-part tasks easier to follow.
• Avoid contradictory instructions.
• State priorities when requirements may compete.
• Review the AI response and refine instructions when necessary.
• Clear does not mean unnecessarily long.

What Comes Next?

The next lesson will focus specifically on specifying output format so that the AI produces information in a structure that is easy to read, evaluate, and use.