Prompting 101: จากทริกสู่การสั่งงานให้ชัดPrompting 101: from tricks to clear requests

Prompting is not about finding one secret sentence that unlocks the model.

The useful version is simpler:

Prompting is the skill of explaining work clearly enough that AI can help without guessing too much.

That makes it a basic skill, like Markdown or terminal. It is not only for developers. Anyone who asks AI to write, summarize, compare, plan, research, translate, or review work is already prompting.

The evolution of prompting

Prompting has gone through a few phases.

Phase 1: one-shot asking

Early everyday AI use looked like this:

Write an email to my customer.

Sometimes it worked. Often it produced something generic because the model did not know the customer, goal, tone, relationship, or boundary.

Phase 2: prompt tricks

People started adding tricks:

Act as a world-class marketer.
Think step by step.
Make it 10x better.

Some tricks helped, especially with older models or narrow tasks. But tricks also made people think prompting was magic. It is not.

Phase 3: structured prompting

Provider docs now converge on a more practical idea: give clear instructions, useful context, constraints, examples, and the desired output shape.

Task: Rewrite this customer email.
Context: The customer is upset because delivery is 3 days late.
Tone: Calm, accountable, not defensive.
Output: Thai email, 120 words or less.
Check: Do not promise a refund. Ask for confirmation of the new delivery address.

This works because the AI has less to invent.

Phase 4: context engineering

For more advanced systems, prompting has expanded into context engineering: deciding what information, documents, examples, tools, and history should be available to the model at the right moment.

For a beginner, the lesson is practical:

Better AI output usually comes from better supplied context, not louder instructions.

Phase 5: outcome-first prompting

The newest model guidance also warns against dragging old, process-heavy prompt stacks into newer models. Stronger models often work better when you define the outcome, constraints, evidence, and final answer shape, then leave room for the model to choose an efficient path.

That does not mean vague prompts are good. It means modern prompts should be clear and scoped, not overloaded.

The modern prompt mindset

Modern prompting has four rules.

1. Say the job.
2. Give the context.
3. Define the output.
4. Check the result.

Use this starter pattern:

I want you to help me [task].

Context:
- [who this is for]
- [what happened]
- [what matters]

Output:
- [format]
- [length]
- [language]
- [tone]

Before answering, ask me up to 3 questions if anything important is missing.

The last line matters. It turns the model from a guessing machine into a collaborator.

Bad prompt, better prompt

Bad:

Make this better.

Better:

Rewrite this announcement for Thai office workers.

Goal: make the message clear, polite, and easy to act on.
Tone: professional but warm.
Length: under 120 words.
Keep: the deadline and the contact person.
Avoid: sounding like a legal notice.

Text:
[paste draft]

The better prompt does not use fancy language. It simply gives the model enough work context.

Practice

Rewrite this weak prompt:

Help me with my presentation.

Use the modern prompt pattern:

Task:

Context:

Audience:

Output:

Constraints:

Ask first if:

Your goal is not to make the prompt long. Your goal is to remove important guessing.

References

• OpenAI Prompting — prompting as an iterative practice, prompt versioning, variables, and evals.
• OpenAI Prompt Engineering — message roles, Markdown/XML structure, few-shot examples, and context.
• OpenAI Prompt Guidance — latest GPT-5.5 guidance on shorter outcome-first prompts, constraints, evidence, and final answer shape.
• OpenAI GPT-5.2 Prompting Guide — current guidance on scope, verbosity, research bars, and eval-driven prompt migration.
• Anthropic Prompt Engineering Overview — start with success criteria and evaluation before tuning prompts.
• Anthropic: Effective Context Engineering for AI Agents — context engineering as the next step beyond prompt wording.
• Google Gemini Prompt Design Strategies — clear instructions, constraints, examples, decomposition, and iteration.
• Microsoft Foundry Prompt Engineering Techniques — clear syntax, task breakdown, affordances, and citation-based grounding.
• Wei et al., 2022: Chain-of-Thought Prompting — the research paper that made step-by-step reasoning prompts famous.
• Wang et al., 2022: Self-Consistency — sampling multiple reasoning paths and choosing the most consistent answer.
• Yao et al., 2022/2023: ReAct — combining reasoning with actions such as tool use and search.
• Sahoo et al., 2024: Systematic Survey of Prompt Engineering — broad taxonomy of prompting techniques and applications.
• Dong et al., 2025: Survey of Context Engineering — newer framing around context retrieval, processing, and management.