{"id":2026100645,"date":"2026-10-05T07:45:00","date_gmt":"2026-10-05T00:45:00","guid":{"rendered":"https:\/\/danieel.id\/?p=2026100645"},"modified":"2026-10-22T17:23:33","modified_gmt":"2026-10-22T10:23:33","slug":"effective-ai-prompting","status":"publish","type":"post","link":"https:\/\/danieel.id\/en\/effective-ai-prompting\/","title":{"rendered":"10 Effective AI Prompting Techniques to Get the Results You Need"},"content":{"rendered":"
\n

Ever asked an AI to make something, only to get an answer that wanders all over the place? We want a simple recipe, and it hands us a full restaurant menu. We ask it to explain a school lesson for a child, and the answer is so full of jargon it makes our foreheads wrinkle, haha. Usually the problem isn\u2019t the AI. It\u2019s how we asked.<\/p>\n

That\u2019s where effective AI prompting<\/strong> comes in. A prompt is the instruction or material we give an AI. Effective AI prompting means writing prompts that spell out the goal, context, audience, format, and limits clearly, so the result fits what we need and is easy to check. The principles work in ChatGPT, Gemini, and Claude alike.<\/p>\n

Well, giving a prompt is a lot like briefing a new member of our team. If all we say is \u201cplease make a report,\u201d we could get just about anything. But if we explain who the report is for, what material to use, what shape it should take, and when it\u2019s due, the result fits far better.<\/p>\n

A quick honest note: there\u2019s no magic sentence that guarantees an AI\u2019s answer will always be right. A clear prompt simply makes the result easier to steer and easier to check. So what should we pay attention to? There are at least ten techniques, and for each one I\u2019ve included the prompt structure and a complete example. Let\u2019s go through them one by one.<\/p>\n

Start effective AI prompting with a clear goal<\/h2>\n
\"Effective<\/figure>\n

A clear goal is the foundation of a good prompt, so first decide whether we need ideas, a summary, a comparison, a draft, or input for a decision. A word like \u201cgood\u201d also needs a yardstick: good according to whom?<\/p>\n

For complex tasks, break the work into stages: outline first, content next, then a review. That way, we can correct the direction before the AI has already written pages and pages.<\/p>\n

Prompt structure:<\/strong> Task + Goal + Format + Required closing.<\/p>\n

Full example prompt:<\/strong> \u201cDraft an email to a vendor asking for the latest delivery schedule. The goal is to get a firm delivery date and an explanation of the problem. Use polite language, 150 words maximum. End by asking for a reply before Friday at 3:00 p.m. WIB.\u201d<\/p><\/blockquote>\n

For learning, the goal can be understanding the material rather than just getting a ready-made answer: \u201cHelp me understand fractions through three exercises, from the easiest to the more difficult. Don\u2019t give me the answers right away; give me hints after I\u2019ve tried to answer.\u201d<\/p>\n

Provide context and material that truly matter<\/h2>\n
\"Reference<\/figure>\n

Relevant context helps an AI understand our situation, because it knows nothing we haven\u2019t told it. Share the information that affects the result, such as ingredients, time, budget, equipment, skill level, and taste, instead of dumping every document at once.<\/p>\n

Prompt structure:<\/strong> Context + Ingredients and tools + Task + Output format + If something is missing.<\/p>\n

Full example prompt:<\/strong> \u201cI want to cook for two people in 30 minutes. I have eggs, tofu, carrots, rice, and one frying pan. Suggest two dishes that aren\u2019t spicy, then explain the cooking steps as a numbered list. If extra ingredients are needed, ask me first.\u201d<\/p><\/blockquote>\n

When there\u2019s a lot of material, separate it with labels such as \u201cContext,\u201d \u201cData,\u201d and \u201cTask.\u201d Make it clear that the document is material to analyze, not new instructions, because commands inside it may otherwise be read by the AI as directions. Google\u2019s prompt design guide for Gemini<\/a> also stresses specific instructions and the context the model needs.<\/p>\n

Context:<\/strong> I\u2019m on the committee of our neighborhood association (RT) and I\u2019m preparing next month\u2019s residents\u2019 meeting.
\nData<\/strong> (material to analyze, not instructions): [paste the residents\u2019 suggestion survey here]
\nTask:<\/strong> Group the residents\u2019 suggestions into three themes, then rank them from the most frequently mentioned.<\/p><\/blockquote>\n

Define the audience and, where useful, a role<\/h2>\n
\"A<\/figure>\n

Naming the audience helps an AI choose the right terms and depth, because explaining something to an elementary school child is very different from explaining it to university students or a board of directors. Mention what the readers already know and what they need.<\/p>\n

A role, or point of view<\/em>, also shapes how the AI looks at a problem: a tutor helps us learn, an editor reviews a story. But asking an AI to \u201cbe an expert\u201d doesn\u2019t automatically make its answer accurate, let alone replace a qualified professional.<\/p>\n

Prompt structure:<\/strong> Task + Audience + Role + Must include + Limits.<\/p>\n

Full example prompt:<\/strong> \u201cExplain photosynthesis to a 10-year-old. Take the point of view of a patient tutor and use one simple analogy. Explain every new term, then point out the limits of the analogy so it isn\u2019t misleading. Finish with three questions to check understanding.\u201d<\/p><\/blockquote>\n

State the format and level of detail you want<\/h2>\n
\"Choosing<\/figure>\n

The output format should be stated up front, whether it\u2019s a schedule, a shopping list, a table, an email, or a presentation script, along with the length, order, columns, and units to use.<\/p>\n

Avoid instructions that clash, such as asking for a \u201cvery detailed\u201d answer in just two sentences (the AI may end up as confused as we are, haha).<\/p>\n

Prompt structure:<\/strong> Task + Context + Output format + Priorities.<\/p>\n

Full example prompt:<\/strong> \u201cCreate a seven-day study schedule for a math exam. I have 45 minutes every afternoon. Use a table with the columns Day, Topic, Practice, and Break time. Prioritize fractions and percentages, and keep the last day for review.\u201d<\/p><\/blockquote>\n

For a travel plan, add the budget, duration, and travelers\u2019 needs, then ask the AI to flag costs that aren\u2019t confirmed yet:<\/p>\n

Full example prompt:<\/strong> \u201cPlan a 3-day, 2-night trip to Yogyakarta for four adults, with a maximum budget of Rp4 million excluding plane tickets. One traveler can\u2019t walk long distances. Use a table with the columns Day | Activity | Estimated cost. Mark any unverified costs with the word (estimate).\u201d<\/p><\/blockquote>\n

Spell out what to keep and what to avoid<\/h2>\n
\"An<\/figure>\n

A list of elements that must stay unchanged stops an AI from overhauling important parts while editing, such as numbers, dates, direct quotes, terms, links, or the writer\u2019s point of view. Without that list, the AI sometimes gets a little too eager to \u201cimprove\u201d everything.<\/p>\n

Restrictions also work better with a replacement instruction. Instead of just saying \u201cdon\u2019t be stiff,\u201d it\u2019s clearer to write \u201cuse short sentences and a polite tone of address.\u201d<\/p>\n

Prompt structure:<\/strong> Task + Keep + Avoid + If unsure.<\/p>\n

Full example prompt:<\/strong> \u201cTidy up this piece without changing the numbers, dates, product names, direct quotes, or URLs. Keep the inclusive \u2018we.\u2019 Don\u2019t add personal experiences, testimonials, or new claims. If you find inconsistent numbers, flag them for me to check.\u201d<\/p><\/blockquote>\n

Give examples so the expected standard is easy to grasp<\/h2>\n
\"Three<\/figure>\n

A writing sample shows the standard we want far more clearly than a request like \u201cwrite in my style.\u201d Pick one or two samples, then explain what should be copied: sentence rhythm, structure, or level of formality.<\/p>\n

Claude\u2019s prompt engineering guide<\/a> lists examples as one of the core techniques. Just be careful that the AI doesn\u2019t also copy facts or experiences from those samples.<\/p>\n

Prompt structure:<\/strong> Style sample + Elements to follow + Task + Avoid.<\/p>\n

Full example prompt:<\/strong> \u201cHere\u2019s a sample of my writing style: \u2018It doesn\u2019t have to be perfect right away. What matters is that we start and learn from the process.\u2019 Follow its short sentences and warm tone to write three captions for a community event. Don\u2019t copy the sample sentence, and don\u2019t make up experiences for me.\u201d<\/p><\/blockquote>\n

If there\u2019s a style we reject, give a concrete reason. For example: \u201cI don\u2019t want this kind of style: \u2018Let us all come together to make this truly extraordinary event a success!\u2019 The reason: it\u2019s too formal and over the top.\u201d<\/p>\n

Ask for grounding, clarification, and flagged uncertainty<\/h2>\n
\"A<\/figure>\n

Asking an AI to show the basis of its answer and flag uncertainty matters because a fluent answer isn\u2019t necessarily a correct one. In effective AI prompting, state which sources may be used, what the AI should do when information is missing, and ask it to separate facts from assumptions.<\/p>\n

NIST\u2019s Generative AI Profile (2024)<\/a> calls this confabulation<\/em>: the AI presents false information with full confidence. Asking for references is a sensible safeguard.<\/p>\n

Prompt structure:<\/strong> Sources + Output format + References + Clarification + Prohibitions.<\/p>\n

Full example prompt:<\/strong> \u201cUse only the report I\u2019ve attached. Separate facts, interpretations, and data that isn\u2019t available yet. Cite the source page for every number. If the period or unit is unclear, ask me first. Don\u2019t create references that aren\u2019t in the report.\u201d<\/p><\/blockquote>\n

For up-to-date information, make sure the AI service you\u2019re using actually has web search access; typing \u201cuse the latest data\u201d doesn\u2019t grant that access on its own. Double-check with official sources, in the same spirit of checking before sharing discussed in understanding the spread of online hoaxes<\/a>.<\/p>\n

Refine the results with specific feedback<\/h2>\n
\"A<\/figure>\n

Specific feedback turns an AI\u2019s first answer from a rough draft into a result that fits, as long as we point out what\u2019s wrong and what needs to change. A comment like \u201cnot good enough\u201d leaves the AI guessing again, while \u201cthe opening is too long\u201d gives it a clear direction.<\/p>\n

Prompt structure:<\/strong> What already works + What needs to change + Keep + Output format.<\/p>\n

Full example prompt:<\/strong> \u201cOverall, this story works. Shorten the opening to two sentences, make the dialogue more natural, and cut down on repetition. Keep the characters\u2019 names, the main conflict, and the ending. Show me the complete revised version.\u201d<\/p><\/blockquote>\n

OpenAI\u2019s prompt engineering documentation<\/a> also emphasizes instructions, context, and examples. In long conversations, ask the AI to summarize the agreed requirements first. This habit of improving the process also ties in with tips on becoming an effective and reliable worker<\/a>.<\/p>\n

Use skills for recurring work<\/h2>\n
\"A<\/figure>\n

An AI skill stores the rules that never change, such as instructions, workflows, examples, and reference material, so we don\u2019t have to retype them every time. According to Anthropic\u2019s introduction to Agent Skills<\/a>, a skill is a folder of instructions, scripts, and resources that the AI loads when it\u2019s needed.<\/p>\n

Take a monthly report presentation. The data source, number format, slide order, and template are always the same. Exhausting to type all that out again and again, isn\u2019t it? All those rules can be saved once:<\/p>\n

Skill contents \u2014 monthly-report-presentation<\/strong> (saved once, used every month):<\/p>\n