Prompt Engineering Basics: How to Get Better Results from AI
Deliberate Academy Editorial Team
Reviewed for accuracy and professional relevance
You're 3 lessons in — don't lose your progress.
Sign up free to save where you are and earn a verified certificate when you pass.
- Explain why vague prompts produce statistically average outputs and specific prompts produce targeted ones
- Apply the four-element prompt framework — role, context, task, and format — to any professional writing request
- Identify the root cause of common prompt failures and select the appropriate fix for each
- Evaluate a weak prompt and rewrite it using the four-element structure to produce a measurably better result
You ask ChatGPT to write a proposal for a new client project. What comes back is generic, misses your firm's tone entirely, and does not reflect the client's specific situation at all. You try again with slightly different wording and get something marginally better but still not usable. Sound familiar? The problem is almost never the AI model. It is the prompt. Professionals who consistently get high-quality outputs from AI tools are not using better tools — they are writing better prompts.
Why Wording Matters More Than You Think
When you submit a prompt to an AI model, you are not issuing a command to a database. You are providing a context window that statistically constrains what the model generates next. Every word in your prompt influences the probability distribution of the output. Vague prompts produce statistically average outputs — competent, generic, and not particularly useful. Specific prompts produce outputs that are anchored to your actual situation, constraints, and quality standards.
This is not intuitive, because we are used to software that either works or does not. If you type a wrong command in a terminal, you get an error. If you type a vague prompt into ChatGPT, you get a response — just a bad one. The absence of an error message makes it easy to conclude the tool did what you asked. It did not. It did what the statistics of your prompt suggested you might want.
The good news is that the skill of writing better prompts is learnable, immediately applicable, and produces measurable results within hours of practice. You do not need to understand the model's architecture to use it well — you just need a systematic approach to specifying what you actually want.
The fastest way to improve your prompts is to read them back as if you are a new employee receiving these as written instructions. If a competent human would not know what to do, neither will the model.
The Four Elements of a Specific Prompt
Most weak prompts are missing one or more of four things: role (whose perspective and expertise level the model should write from), context (the specific information — your company, your audience, your constraint — that moves the output from a generic average to your actual situation), task (the precise action you want, not just the topic), and format (the structure, length, and shape you want back). For the full breakdown of each element, plus system prompts and few-shot examples, see Prompting AI Effectively. The rest of this lesson applies those four elements to the specific prompt failures you will hit most often in day-to-day business writing.
Proposal Drafting — B2B Sales
Context
An account director was responsible for producing customized proposals for prospective clients — each proposal needed to reflect the client's specific industry, pain points, and business size, and match the company's professional but conversational tone. She had tried using AI to help but found the outputs consistently generic: technically structured but missing the contextual specificity that made proposals feel tailored rather than templated.
Action
She rewrote her prompting approach using the four-element structure. For each proposal, her prompt now specified role (account director at a technology reseller writing for a specific industry decision-maker), context (the prospect's sector, team size, stated challenge, and the competing solution they were currently using), task (draft the executive summary and two problem-solution sections addressing the specific pain points discussed in the discovery call), and format (professional tone, conversational register, no jargon, under 300 words per section). She saved a reusable template with placeholder fields and updated the context block before each use.
Outcome
First-draft proposals required significantly less revision before being sent, and the account director reported that proposals now passed internal review with minor edits rather than extensive rewrites. She attributed the improvement almost entirely to the context element: once she supplied the prospect-specific details rather than leaving the model to generate generic positioning, the output became genuinely useful as a starting point. She trained two junior account managers on the same template structure within the following month.
A product manager asks AI to 'help with the roadmap presentation.' The output is generic and misses the company's specific priorities. Which single addition to the prompt would most likely produce a usable result?
Select one answer.
Before and After: Four Prompt Comparisons
Example 1: Meeting summary
Weak: "Summarize this meeting."
Strong: "You are a project manager. Summarize the following meeting transcript in three sections: decisions made, action items (with owner and deadline if mentioned), and open questions. Use bullet points. Keep the summary under 200 words."
Example 2: Job description
Weak: "Write a job description for a marketing manager."
Strong: "Write a job description for a Marketing Manager role at a 40-person B2B SaaS company targeting mid-market HR teams. The role owns demand generation and content. Required experience: 4+ years in B2B marketing, familiarity with HubSpot. Tone: professional but direct. Format: Job summary (2 sentences), Responsibilities (6 bullets), Requirements (5 bullets), Nice to have (3 bullets). Avoid jargon and overly aspirational language."
Example 3: Data interpretation
Weak: "Analyze this data."
Strong: "You are a business analyst. The table below shows our monthly website traffic, leads, and trial signups for the past six months. Identify the two most significant trends, flag any anomalies, and suggest one hypothesis for each that our team should investigate. Write in plain English, no jargon."
Example 4: Email response
Weak: "Help me reply to this client email."
Strong: "Draft a reply to the client email below. The client is frustrated that the project has slipped by two weeks. We are responsible for the delay. The reply should: acknowledge the delay directly without excessive apology, give one concrete reason (team capacity), confirm the new delivery date (June 20), and offer a brief status call. Tone: professional, accountable, forward-looking. Keep it under 150 words."
You do not need to write every prompt this way from scratch. Build a personal library of your best-performing prompts and reuse them as templates, updating the context-specific sections as needed. This is covered in depth in the Prompt Engineering for Business course.
The Fastest Fixes for Common Prompt Problems
Output is too generic: Add specific context about your company, audience, or situation. The model cannot know your particulars — you must supply them.
Output ignores key constraints: State constraints explicitly as requirements, not assumptions. If the output must be under 100 words, say so. If it must not mention a competitor, say so.
Output is the wrong length: Always specify length. "Short" means different things to different people — and to the model, it means nothing specific.
Output is in the wrong format: Specify format explicitly. "Bullet points," "numbered list," "table with three columns," "executive summary paragraph followed by detailed bullets."
Output misses the tone: Give the model a sample. "Match the tone of this example: [paste your example]" is more reliable than describing tone in adjectives.
Still getting poor results after revision: Break the task into smaller steps. Instead of asking for a complete strategic document in one prompt, generate the outline first, then expand each section separately. This approach — called chaining — dramatically improves quality on complex outputs.
A project manager asks AI to help draft a client update email. Their prompt reads: 'Write an email to my client about the project.' The output is vague, overly formal, and misses the key issue the client is waiting to hear about. Which of the following is the most appropriate next step?
Select one answer.
Exercise
Your Task
Choose one specific recurring professional task — a type of document, email, or analysis you produce regularly. Write the weak version of the prompt you would have used before this lesson (one sentence, no structure). Then write the strong version using all four elements: role, context, task, and format. Run both in the same AI tool and write two sentences comparing the outputs: what the strong prompt produced that the weak one did not, and which of the four elements made the biggest difference in this case.
Success looks like
- Your strong prompt includes all four elements — a specific role, the real context of your situation, a precisely defined task, and an explicit output format
- The comparison identifies a concrete difference in the output — not just 'it was better' but what specifically changed
- You have identified which single element had the highest impact, which tells you where your previous prompts were most underspecified
Watch out for
- Writing a weak prompt that is intentionally terrible — the comparison is only useful if the weak version is what you would genuinely have written before reading this lesson
- Adding all four elements but keeping the context vague — the context element is where most of the quality gain comes from, and generic context produces generic output even with the other elements in place
Hint
The role element changes the model's register and depth; the context element is what eliminates generic output; the task element prevents the wrong deliverable; and the format element prevents the right content in an unusable structure. If you are unsure which matters most for your task, try removing each one in turn.
Try It: AI-Graded Practice
The exercise below grades your rewrite automatically against the four-element framework, so you get direct feedback on whether it holds up.
- Vague prompts produce statistically average outputs — specific prompts produce outputs anchored to your actual situation, audience, and requirements.
- Every effective prompt addresses four elements: role, context, task, and format — missing any one of these is typically the cause of a weak output.
- The fastest prompt improvements come from adding specific context about your situation, stating constraints explicitly, and specifying output length and format.
- When a single prompt is not producing good results, break the task into sequential steps — complex outputs almost always benefit from a chained approach.
- Build a library of your best prompts and reuse them as templates — prompt engineering compounds over time, and each refinement improves every future use.