How to Write AI Prompts That Actually Work

Why Most AI Prompts Fail

The fundamental reason most AI prompts fail is not that the AI is incapable — it is that the prompt does not give the AI enough information to succeed. Large language models are trained to produce the most statistically likely response to a given input. When the input is vague, the most likely response is a generic, middle-of-the-road output that satisfies no one in particular.

Consider what happens when you ask a human expert for help. You would not walk up to a lawyer and say "write me a contract." You would explain the parties involved, the nature of the agreement, the key terms, and any specific concerns. The more context you provide, the better the advice. AI is exactly the same — it just requires that context to be written down explicitly.

The Prompt Writing Formula

A reliable prompt follows a consistent structure. You do not need to use all components for every task, but knowing what each component does allows you to include the right ones for each situation.

[ROLE] You are a [specific expert with relevant experience].

[CONTEXT] [Background information the AI needs to understand the task: who you are, what you're building, who the audience is, what the goal is].

[TASK] [Exactly what you want produced, with no ambiguity].

[FORMAT] [How the output should be structured: length, sections, style, tone].

[CONSTRAINTS] [What to avoid, what to include, any non-negotiable requirements].

[EXAMPLES] (optional) [One or two examples of the quality or style you want].

This formula works because it mirrors how you would brief a human professional. The role establishes expertise. The context provides the necessary background. The task states the deliverable. The format specifies the structure. The constraints prevent common failure modes. Together, they eliminate the assumption gap.

Building a Prompt Step by Step

The best way to understand prompt writing is to watch a weak prompt transform into a strong one through deliberate improvement. Here is a real example from start to finish.

Write a LinkedIn post about my new product launch.

This prompt will produce a generic LinkedIn post that could be about any product. It has no role, no context, no format guidance, and no constraints. The AI will fill every gap with defaults.

You are a senior B2B content strategist who specialises in LinkedIn marketing for SaaS companies.

Write a LinkedIn post about my new product launch.

Better — the role establishes expertise and domain knowledge. But the task is still too vague. The AI still does not know what the product is, who the audience is, or what the post should achieve.

You are a senior B2B content strategist who specialises in LinkedIn marketing for SaaS companies.

Context: I am launching a project management tool called "FlowDesk" that helps freelancers track client projects, invoices, and deadlines in one place. My LinkedIn audience is primarily freelancers and small agency owners aged 28–45. The launch is happening next Monday. I want to build anticipation and drive sign-ups to the waitlist.

Write a LinkedIn post about my new product launch.

You are a senior B2B content strategist who specialises in LinkedIn marketing for SaaS companies.

Context: I am launching a project management tool called "FlowDesk" that helps freelancers track client projects, invoices, and deadlines in one place. My LinkedIn audience is primarily freelancers and small agency owners aged 28–45. The launch is happening next Monday. I want to build anticipation and drive sign-ups to the waitlist.

Task: Write a LinkedIn post that builds anticipation for the launch and drives clicks to the waitlist sign-up page.

Format:
- Opening hook: 1 sentence that stops the scroll (do not start with "I" or "We")
- Body: 3–4 short paragraphs, each 1–3 sentences
- CTA: 1 sentence with a clear action
- Total length: 150–200 words
- Include 3–5 relevant hashtags at the end

Constraints:
- Tone: Conversational and authentic, not corporate
- Do not use buzzwords like "game-changer", "revolutionary", or "excited to announce"
- Do not use emojis except sparingly (maximum 2)

⚡ CONTENT STRATEGIST — LINKEDIN POST MODE

PRODUCT/SERVICE: [Name and one-sentence description]

AUDIENCE: [Job title, industry, company size, pain points]

POST OBJECTIVE: [awareness / waitlist sign-ups / engagement / thought leadership]

KEY MESSAGE: [The single most important thing the reader should take away]

SOCIAL PROOF (optional): [Statistic, testimonial, or result to include]

FORMAT:
- Hook style: [question / bold claim / story opener / statistic]
- Length: [short 100w / medium 200w / long 300w]
- CTA: [what you want them to do]
- Hashtags: [3–5 relevant tags]

TONE: [conversational / authoritative / vulnerable / inspiring]

CONSTRAINTS:
- Avoid: [words or phrases to exclude]
- Must include: [required elements]

Writing Prompts for Different Task Types

Different task types require different prompting strategies. Here are the key adjustments to make for the most common AI use cases.

For writing tasks (blog posts, emails, copy): Focus on tone, audience, and format. Specify the reading level, the emotional response you want to evoke, and any structural requirements. Provide an example of the style you want if possible.

You are a [type of writer] with expertise in [domain].

Audience: [Who will read this? Their knowledge level, role, and what they care about]
Purpose: [What should this writing achieve? Inform / persuade / entertain / convert]
Tone: [e.g. authoritative but accessible / warm and direct / formal and precise]
Length: [word count or range]
Format: [sections, headings, bullet points, or flowing prose]

Task: Write [specific piece] about [topic].

Key points to cover:
- [point 1]
- [point 2]
- [point 3]

For analytical tasks (research, summaries, evaluations): Specify the depth of analysis required, the criteria for evaluation, and the format for presenting findings. Ask for reasoning to be shown explicitly. Chain-of-thought prompting is particularly effective here.

For code tasks (generation, debugging, refactoring): Always specify the programming language, framework version, and any constraints on the approach. Describe the desired behaviour rather than the implementation. Include the existing code context and any error messages.

For image generation (Midjourney, DALL-E, Stable Diffusion): Use the SSSAC framework — Subject, Setting, Style, Atmosphere, Camera. Be specific about artistic style, lighting, and composition. Use negative prompts to exclude unwanted elements. See the prompt engineering techniques guide for image-specific strategies.

Iterating and Refining Your Prompts

The first output is rarely the final output. Professional AI users treat the first response as a draft to be refined, not a finished product. The key is knowing how to give effective refinement instructions.

Five Prompt Writing Mistakes to Avoid

Even experienced AI users make these mistakes. Recognising them is the first step to eliminating them from your workflow.

• ▸Writing for yourself, not the AI: The AI does not know what you know. Everything you consider obvious background knowledge needs to be stated explicitly.
• ▸Using ambiguous language: Words like "good", "professional", "detailed", and "comprehensive" mean different things to different people — and to the AI. Replace them with specific, measurable criteria.
• ▸Asking multiple unrelated questions in one prompt: Each prompt should have one clear objective. If you need multiple outputs, use multiple prompts.
• ▸Not specifying the output format: Without format guidance, the AI will choose its own structure. Always specify whether you want prose, bullet points, a table, a numbered list, or a specific document structure.
• ▸Giving up after one attempt: If the first output is not what you wanted, the answer is almost always a better prompt — not a different AI tool.

Using EMOs as Ready-Made Prompt Templates

Building great prompts from scratch takes time and practice. EMOs (Emoji-Optimised Operators) are professionally engineered prompt systems that have already been built, tested, and optimised for specific tasks. Instead of spending an hour crafting the perfect prompt for a complex task, you can start with an EMO and customise it for your specific needs in minutes.

Each EMO is built on a named framework (RISEN, CAMP, CO-STAR, SSSAC, or similar) and includes all the components of a Level 4 prompt: role, context structure, task definition, format specification, and constraints. They are designed to be used as templates — fill in the bracketed fields with your specific details and the prompt is ready to use.

The EMOAi Shop contains EMOs for every major AI tool and use case — from content creation and business analysis to image generation and video scripting. If you want to skip the prompt-building process entirely and go straight to professional results, EMOs are the fastest path.