How to prompt AI content: the CLEAR framework
A repeatable prompt structure — context, length, expectations, audience, role — plus the four mistakes behind most unusable AI output.

Read the guide, then turn it into a concrete plan for your brand.
Most bad AI output is not a model problem. It is a brief problem. The model answered exactly the question you asked; the question was just vaguer than you realised. This tutorial gives you a repeatable structure for writing prompts that produce usable text on the first or second attempt instead of the sixth.
The CLEAR framework
Five things decide whether a prompt lands. Write them down every time until the habit is automatic.
- Context — the background the model cannot infer: the product, the campaign, what already happened.
- Length — a number, not an adjective. "Short" means nothing; "90 words" means something.
- Expectations — the quality bar and style rules the output has to satisfy.
- Audience — who reads it, and what they already know.
- Role — the expertise the model should write from.
A prompt missing Context and Audience produces generic copy. A prompt missing Length produces an essay when you wanted a caption. A prompt missing Expectations produces something you rewrite by hand, which defeats the point.
Four moves that fix most prompts
1. Be specific
Vague: Write about marketing.
Specific: Write a 500-word blog post about email marketing strategies
for e-commerce businesses targeting millennials.
The second version constrains topic, length, industry, and audience. Each constraint removes a branch the model would otherwise have to guess.
2. Provide context
Vague: Create a product description.
Specific: Create a product description for our new eco-friendly water
bottle. Key features: BPA-free, 24-hour insulation, 750ml capacity, made from
recycled materials. Target audience: environmentally conscious outdoor
enthusiasts.
3. Define the format
Vague: Give me social media content.
Specific: Create 5 Instagram captions for a fitness brand. Each caption
should be 2-3 sentences, include a call-to-action, and use 3-5 relevant
hashtags.
4. Specify tone
Vague: Write an email.
Specific: Write a professional yet friendly follow-up email to a
potential client who showed interest in our consulting services. Use a
conversational tone while maintaining business professionalism.
Advanced techniques
Chain of thought
Break a complex task into ordered steps so the model reasons before it writes.
First, analyze the target audience for luxury watches.
Then, identify their main pain points.
Finally, write a product description that addresses these pain points.
Few-shot prompting
Two examples teach a house style faster than a paragraph describing it.
Write product titles in this style:
Example 1: "Premium Leather Wallet - RFID Protection, Slim Design"
Example 2: "Organic Cotton T-Shirt - Sustainable, Ultra-Soft Comfort"
Now create a title for noise-canceling headphones.
Role-based prompting
Act as an experienced SEO copywriter. Write a meta description for a
blog post about remote work productivity tips. Include relevant keywords
and keep it under 160 characters.
Constraint-based prompting
Write a product description with these constraints:
- Exactly 100 words
- Include 3 benefit statements
- Use power words: "revolutionary", "transform", "essential"
- End with a clear call-to-action
Reusable skeletons
Blog post
Topic: [Your Topic]
Target audience: [Describe audience]
Word count: [Number]
Tone: [Professional/Casual/Friendly]
Include: Introduction, 3-5 main points with subheadings, conclusion with CTA
SEO keywords: [List keywords]
Social media post
Platform: [Instagram/X/LinkedIn]
Brand voice: [Describe brand personality]
Goal: [Engagement/Sales/Awareness]
Include: Hook, value proposition, hashtags, CTA
Character limit: [Platform-specific limit]
Marketing email
Email type: [Welcome/Promotional/Newsletter]
Recipient: [Customer segment]
Primary goal: [Conversion/Engagement/Information]
Length: [Short/Medium/Long]
Include: Personalization, clear value, single CTA
Four mistakes worth naming
- Too vague. "Write something good" gives the model nothing to optimise against.
- Information overload. Twenty requirements in one prompt means the model quietly drops some. Keep it to three to five that matter.
- Conflicting instructions. "Be brief but comprehensive and detailed" forces the model to pick, and it will pick wrong.
- Missing context. "Rewrite this better" has no target. "Rewrite this paragraph to be more engaging for teenage readers" does.
Iterate deliberately
Change one variable at a time so you learn which change caused the improvement.
- First pass:
Write about coffee. - Second:
Write a 300-word guide on brewing the perfect espresso at home. - Third:
Write a beginner-friendly 300-word guide on brewing espresso at home, including equipment needed and step-by-step instructions.
Judge each result on four axes: does it match your intent, is it clean, does it cover every requirement, and can you ship it with light editing. When the answer to all four is yes, save the prompt. A personal prompt library compounds faster than any single clever trick.
Next
- How to prompt AI images, videos, and audio — the same discipline applied to visual and audio generation.
- 12 AI image prompt briefs to hand your agent — describe the job, let the agent write the prompt.