What is prompt engineering?
Prompt engineering is the practice of crafting instructions that help AI systems produce better, more useful outputs. When you interact with AI tools like ChatGPT, Claude, or Gemini, the way you phrase your request directly impacts the quality of the response you receive.
Consider it similar to giving directions to someone unfamiliar with your city. Saying "go to the store" produces different results than "take the second left on Main Street, then look for the blue building with the pharmacy sign on the corner." Both requests aim for the same destination, but one provides the clarity needed for success.
This guide introduces three foundational techniques that improve AI interactions for anyone—whether you're writing emails, making business decisions, or building products.
Technique 1: Be specific with your instructions
The most common mistake in AI interactions is vague requests. AI systems perform significantly better when instructions are clear and detailed.
The principle
Vague instructions force the AI to make assumptions. These assumptions may not align with your intentions. Specific instructions reduce guesswork and increase the likelihood of receiving what you actually need.
What this looks like
Vague instruction:
"Write something about our product."
Specific instruction:
"Write a 150-word product description for our project management software. Target audience: small business owners. Tone: professional but approachable. Include: key features (task tracking, team collaboration, deadline management) and one customer benefit."
The second request defines length, audience, tone, content requirements, and purpose. The AI has everything needed to produce a relevant output.
Practical examples
Everyday use:
"Explain how compound interest works to someone with no financial background. Use a simple example with $1,000 saved over 5 years. Keep the explanation under 100 words."
Product Manager:
"Summarize the key user complaints from these support tickets. Group them into categories. For each category, note how many tickets mention this issue and suggest one potential solution. Format as a table."
Founder:
"Analyze whether we should expand to the European market. Consider: our current resources ($500K runway, 12-person team), the regulatory requirements (GDPR), and timeline to first revenue. Provide a recommendation with reasoning."
How companies use this: Specific instructions in AI products
Notion AI (writing assistant):
"Summarize this document in 3 bullet points. Focus on action items and decisions made. Keep each bullet under 20 words."
Notion's AI doesn't just "summarize"—it provides specific parameters (3 bullets, focus areas, length limits) to ensure consistent, useful outputs.
Grammarly (writing enhancement):
"Rewrite this sentence to be more concise. Maintain the original meaning. Target a professional tone. Reading level: undergraduate."
Grammarly's suggestions are powered by specific instructions about length, meaning preservation, tone, and audience level.
Canva Magic Write (design copy):
"Write a headline for a Instagram post promoting a summer sale. Maximum 8 words. Tone: exciting and urgent. Include an emoji."
Canva constrains the AI with platform-specific requirements (character limits, tone for social media) to produce immediately usable copy.
Technique 2: Provide relevant context
AI systems lack knowledge of your specific situation. Providing context helps them tailor responses to your actual needs rather than generating generic answers.
The principle
Context includes any information that shapes what a "good" answer looks like for your situation. This might include your role, your audience, your constraints, or relevant background information.
Without context, AI provides general responses. With context, AI provides relevant responses.
What this looks like
Without context:
"How should I respond to this customer complaint?"
With context:
"I'm a customer success manager at a B2B software company. A long-term client (3 years, $50K annual contract) complained that our latest update removed a feature they relied on. They're frustrated and mentioned considering alternatives. How should I respond to acknowledge their concern while explaining that we're working on a solution? Our engineering team estimates 2 weeks to restore the feature."
The contextual version enables a response that fits the specific relationship, stakes, and available solutions.
Practical examples
Everyday use:
"I need to decline a wedding invitation from a close friend. We've been friends for 15 years, but I have a family commitment that same weekend that I cannot change. Help me write a message that's warm and genuine while being honest about why I can't attend."
Product Manager:
"Our mobile app has a 3.2 star rating. Main complaints: slow loading times and confusing navigation. We have 6 weeks of engineering time before our next release. Which issue should we prioritize, and why? Consider that 60% of our users are on older devices."
Founder:
"We're a Series A startup with 18 months of runway. Our competitor just raised $50M and announced they're entering our market segment. Our advantages: better product-market fit with mid-market customers and 40% lower pricing. What strategic options should we consider?"
How companies use this: Context-aware AI products
Spotify (personalized recommendations):
"Recommend songs for this user. Context: They listen primarily to indie rock (70%), some electronic (20%), rarely pop (10%). Current time: Friday 8pm. Recent listening: upbeat tracks. Last skipped: slow ballads. Listening history: 3 years."
Spotify's AI uses extensive context—listening history, time of day, recent behavior—to make recommendations that feel personally relevant.
LinkedIn (job matching):
"Match this candidate to relevant jobs. Context: 8 years experience in marketing, currently Senior Manager level, location preference: remote or San Francisco, salary expectation: $150-180K, recently viewed: tech startups, has applied to: product marketing roles."
LinkedIn's matching considers explicit preferences plus behavioral signals to surface relevant opportunities.
Duolingo (adaptive learning):
"Generate a practice exercise for this learner. Context: Learning Spanish, intermediate level (B1), struggles with subjunctive tense (failed 60% of recent exercises), strong with vocabulary (92% correct), prefers short exercises, currently on a 45-day streak, learns best in evening sessions."
Duolingo personalizes exercises based on skill gaps, learning patterns, and engagement history.
Netflix (content recommendations):
"Recommend content for this profile. Context: Watches primarily thriller and documentary genres, average session: 2 hours, typically watches after 9pm, recently finished: true crime series, abandoned after 10 minutes: romantic comedies, shares account with family (but this is their personal profile)."
Netflix uses viewing patterns, time preferences, and completion rates to recommend content likely to engage each specific user.
Technique 3: Specify your desired output format
Telling the AI how you want information structured often matters as much as what information you request. Format specification saves time and produces immediately usable outputs.
The principle
Different situations require different formats. A brainstorming session benefits from bullet points. A board presentation requires structured analysis. An email needs conversational prose.
When you specify format, you receive outputs that fit directly into your workflow without additional editing.
What This Looks Like
Without format specification:
"Give me ideas for improving customer retention."
With format specification:
"Give me 5 ideas for improving customer retention. For each idea, provide:
- •The idea (one sentence)
- •Why it works (one sentence)
- •Implementation difficulty (Low/Medium/High)
- •Expected impact (Low/Medium/High)
Format as a numbered list."
The second request produces an organized, comparable set of options ready for team discussion.
Practical examples
Everyday use:
"Create a weekly meal plan for a family of four. Format as a table with columns for: Day, Breakfast, Lunch, Dinner. Include one vegetarian dinner. Keep meals simple (under 30 minutes to prepare)."
Product Manager:
"Compare these three feature options for our roadmap. Create a table with columns for: Feature Name, User Impact, Engineering Effort (weeks), Revenue Potential, Recommendation. End with a one-paragraph summary of which feature to prioritize."
Founder:
"Analyze our Q3 performance. Structure your analysis as:
- •Executive Summary (3 bullet points)
- •Key Wins (what went well)
- •Areas for Improvement (what didn't)
- •Q4 Priorities (recommended focus areas)
Keep the entire analysis under 500 words."
How companies use this: Structured outputs in AI products
Jasper AI (marketing content):
"Generate a Google Ad. Format requirements:
- •Headline 1: Maximum 30 characters
- •Headline 2: Maximum 30 characters
- •Headline 3: Maximum 30 characters
- •Description 1: Maximum 90 characters
- •Description 2: Maximum 90 characters
Include the keyword 'project management software'. Tone: professional."
Jasper structures outputs to match exact platform requirements—ads that are immediately ready to deploy.
Otter.ai (meeting transcription):
"Process this meeting recording. Output format:
- •Summary: 3-5 sentences covering main topics
- •Action Items: Bulleted list with owner name and deadline if mentioned
- •Key Decisions: Numbered list of decisions made
- •Follow-ups Needed: Questions that remained unresolved
Timestamp important moments."
Otter's AI produces structured meeting notes that integrate directly into workflow tools.
HubSpot AI (CRM insights):
"Analyze this sales call. Provide output in this format:
- •Deal Stage Recommendation: [Stage name]
- •Confidence Score: [Percentage]
- •Key Buyer Signals: [Bulleted list]
- •Objections Raised: [Bulleted list]
- •Suggested Next Steps: [Numbered list]
- •Risk Factors: [If any]"
HubSpot formats AI insights to match CRM fields and sales workflow stages.
Slack AI (channel summaries):
"Summarize this channel's activity from the past 24 hours. Format:
- •Key Discussions: Top 3 conversation threads with brief summary
- •Decisions Made: Any conclusions reached
- •Action Items: Tasks mentioned with @mentions preserved
- •Unresolved Questions: Topics needing follow-up
Keep total summary under 200 words. Link to original messages."
Slack's AI produces summaries formatted for quick scanning, with links back to source conversations.
Putting it together
These three techniques work together. A well-crafted prompt typically includes:
- •Specific instructions — What exactly you want
- •Relevant context — Information that shapes the right answer
- •Output format — How you want it structured
Combined example
Weak prompt:
"Help me with my presentation."
Strong prompt:
"I'm presenting our product roadmap to the executive team next week (context). Create an outline for a 15-minute presentation (specific instruction) that covers: Q3 accomplishments, Q4 priorities, and resource needs. Format as bullet points with timing estimates for each section (output format). The executives care most about revenue impact and competitive positioning (additional context)."
The strong prompt provides everything the AI needs to generate a useful, relevant response.
Key takeaways
- •Specificity reduces guesswork. The more precise your instructions, the closer the output matches your needs.
- •Context enables relevance. AI cannot read your mind. Provide the background information that shapes what "good" looks like.
- •Format determines usability. Specifying structure produces outputs you can use immediately.
These three techniques form the foundation of effective prompt engineering. Mastering them improves every AI interaction, regardless of the tool you use or the task you're completing.
What's next
Part 2 introduces intermediate techniques: Few-Shot Learning (teaching AI through examples) and Chain-of-Thought Reasoning (getting AI to show its work). These methods produce higher-quality outputs for complex tasks.
Part 3 covers advanced techniques for specific professional contexts: validation, critique, and multi-perspective analysis.
Part 4 addresses testing and optimization: how to systematically improve prompts and compare results across different AI models.
This is Part 1 of a four-part series on prompt engineering techniques.
