Introduction
I’ve been using AI tools daily since starting and completing my MSc in Data Science and Artificial Intelligence at the University of Hull. As a consultant helping businesses implement AI—and through my work at TIIQU building research databases and creating content for platforms like My Thriving Child—I’ve tested thousands of prompts across ChatGPT, Claude, Gemini, and other tools.
Here’s what I’ve learned: when someone says “AI doesn’t work well,” what they really mean is “I’m prompting badly.”
The key difference? Mediocre AI outputs versus genuinely useful results isn’t about the tool. Instead, it’s about how you communicate with it. Good prompting is a skill based on clarity, context, and intent.
In this article, I’m not giving you theory. Rather, I’m showing you the exact techniques I use every day for data analysis, web scraping, content creation, and client work. These are real examples from my actual projects, complete with before-and-after comparisons.
Let’s dive in.
Technique 1: Use Role/Persona Prompts
Firstly assigning the AI a specific role shapes its expertise, perspective, and tone. This is the single most powerful technique I use.
Real Example from My Work at TIIQU
When I was building web scraping systems to extract researcher information from university websites, I started with a vague prompt. Here’s what happened:
Vague Prompt:
Write Python code to scrape researcher information from a university website.
View ChatGPT response – Bad scraping prompt + generic code response
The result? Basic code with no error handling, no structure, and nothing I could actually use in production.
Then I applied the role technique:
Good Prompt:
Act as a senior Python developer specializing in web scraping for academic data extraction.
Context: I'm building a database of UK university researchers for organizational analysis at TIIQU. I need to scrape researcher profiles from multiple university websites.
Task: Write production-quality Python code to extract:
- Researcher name
- Email address
- Research areas/interests
- Department
- Publications count (if listed)
Requirements:
- Use BeautifulSoup and requests libraries
- Include robust error handling (timeouts, missing fields, connection errors)
- Handle different HTML structures across universities
- Add rate limiting (2-second delay between requests)
- Include detailed comments
- Output to CSV with proper headers
- Log any failed extractions
Constraints:
- Code should be modular (separate functions for fetching, parsing, saving)
- Must handle missing data gracefully (use "N/A" for unavailable fields)
- Include example usage
View ChatGPT response – Good scraping prompt + professional code response
The Difference
The second prompt gave me production-ready code with proper error handling, modular structure, CSV export functionality, and rate limiting. Consequently, this saved me approximately 4 hours of debugging. Additionally, it worked reliably across multiple university websites with different HTML structures.
Why It Works
Role prompts provide context about depth, expertise level, and audience. Therefore, the AI doesn’t just generate code—it generates code from the perspective of an experienced developer who understands production requirements.
Technique 2: Give Context Before the Task
AI models have no memory of your previous work or understanding of your situation. However, context eliminates generic responses and tailors output to your specific needs.
Real Example from My AI Consulting Work
When analyzing client data, I learned this lesson the hard way.
Bad Prompt:
Analyse this sales data and give me insights.
[Please upload data as needed]
View ChatGPT response – Bad data analysis prompt + generic response
The response was surface-level: “Sales are declining” and “Some products perform better than others.” Unfortunately, nothing I could present to a client.
Good Prompt:
Act as a senior business intelligence analyst specializing in e-commerce and retail analytics.
Context: I'm consulting for a UK-based online retailer. They've experienced a 15% revenue decline over the past 6 months despite stable website traffic. The CEO needs 3 specific, actionable recommendations for their Q1 strategy meeting next week.
Data: 6 months of transactional data including date, product, category, revenue, units sold, region (London, Manchester, Birmingham, Leeds), and customer type (New vs Returning).
Task: Analyze this data and provide:
1. Top 3 data-driven insights explaining the revenue decline
2. For each insight, explain:
- What the data shows (specific numbers/percentages)
- Why this matters for their business
- One concrete action they can take immediately
3. Identify which customer segment or region offers the best opportunity for recovery
Output format:
- Executive summary (3 sentences)
- Insight 1: [Pattern] → [Why it matters] → [Action]
- Insight 2: [Pattern] → [Why it matters] → [Action]
- Insight 3: [Pattern] → [Why it matters] → [Action]
- Recommended priority focus
Constraints:
- Keep recommendations specific and actionable
- Use business language, not statistical jargon
- Focus on what they can implement in the next 30 days
[Sample data provided]
View ChatGPT response – Good data analysis prompt + detailed structured response
The Difference
A vague prompt (“analyse this data”) produces descriptive insights because the goal, audience, and actions are unclear. In contrast, a well-prompted request defines the role, business context, and constraints (CEO audience, revenue decline, 30-day actions). As a result, this forces prioritisation and actionable recommendations.
Furthermore, the difference is outcome-driven. Vague prompts explain what happened, while well-prompted prompts explain what to do next. Ultimately, this results in clearer decisions, faster execution, and higher business value.
I presented this analysis directly to the client. They implemented all three recommendations.
Why It Works
Context narrows focus. Without it, AI gives you generic observations. With it, you get tailored insights that reflect your actual business situation, constraints, and goals.
Technique 3: Break Complex Tasks Into Steps
For multi-part tasks, guide the AI with numbered steps. This leverages AI’s strength in sequential reasoning.
Real Example from Content Creation
When creating educational content for My Thriving Child, I needed a podcast script about supporting children through divorce.
Bad Prompt:
Write a podcast script about helping children through divorce.
View ChatGPT response – Generic advice that could have come from any parenting blog
Good Prompt:
Act as a child therapist and content creator specializing in family transitions.
Context: I'm creating a 5-minute podcast episode for parents navigating divorce. The audience is primarily mothers aged 30-45 dealing with guilt and uncertainty about their children's wellbeing.
Task - Complete these steps in order:
Step 1: Identify the top 2 emotional concerns parents have about their children during divorce
Step 2: For each concern, provide one evidence-based strategy that's immediately actionable
Step 3: Include a real-world example scenario showing how to apply each strategy
Step 4: Structure this as a conversational podcast script with natural transitions
Step 5: End with a reassuring message that validates their efforts
Constraints:
- Empathetic tone, not clinical
- No jargon
- Each strategy should take 60-90 seconds to explain
- Include a brief intro and outro
The Difference
Prompt 1: Broad, general podcast about helping children through divorce. Audience unspecified. Flexible tone. Longer, multi-segment script.
Prompt 2: Highly specific—5-minute episode for mothers 30–45. Structured steps (concerns, strategies, examples). Actionable, 60–90 second strategies. Empathetic tone. Real-world focus. Validates parental efforts.
In short, Prompt 1 is general guidance, while Prompt 2 is targeted, concise, and audience-focused.
Why It Works
Breaking tasks into steps mirrors how humans solve complex problems. Accordingly, AI follows the structure you provide, resulting in more organized, coherent outputs.
Technique 4: Add Constraints for Precision
Constraints eliminate fluff. Specifically, specify word count, tone, format, or audience to get focused, actionable responses.
Real Example from Technical Documentation
When explaining AI concepts to non-technical clients:
Effective Prompt:
Explain how Claude's Constitutional AI training works.
Constraints:
- Maximum 150 words
- Audience: Business executives with no technical background
- Tone: Professional but conversational
- Avoid: Jargon like "RLHF," "transformer architecture," "neural networks"
- Include: One practical implication for businesses considering Claude API
- Format: 3 short paragraphs
The result: Concise, business-appropriate explanation that clients could understand and use in decision-making conversations with their teams.
Why It Works
AI models can be verbose. However, constraints force focused, relevant responses that match your specific requirements.
Technique 5: Request Examples, Not Just Theory
Examples transform abstract advice into practical tools you can use immediately.
Real Example from Client Presentations
When creating dashboards for consulting clients, I learned that abstract principles don’t stick. Instead, clients need to see the difference.
Weak Prompt:
Explain data visualization best practices.
Result: Generic advice like “use clear labels” and “choose appropriate chart types.” True, but not actionable.
Strong Prompt:
Explain 3 data visualization best practices for business dashboards. For each principle, provide:
- One example of doing it wrong (describe what makes it confusing)
- One example of doing it right (describe what makes it effective)
- When this principle matters most
- One tool or technique to implement it
Context: I'm presenting monthly performance dashboards to C-suite executives who have 5 minutes to review them.
The Difference
The improved prompt gave me specific guidance on executive dashboards. For example: avoid 3D charts (hard to read quickly), use consistent color coding across months (enables pattern recognition), and lead with the key metric (don’t bury insights). Moreover, each recommendation included a concrete “before and after” description I could show clients.
Why It Works
Concrete examples build trust and enable immediate application. Meanwhile, theory alone requires you to figure out implementation—examples show you exactly how.
Technique 6: Iterate with Feedback
Don’t expect perfection on the first attempt. Instead, refine outputs conversationally.
Real Workflow from Data Visualization
After getting the data analysis for my e-commerce client:
Iteration 1:
Create Python visualization code for these insights:
1. Revenue trend by region (line chart)
2. Customer retention comparison (bar chart)
3. Top 10 products by revenue decline (horizontal bar)
Use matplotlib. Include proper labels and professional styling.
Iteration 2:
Good start. Make these adjustments:
- Use a color scheme suitable for executive presentations (blues and grays)
- Increase font sizes for readability in slideshow format
- Add data labels on the bar charts
- Export as high-resolution PNG files
The Difference
The first version was functional. However, the iteration made it presentation-ready. Total time: 5 minutes of prompting instead of 30 minutes of manual code editing.
Why It Works
Iteration mimics human collaboration. Furthermore, AI excels at refinement—use this to your advantage rather than demanding perfection immediately.
Technique 7: Use Negative Instructions
Explicitly state what to avoid to prevent off-target outputs. This is especially useful when AI tends to over-explain or use inappropriate tone.
Real Example from API Documentation
When writing technical explanations for business clients evaluating Claude API for their organization:
My Initial Prompt:
Explain Claude API pricing and implementation for a potential client.
Result: A technical deep-dive with code samples, version numbers, and competitor comparisons—overwhelming for non-technical decision-makers.
Improved Prompt with Negatives:
Explain Claude API pricing and implementation for a non-technical CFO evaluating AI tools.
Do NOT:
- Discuss competitor pricing (they'll research that separately)
- Include code samples (CFO is non-technical)
- Mention specific version numbers (they change frequently and add confusion)
- Use terms like "API endpoints," "tokens," or "latency"
- Exceed 200 words
DO:
- Compare costs to familiar subscription models (like software licenses)
- Focus on implementation timeline and support requirements
- Highlight cost predictability vs usage-based pricing models
- Explain what influences monthly costs in business terms
The Result
A clear, concise explanation the CFO could understand and present to their board without technical translation. Specifically, it focused on budget predictability and business value rather than technical specifications.
Why It Works
Negative constraints prevent common AI tendencies like over-explaining, using jargon, or veering off-topic. Additionally, they’re particularly effective when you know the pitfalls to avoid.
Technique 8: Leverage Reusable Templates
Standardize prompts for consistent quality across similar tasks.
My Standard Consulting Analysis Template
Act as a [ROLE with specific expertise].
Context: [Client situation, constraints, goals]
Task: [What you need done]
- Specific deliverable 1
- Specific deliverable 2
- Specific deliverable 3
Output format: [Structure you want]
Constraints:
- [Limit 1]
- [Limit 2]
- [Limit 3]
[Data or additional context]
Why It Works
Templates ensure you never miss critical context. Consequently, they create repeatable systems for high-quality outputs across projects.
Technique 9: Specify Output Format Explicitly
Format specifications prevent back-and-forth revisions. Moreover, they ensure you get exactly what you need in the structure you need it.
Real Example from Research Summarization
When I needed to quickly review 15 research papers for a client’s literature review on AI implementation in healthcare, consistency was critical.
Instead of:
Summarize this research paper.
Result: Each summary had different structure, length, and focus. I couldn’t compare papers effectively or extract consistent insights.
I used:
Summarize this research paper using this exact structure:
## Key Question
[One sentence stating the research question]
## Methodology
[2-3 sentences on approach and sample size]
## Main Findings
[Bullet points, 3-5 most significant findings]
## Implications for Healthcare AI Implementation
[One paragraph on practical applications]
## Limitations
[2-3 sentences on study constraints or gaps]
Maximum 300 words total. Use bullet points for findings only; write everything else in paragraph form.
The Benefit
Consistent summaries across all 15 papers that I could directly copy into the client report. Additionally, I could easily compare methodologies, identify research gaps, and extract implementation recommendations. What would have taken 2 full days of reading and note-taking took 4 hours.
Why It Works
Explicit formatting eliminates ambiguity. As a result, you get exactly what you need in the structure you need it, enabling faster comparison, analysis, and use of the outputs.
Technique 10: Treat AI as a Collaborative Partner
Use dialogue: draft, review, refine. Don’t demand final answers immediately.
My Typical Workflow
Step 1: “Draft a first version of this analysis. I’ll review and suggest changes.”
[Review output]
Step 2: “This is good. Now emphasize the financial implications and add a risk assessment section.”
[Review again]
Step 3: “Perfect. One final adjustment: make the executive summary more urgent in tone.”
Why It Works
AI excels at refinement over one-shot perfection. Therefore, collaborative iteration produces better results than demanding everything upfront.
The Core Mistake to Avoid
Expecting AI to read your mind.
AI amplifies your clarity—it never creates it. Vague inputs yield vague outputs. In contrast, specific, contextualized prompts yield specific, useful results.
Practical Application Guide
Start with these 3 techniques today:
- Add a role to your next prompt – “Act as a [specific expert]…”
- Provide context – Explain your situation, constraints, and goals
- Iterate don’t expect perfection – Refine outputs conversationally
Then expand to:
- Use constraints (word count, format, tone)
- Request specific examples
- Break complex tasks into numbered steps
For advanced users:
- Build reusable templates for common tasks
- Use negative instructions to prevent unwanted outputs
- Specify exact output formats
- Treat AI as a collaborative tool, not a magic button
Real Results from My Work
These techniques have enabled me to:
- Build production-ready web scraping systems at TIIQU in hours instead of days
- Deliver boardroom-ready analytics to consulting clients without extensive revision
- Create publication-quality content for platforms like My Thriving Child on first drafts
- Reduce time spent on technical documentation by approximately 60%
The difference isn’t the AI tools—it’s how you communicate with them.
Conclusion
Good prompting isn’t about tricking AI or finding magic formulas. Rather, it’s about clear communication: providing context, specifying requirements, and iterating toward better results.
Start with one technique from this article. Apply it to your next AI interaction. You’ll immediately see the difference between generic outputs and genuinely useful ones.
Ultimately, the people who thrive with AI aren’t those with the most sophisticated tools—they’re those who know how to communicate clearly about what they actually need.
Want to Go Deeper?
If you’re a business leader curious about implementing AI in your organization, I offer free 20-minute AI audits. We’ll discuss your specific challenges and identify practical opportunities for AI to add value.
Or email me directly: contact@artificialintelligence-tech.com
For more AI tool reviews, comparisons, and practical guides, visit artificialintelligence-tech.com
About the Author
I’m a Data Scientist (MSc, University of Hull) specializing in AI implementation and consulting. I help businesses cut through AI hype and implement practical solutions that deliver measurable results.
Leave a Reply