Download to read offline
![A TEMPLATE for a self
reflecting system
GENERATE:
Create a [LinkedIn ad / Email / Landing page hero] about [product
for [target audience] in [country].
EVALUATE:
Evaluate the generated text based on:
• Clarity (does the ICP understand the offer?)
• Differentiation (does it stand out vs. competitors?)
• Persuasiveness (does it create desire to act?)
• CTA strength (is the next step clear?)
REFLECT:
What needs improvement? (2–3 sentences)
REGENERATE:
Rewrite it based on the reflection so that it becomes:
• More specific (backed by metrics)
• More persuasive (with social proof)
• With a clearer CTA](https://image.slidesharecdn.com/workshop-theartofpromptengineering-bestpracticesforeffectiveaiinteractionsbyaurimaspauliusgircys-251124181258-38e63310/75/The-Art-of-Prompt-Engineering-Best-Practices-for-Effective-AI-Interactions-By-Aurimas-Paulius-Gircys-21-2048.jpg)
Uploaded byAurimas Paulius Girčys
The Art of Prompt Engineering - Best Practices for Effective AI Interactions By Aurimas Paulius Gircys
Workshop: The Art of Prompt Engineering – Best Practices for Effective AI Interactions Presenter: Aurimas Paulius Girčys, APG Media Core Philosophy: "A good prompt is not a question to AI, but a mirror of your thinking." 1. The Challenge: Moving Beyond "Weak" Prompts Most users default to Zero-Shot prompting (e.g., "Write a blog post about cats"), characterized by a lack of context or reasoning. While useful for simple tasks, these "weak prompts" often produce generic, shallow, or hallucinated responses. To instantly upgrade basic queries, the workshop introduces the CIA Framework: Clarity: Define precise objectives to target responses. Include Best Practices: Ground the AI in professional standards (e.g., "Act as a senior strategist"). Action Orientation: Request immediate, executable steps. 2. Advanced Methodologies The workshop pivots from simple prompting to engineering AI as a reasoning engine using two primary techniques: A. Chain-of-Thought (CoT) Prompting Instead of asking for a final result immediately, CoT instructs the model to "think step by step." How it works: It forces the AI to break down complex problems (like budget calculations) into intermediate logical steps. Self-Consistency: For high-stakes tasks, generate 3–5 different reasoning paths and select the most frequent conclusion to filter out errors. B. Self-Reflecting Systems This technique creates a continuous improvement loop, significantly increasing output quality. Generate: AI creates the initial draft. Evaluate: AI assesses the draft against specific criteria (e.g., "Does this have a clear USP?"). Reflect: AI identifies specific weaknesses or missing elements. Regenerate: AI rewrites the content based on its own feedback. Result: Demonstrated a score increase from 46/50 to 50/50 for ad copy by automatically refining weak calls-to-action. 3. Implementation & Automation These strategies should be codified into automated workflows rather than manual typing: AI Assistants: Build custom system instructions that automatically apply CoT and Self-Reflection to every query. "Vibe Coding": Create lightweight apps (e.g., PromptCraft) that rewrite user input into professional prompts before execution. Scale via API: Use tools like Google Sheets to run the "Generate → Evaluate → Reflect" cycle on large datasets automatically. Conclusion Effective prompt engineering requires a shift in mindset: users must teach the AI how to think before asking what to produce. By adopting Self-Reflecting Systems and Chain-of-Thought reasoning, organizations can transform AI outputs from generic text into professionally structured, strategic assets. Key Tools: PromptCraft (Prompt rewriting), Waikay.io (AI Search Optimization), Jupitron.ai (Ad generation).
More Related Content
Similar to The Art of Prompt Engineering - Best Practices for Effective AI Interactions By Aurimas Paulius Gircys
The Art of Prompt Engineering - Best Practices for Effective AI Interactions By Aurimas Paulius Gircys
- 1. WORKSHOP: THE ARTOF PROMPT ENGINEERING: BEST PRACTICES FOR EFFECTIVE AI INTERACTIONS Aurimas Paulius Girčys, APG Media
- 2. A good promptis not a question to AI, but a mirror of your thinking.
- 3. Our agenda Zero shotprompts Chain-of-thought prompting Self-reflecting systems Building an AI assistant to implement prompting techniques in action Vibe-coding an app that both rewrites the prompts and executes them
- 4. Zero shot iswhat we usually do What is the average price per square meter for newly built apartments in Vilnius? Write me a 500-word blog post about cats.
- 5. Zero shot iswhat we usually do quick questions Brainstorming exploring unknown domains simple tasks explanations summaries translations idea generation
- 6. I call them: “WEAKPROMPTS”
- 7. CIA in action CLARITYand precision: Create queries with clear, precise tasks and objectives so that responses are well- targeted. INCLUDE best practices: Incorporate references to best practices in your field so that responses are grounded in professional standards. ACTION orientation: Include actionable tasks so that AI can take immediate steps and make advice more practical.
- 8. A good Chainof Thought prompt doesn’t just demand an answer, it teaches the AI how to think it through.
- 9. Chain of Thought prompting Insteadof simply asking for the final answer, you instruct the model to “think step by step” or provide an example that includes the reasoning process. This encourages the model to break down complex problems into intermediate steps, which leads to much more accurate results.
- 10. Chain of Thoughtprompting Question: I’m planning a Google Ads + LinkedIn ABM campaign for the German market. Budget: €20,000 over 3 months. Product: B2B SaaS (ACV €15k). Goal: 30 MQL. Distribute the budget across channels and calculate CPL using the CoT methodology. THE AI RESPONSE SHOULD BE: Step 1: Benchmark CPL in Germany Google Search: €80–150 per MQL LinkedIn ABM: €120–250 per MQL Step 2: Budget split (60/40 Google/LinkedIn) Google: €12,000 LinkedIn: €8,000 Step 3: Expected number of MQLs Google: 12,000 / 120 ~100 MQL (with optimization) ≈ LinkedIn: 8,000 / 180 ~44 MQL ≈ Conclusion: With this budget, you can achieve 30–35 MQL over 3 months.
- 11. I prefer COTbecause it doesn’t just spit out the answer. But how do you know if the provided answer is correct?
- 12. The solution couldbe: 1. Self consistent C-O-T prompts 2. Self reflecting systems
- 13. Self Consistency Chainof Thought prompting This system extends CoT: the model generates SEVERAL DIFFERENT reasoning paths for the same problem, then selects the answer that appears most frequently. This makes the result MORE RELIABLE. 📋 CONCEPT: Generate 3–5 reasoning paths choose → the most frequently recurring answer. ✅ BEST USED FOR: • Critical CAC/LTV calculations • Market sizing forecasts • ROI scenario modeling • Pricing strategy development 🔧 HOW IT WORKS: You run the same CoT prompt 3–5 times. If 4 out of 5 answers suggest “€120 CPL” and one says “€95 CPL,” you select €120 as the most reliable estimate.
- 14. SELF-reflecting system 🔄 Thissystem allows AI to SELF-EVALUATE and IMPROVE the content it has generated. Process: Generate – Create the initial version 1 ️ 1️⃣ Evaluate – Assess it based on criteria 2️⃣ Reflect – Describe what needs improvement 3 ️ 3️⃣ Regenerate – Produce an improved version 4️⃣ 📋 CONCEPT: AI learns from its own mistakes and improves the content. ✅ BEST USED FOR: • Marketing copy optimization • Email sequence improvement • Refining landing page text • Pitch deck content enhancement
- 15. A self-reflecting system doesn’tjust generate answers — it holds up a mirror, showing you the clarity and gaps in your own
- 16. SELF-reflecting system example 🎯TASK: “Create Google Search ad copy for CNC machine export to Germany” GENERATE (Initial): 1 ️ 1️⃣ H: “CNC machines for industry” D: “High-quality machines. Contact us.” EVALUATE: 2️⃣ ❌ Too generic ❌ No USP ❌ Weak CTA REFLECT: 3 ️ 3️⃣ “Need to: • Add CE/ISO mention • Specify delivery time • Strengthen CTA with ‘demo’ or ‘quote’” REGENERATE (Improved): 4️⃣ H: “CE-Certified CNC Machines | 4-Week Delivery” D: “Production equipment with IoT integration. ISO 9001. Free consultation. Book a demo ” →
- 17.
- 18. Step-by-step explanation ofself-reflecting sytem
- 19. Self reflecting systemin practice X
- 20. Another example ofSelf reflecting system Generate: Write ad copy for a Facebook post about coffee machine rentals for businesses. Evaluate: Assess the generated text based on the following criteria: clarity, persuasiveness, target audience fit, and call-to-action (CTA). Reflect: Based on the evaluation, describe what needs to be improved in the text to make it more effective in an advertising context. Regenerate: Rewrite the original ad using the suggestions from the reflection step so that the result is higher quality and more impactful.
- 21. A TEMPLATE fora self reflecting system GENERATE: Create a [LinkedIn ad / Email / Landing page hero] about [product for [target audience] in [country]. EVALUATE: Evaluate the generated text based on: • Clarity (does the ICP understand the offer?) • Differentiation (does it stand out vs. competitors?) • Persuasiveness (does it create desire to act?) • CTA strength (is the next step clear?) REFLECT: What needs improvement? (2–3 sentences) REGENERATE: Rewrite it based on the reflection so that it becomes: • More specific (backed by metrics) • More persuasive (with social proof) • With a clearer CTA
- 22. What‘s next? Zero shotprompts Chain-of-thought prompting Self-reflecting systems Building an AI assistant to implement prompting techniques in action Vibe-coding an app that both rewrites the prompts and executes them
- 23. Building an AIAssistant
- 24. How did webuild it?
- 25. Vibe code aprompt improver https://prompt-engineering-architect-aurimas4.replit.app/
- 26. Vibe code aprompt improver https://prompt-engineering-architect-aurimas4.replit.app/
- 27. Improving prompts overAPI https://docs.google.com/spreadsheets/d/1C5JdUpnsOsB- bph3T4OHxp3kJQINPlQDKYjD6LC9S7M/edit?usp=sharing
- 28. Improving prompts overAPI https://docs.google.com/spreadsheets/d/1C5JdUpnsOsB- bph3T4OHxp3kJQINPlQDKYjD6LC9S7M/edit?usp=sharing
- 29.
- 30.
- 31.