€20 focused prompt optimization · Client instruction work · Human-led review
AIAI ExpertsBook AI support
Home / Case Studies / Prompt Optimization
AI prompt & system instruction optimization case studies · €20

Two client prompt rewrites that turned vague instructions into clearer, testable systems.

These case studies are based on focused prompt and instruction optimization delivered for clients. They show what can fit into a compact engagement: one existing instruction system, a clear objective, focused restructuring, a small test set and practical handover notes.

Case Study A · Support prompt cleanup

Rewrite of an overgrown customer-support prompt with conflicting tone and policy rules

A client had one long support prompt that had accumulated changes over time. It contained competing tone rules, repeated policy instructions and weak behavior for missing information.

Service€20 · up to 40 minutes
InputOne existing support prompt
Main issueConflicting rules + weak fallback behavior
OutputRewritten prompt + mini test checklist

What the client brought

  • The current prompt/system instruction.
  • Examples of weak answers.
  • The intended tone and answer length.
  • Non-negotiable policy and escalation rules.

A→Z optimization

  1. Clarify objectiveDefined the exact job: draft support responses using approved information while asking for missing details and avoiding unsupported commitments.
  2. Find instruction conflictsMarked duplicate rules, contradictions, vague behavior requirements and policy text that needed separation from response behavior.
  3. Restructure hierarchyRebuilt the prompt into role, objective, allowed context, hard boundaries, clarification behavior, tone and output format.
  4. Add fallback behaviorSpecified what to do when information was missing, when a human was required and when policy did not support a definitive answer.
  5. Test representative inputsRan normal FAQ, incomplete request, complaint and unsupported policy scenarios to validate the new structure.
  6. HandoverDelivered the rewritten prompt plus concise notes explaining changes and what to retest after future edits.

Artifacts delivered

Clean rewritten support prompt
Role/objective block
Boundary + escalation block
Tone/output-format rules
Focused test scenarios
Next-step notes

Session complete when

  • The rewritten prompt had no obvious internal contradictions.
  • Missing information triggered clarification or escalation behavior.
  • Representative tests followed the intended tone and answer structure.
  • The client had a reusable final version and retest guidance.
Case Study B · Research prompt rewrite

Turn “research this topic and summarize it” into a structured evidence brief

A client used AI for recurring research but received generic prose, inconsistent formatting and weak separation between evidence, inference and missing information.

Service€20 · up to 40 minutes
InputOne recurring research prompt
Main issueVague task + inconsistent evidence handling
OutputStructured research prompt + examples

What the client brought

  • The current research prompt.
  • An example of the intended final brief.
  • Typical topic and audience inputs.
  • Source, recency or uncertainty requirements where relevant.

A→Z optimization

  1. Define the decision/outputClarified whether the research supported comparison, meeting preparation, content planning or another specific use.
  2. Separate inputs from reusable rulesCreated variables for topic, audience, geography, timeframe and research questions instead of rewriting the full prompt each time.
  3. Add evidence rulesDefined how to separate findings, uncertainty, assumptions and missing information, with source references where the platform supported them.
  4. Standardize the briefCreated a repeatable structure for executive summary, findings, evidence, uncertainty, implications and follow-up questions.
  5. Test across topicsRan straightforward and ambiguous research inputs to expose places where generic filler still appeared.
  6. Finalize reusable templateDelivered the final research prompt, example input block and a compact review checklist.

Artifacts delivered

Reusable research prompt
Input variable template
Evidence/uncertainty rules
Structured brief schema
Example test inputs
Review checklist

Session complete when

  • The prompt clearly separated task inputs from reusable instructions.
  • The output structure remained consistent across tested topics.
  • Uncertainty and missing evidence were surfaced rather than hidden.
  • The client could reuse the template without rebuilding it from scratch.
The €20 service covers one focused prompt or instruction framework. Complex multi-agent systems, large knowledge bases, integrations or extensive evaluation work require a larger scope.
Fixed-price focused service

Have one prompt that needs a professional cleanup?

Book the AI Prompt & System Instruction Optimization session for €20, up to 40 minutes.

View the €20 prompt service View pricing