Educational and Tutorial Content in Prompt Engineering

Educational and tutorial content for prompt engineering comprises structured materials—guides, curricula, examples, and exercises—designed to teach people how to systematically design and refine prompts for large language models and related generative systems ⁴⁷⁹. Its primary purpose is to translate rapidly evolving research and best practices into repeatable, learnable workflows that non-experts and experts alike can apply in real tasks ³⁸. This content matters because prompt quality strongly determines model performance, safety, and reliability, especially when LLMs are used in education, communication, and decision support ³⁴⁷. As generative AI becomes embedded across domains, robust educational resources on prompt engineering are a key mechanism for scaling safe and effective use.

Overview

The emergence of educational and tutorial content in prompt engineering is closely tied to the rapid development and deployment of instruction-tuned large language models, particularly following the release of systems like ChatGPT ⁴. As these models demonstrated remarkable sensitivity to how instructions are phrased—where small changes in wording could dramatically alter output quality—the need for systematic training became apparent ³⁷. Early adopters in education, enterprise, and research quickly discovered that effective use of LLMs required more than casual experimentation; it demanded a disciplined approach to crafting, testing, and refining prompts ¹².

The fundamental challenge that educational content addresses is the gap between LLM capability and user ability to elicit that capability reliably. Unlike traditional software with deterministic APIs, LLMs respond to natural-language instructions in probabilistic ways, exhibiting both impressive flexibility and frustrating inconsistency ³⁴. Users must learn to think of prompts as “soft programs”—natural-language specifications that shape model behavior through linguistic cues, context, constraints, and examples ⁵⁷. Without structured guidance, users often produce vague, ambiguous prompts that yield irrelevant, biased, or unsafe outputs, undermining trust and limiting adoption ²⁹.

Over time, the practice has evolved from ad-hoc tips shared in forums to comprehensive frameworks grounded in empirical research and instructional design principles ¹³⁶. Modern educational materials incorporate findings from prompt-engineering research—such as chain-of-thought reasoning, few-shot learning, and role-based prompting—and package them into scaffolded learning experiences with worked examples, exercises, and reflection activities ⁴⁷. This evolution reflects a maturation from “prompt hacking” toward prompt engineering as a teachable, transferable skill set that bridges human-computer interaction, software specification, and domain expertise ³⁵⁸.

Key Concepts

Instruction Clarity and Specificity

Instruction clarity refers to the precision and explicitness with which a prompt communicates the desired task, format, constraints, and evaluation criteria to the model ³⁷⁹. Clear prompts reduce ambiguity, helping the model generate outputs that align closely with user intent. Educational materials emphasize that vague requests like “Tell me about photosynthesis” often yield generic responses, whereas specific instructions—”Explain photosynthesis to a 7th-grade student in 200 words, using an analogy to a factory”—produce targeted, appropriate content ¹⁶.

Example: A high school biology teacher wants to generate a quiz on cellular respiration. An unclear prompt might be: “Make a quiz about cells.” A clear, specific prompt would be: “Create a 10-question multiple-choice quiz on cellular respiration for 10th-grade biology students. Include questions on glycolysis, the Krebs cycle, and the electron transport chain. Provide four answer choices per question, with one correct answer and three plausible distractors. Include an answer key with brief explanations.” This specificity ensures the output matches the curriculum scope, difficulty level, and format needed for classroom use ²⁶.

Context Provisioning

Context provisioning involves supplying the model with relevant background information, constraints, audience characteristics, and situational details that condition its response ³⁷. Because LLMs generate text based on the context window—the token budget available for instructions, examples, and conversation history—effective prompts front-load essential context to guide the model’s reasoning and tone ³⁴. Educational content teaches learners to identify what context is necessary (grade level, prior knowledge, cultural considerations) and how to encode it efficiently ²⁶.

Example: A university instructor designing an AI-assisted writing tutor for international students needs to provide context about the audience. Instead of prompting “Give feedback on this essay,” the instructor writes: “You are an academic writing tutor for undergraduate international students whose first language is not English. Review the following essay introduction for clarity, thesis strength, and organization. Provide constructive feedback in simple, encouraging language. Highlight one strength and suggest one specific revision.” This context ensures the feedback is appropriately pitched, supportive, and actionable for the target learners ⁶⁷.

Role Specification

Role specification is the technique of assigning the model a persona, expertise level, or functional role within the prompt to shape its tone, depth, and perspective ³⁴⁷. Research and practice show that prompts beginning with phrases like “You are an expert statistician” or “You are a patient elementary school teacher” can significantly improve response relevance and style ⁴⁹. Educational materials present role prompting as a powerful lever for aligning outputs with domain norms and user expectations ¹⁷.

Example: A corporate trainer developing onboarding materials for new software engineers might prompt: “You are a senior software engineer mentoring a junior developer on their first day. Explain the concept of version control using Git in a friendly, non-technical way. Use a real-world analogy and provide three practical tips for getting started.” The role cue (“senior engineer mentoring a junior”) primes the model to adopt an accessible, supportive tone and to structure the explanation pedagogically, rather than delivering a technical manual ³⁷.

Few-Shot and Zero-Shot Prompting

Few-shot prompting involves including one or more labeled examples within the prompt to demonstrate the desired input-output pattern, leveraging the model’s in-context learning ability ⁴⁷. Zero-shot prompting, by contrast, provides only instructions without examples, relying on the model’s pre-trained and instruction-tuned knowledge ⁴. Educational content teaches when to use each approach: zero-shot for straightforward tasks where instructions suffice, and few-shot when the desired format, style, or reasoning pattern is complex or non-standard ³⁹.

Example: A curriculum developer wants the model to generate vocabulary flashcards in a specific format. A zero-shot prompt might be: “Create five vocabulary flashcards for 8th-grade English, each with a word, definition, and example sentence.” A few-shot prompt would include an exemplar:

Create vocabulary flashcards in this format:

Word: Ubiquitous
Definition: Present, appearing, or found everywhere.
Example: Smartphones have become ubiquitous in modern society.

Now create five flashcards for 8th-grade English on the theme of persuasive writing.

The example clarifies the exact structure and style, reducing variability and ensuring consistency across generated flashcards ⁴⁶⁷.

Chain-of-Thought Prompting

Chain-of-thought prompting is a technique that instructs the model to articulate its reasoning process step-by-step before arriving at a final answer ⁴. Research demonstrates that prompts like “Think step by step” or “Show your work” significantly improve performance on mathematical, logical, and multi-step reasoning tasks ⁴. Educational materials highlight this method for generating worked examples, explanations, and diagnostic feedback, where transparency of reasoning is pedagogically valuable ¹³⁶.

Example: A math teacher wants the model to generate a worked solution for a quadratic equation problem. Instead of prompting “Solve x² – 5x + 6 = 0,” the teacher writes: “Solve the equation x² – 5x + 6 = 0. Show each step of your reasoning: identify the method (factoring, quadratic formula, or completing the square), perform the calculations, and check your solution.” The model then produces a step-by-step solution that students can follow, making the reasoning process explicit and supporting learning ⁴⁶.

Iterative Refinement and Debugging

Iterative refinement is the practice of treating prompt design as a cyclical process: draft a prompt, evaluate the output, diagnose issues, and revise the prompt accordingly ¹³⁷. Educational content frames this as analogous to software debugging, where failures (ambiguous outputs, factual errors, inappropriate tone) are systematically analyzed and addressed through prompt adjustments ¹⁹. This concept emphasizes that effective prompting is rarely achieved on the first attempt and that learning to iterate is a core skill ³⁵.

Example: A professional development facilitator creates a prompt to generate discussion questions for a workshop on equity in education. The initial prompt—”Generate discussion questions about equity”—yields overly broad, generic questions. The facilitator diagnoses the issue (lack of specificity and context) and revises: “Generate five discussion questions for K-12 educators on equity in classroom assessment practices. Each question should prompt reflection on personal practice, reference a specific challenge (e.g., bias in grading, accessibility), and encourage sharing of strategies.” The revised prompt produces focused, actionable questions. The facilitator documents both versions and the rationale, building a reusable template for future workshops ¹³⁶.

Safety and Bias Mitigation

Safety and bias mitigation in prompt engineering involves designing prompts that minimize the risk of harmful, biased, or inappropriate outputs ²⁴⁹. Educational materials teach users to include explicit constraints (e.g., “Avoid stereotypes,” “Use inclusive language”), to review outputs critically, and to understand that models can reproduce biases present in training data ²⁷. This concept is especially critical in educational and public-facing applications, where outputs directly affect learners and stakeholders ²⁵.

Example: A school district develops an AI tool to generate personalized reading recommendations for students. To mitigate bias, the prompt includes: “Recommend five age-appropriate books for a 6th-grade student interested in science fiction. Ensure the list includes diverse authors (gender, race, cultural background) and protagonists. Avoid reinforcing stereotypes about scientists or explorers.” The district also implements a review process where librarians check recommendations before sharing them with students, combining prompt-level safeguards with human oversight ²⁶⁹.

Applications in Educational Contexts

Educational and tutorial content in prompt engineering finds application across multiple phases of teaching, learning, and instructional design, each with distinct goals and techniques.

Lesson Planning and Curriculum Development: Educators use prompt engineering to generate lesson plans, learning objectives, activity sequences, and differentiated materials ¹²⁶. For example, a middle school science teacher might prompt: “Design a 45-minute lesson plan on the water cycle for 6th graders, including a 10-minute demonstration, a hands-on activity, and formative assessment questions. Align with Next Generation Science Standards MS-ESS2-4.” The resulting plan provides a structured starting point that the teacher refines based on classroom context, saving preparation time while maintaining pedagogical quality ¹⁶.

Formative Assessment and Feedback: Prompt engineering enables the creation of quizzes, practice problems, and automated feedback on student work ²⁶⁷. A university writing instructor might use a prompt like: “Review this student essay introduction. Identify the thesis statement, assess its clarity and specificity, and suggest one concrete revision to strengthen the argument. Use encouraging, constructive language appropriate for a first-year undergraduate.” This application supports timely, personalized feedback at scale, though educators are trained to review AI-generated feedback for accuracy and appropriateness before sharing it with students ²⁶.

Interactive Tutoring and Socratic Dialogue: Advanced applications involve prompts that configure the model as an interactive tutor, asking guiding questions rather than providing direct answers ³⁶⁷. For instance: “You are a Socratic tutor helping a high school student understand the concept of slope in algebra. The student has just asked, ‘What is slope?’ Respond with a guiding question that prompts them to think about how steepness relates to rise and run. Do not give the definition directly.” This approach supports inquiry-based learning and metacognition, aligning with constructivist pedagogical principles ⁶⁷.

Professional Development and Educator Training: Educational institutions and technology providers offer workshops, online courses, and certification programs that teach educators prompt-engineering skills ¹⁵⁸. These programs typically include hands-on exercises where participants design prompts for their own teaching contexts, peer review each other’s work, and build shared prompt libraries. For example, a district-wide training might have teachers collaboratively develop prompt templates for generating rubrics, parent communication letters, and individualized education plan (IEP) goal statements, fostering both skill development and community of practice ¹²⁵.

Best Practices

Use Explicit Structure and Formatting Constraints

Educational materials consistently recommend that prompts include explicit structural cues—numbered steps, bullet points, section headings, word limits, and formatting requirements ³⁷⁹. The rationale is that LLMs are highly responsive to structural signals, and clear formatting constraints reduce ambiguity and improve output consistency. Structured prompts also make it easier to evaluate whether the model has followed instructions.

Implementation Example: A curriculum coordinator generating unit assessments writes: “Create a unit test on the American Civil War for 11th-grade U.S. History. Structure: (1) 10 multiple-choice questions, (2) 3 short-answer questions (2-3 sentences each), (3) 1 essay prompt. For multiple-choice, provide four options labeled A-D. For the essay prompt, include a rubric with criteria for thesis, evidence, and analysis. Limit total length to 1,200 words.” This explicit structure ensures the output matches the required format and can be directly adapted for classroom use ³⁷⁹.

Incorporate Audience and Context Information

Effective prompts specify the target audience’s characteristics—age, prior knowledge, language proficiency, and learning goals—so the model can tailor content appropriately ²⁶⁷. The rationale is that generic outputs often miss the mark on difficulty, vocabulary, or cultural relevance. By front-loading audience context, users increase the likelihood of generating materials that are pedagogically sound and accessible.

Implementation Example: An ESL instructor designing reading comprehension exercises prompts: “Create a reading passage about renewable energy for adult English language learners at CEFR B1 level (intermediate). Use vocabulary appropriate for learners with 3-5 years of English study. Include a glossary of five key terms. Follow the passage with five comprehension questions: three factual recall and two inferential.” The audience specification (adult, B1 level, vocabulary range) ensures the passage is neither too simple nor too complex, and the glossary supports vocabulary development ²⁶⁷.

Build Evaluation and Verification Loops

Best practices emphasize that AI-generated content must be reviewed, fact-checked, and validated by domain experts before use in high-stakes settings ²⁶⁹. The rationale is that LLMs can produce plausible but incorrect information (hallucinations), exhibit biases, or generate content misaligned with institutional policies. Educational materials teach users to cross-reference outputs against trusted sources, pilot materials with small groups, and document quality-assurance steps.

Implementation Example: A school district implementing AI-assisted quiz generation establishes a review protocol: (1) Teachers generate quizzes using standardized prompts, (2) Department heads verify factual accuracy against textbooks and standards documents, (3) Quizzes are piloted with a small group of students, (4) Feedback is used to refine prompts and outputs. The district maintains a shared repository of vetted prompts and quizzes, with version notes documenting changes and rationales. This loop ensures quality and builds institutional knowledge over time ²⁶⁹.

Document and Share Effective Prompts

Educational content encourages users to log successful prompts, rationales, and performance notes in shared libraries or knowledge bases ³⁹. The rationale is that prompt engineering is context-dependent and iterative; documenting what works accelerates learning, supports collaboration, and enables reuse. Shared libraries also help organizations establish standards and reduce redundant effort.

Implementation Example: A university teaching center creates a prompt library in its learning management system, organized by discipline and task type (e.g., “Biology > Lab Report Feedback,” “History > Primary Source Analysis”). Each entry includes the prompt text, intended use case, example output, and notes on limitations or required follow-up. Faculty contribute prompts and rate their effectiveness, creating a living resource that evolves with practice. New instructors onboarding to AI tools start with vetted templates, reducing the learning curve ³⁹.

Implementation Considerations

Tool and Format Choices

Organizations must decide which platforms, interfaces, and delivery formats best suit their educational goals and technical capacity ³⁷. Options range from web-based chat interfaces (e.g., ChatGPT, Claude) to API integrations within learning management systems, custom applications, or prompt playgrounds for experimentation ³⁷. The choice depends on factors such as user technical skill, need for customization, data privacy requirements, and budget. For example, a K-12 district prioritizing student data privacy might opt for an on-premises or education-specific LLM deployment with strict access controls, while a university research group might use API access for flexibility and integration with existing workflows ³⁷⁹.

Specific Example: A community college implements prompt-engineering training for faculty using a tiered approach. Beginners use a simplified web interface with pre-loaded prompt templates for common tasks (syllabus generation, discussion prompts). Intermediate users access a prompt playground where they can experiment with parameters like temperature and token limits. Advanced users receive API keys and documentation for integrating LLM calls into custom scripts or learning analytics dashboards. This tiered structure accommodates diverse technical comfort levels and use cases ³⁷.

Audience-Specific Customization

Educational content must be tailored to the prior knowledge, roles, and goals of different audiences—K-12 teachers, higher education faculty, instructional designers, students, or administrators ¹²⁶. Customization includes adjusting terminology (avoiding jargon for novices, using precise technical language for experts), selecting relevant examples from the audience’s domain, and aligning exercises with authentic tasks. For instance, training for elementary teachers emphasizes prompts for differentiated reading materials and social-emotional learning scenarios, while training for computer science faculty focuses on code generation, debugging assistance, and autograding ¹⁶⁸.

Specific Example: A state education agency develops three versions of a prompt-engineering workshop. The K-5 version uses examples like generating read-aloud questions and behavior management scripts, with hands-on activities creating morning meeting prompts. The 6-12 version covers quiz generation, essay feedback, and project-based learning scaffolds, with exercises designing prompts for argumentative writing rubrics. The administrator version focuses on communication templates (parent letters, policy summaries) and data analysis prompts for enrollment or achievement trends. Each version uses role-appropriate language and examples, increasing relevance and engagement ¹²⁶.

Organizational Maturity and Governance

Implementation success depends on organizational readiness—existing policies on AI use, data governance, professional development infrastructure, and cultural attitudes toward technology ²⁵⁷. Mature organizations establish clear guidelines on acceptable use, data privacy, intellectual property, and quality assurance before scaling prompt-engineering training ²⁹. They also invest in ongoing support, such as help desks, communities of practice, and regular updates as models and best practices evolve ¹⁵.

Specific Example: A large urban school district adopts a phased rollout. Phase 1 (pilot year): A volunteer cohort of 20 teachers receives intensive training, co-designs prompt templates, and documents lessons learned. The district drafts an AI use policy covering student data, attribution, and prohibited uses (e.g., no AI-generated grades without human review). Phase 2 (expansion): Based on pilot feedback, the district refines training materials, establishes a prompt library, and trains 200 additional teachers. An AI task force meets quarterly to review incidents, update guidelines, and share innovations. Phase 3 (scale): Training becomes part of standard onboarding for all new teachers, with annual refreshers on emerging techniques and safety practices. This staged approach builds capacity, governance, and cultural buy-in ²⁵⁷.

Integration with Existing Workflows

Effective implementation embeds prompt engineering into existing instructional design, curriculum development, and assessment processes rather than treating it as a standalone tool ¹³⁶. This requires mapping where AI assistance adds value (e.g., generating first drafts, brainstorming ideas, differentiating materials) and where human expertise remains essential (e.g., final content approval, relationship-building, ethical judgment). Training materials should demonstrate integration points and provide workflow templates ³⁶⁹.

Specific Example: A university instructional design team integrates prompt engineering into its course development process. When faculty request a new online course, the intake form now includes a section on AI-assisted content generation. Instructional designers use standardized prompts to generate draft learning objectives, discussion questions, and quiz items based on faculty input. These drafts are reviewed in collaborative meetings where faculty refine content for accuracy and alignment with course goals. The final syllabus credits AI assistance in the acknowledgments section, modeling transparency. This integration reduces development time by approximately 30% while maintaining quality standards ³⁶⁹.

Common Challenges and Solutions

Challenge: Overreliance and Academic Integrity Concerns

Students and educators may use AI to bypass learning processes—students submitting AI-generated essays without engagement, or teachers relying on AI-generated lesson plans without critical review ²⁶. This overreliance undermines learning objectives, raises academic integrity issues, and can perpetuate errors or biases present in AI outputs. The challenge is especially acute in educational settings where the goal is skill development, not just task completion ²⁵.

Solution:

Educational materials should explicitly frame AI as a collaborative tool that supports, rather than replaces, human thinking and creativity ²⁶. Training includes exercises that require learners to critique, revise, and extend AI outputs, making engagement with content unavoidable. For example, a writing assignment might require students to generate an essay outline with AI, then write a reflection analyzing the outline’s strengths and weaknesses, and finally compose the essay incorporating their critique. Rubrics assess both the final product and the reflection, rewarding thoughtful use of AI ²⁶. Institutions also establish clear policies defining acceptable AI use for different assignment types, with examples and rationales, and educate students on why these boundaries support learning ²⁵.

Challenge: Inconsistent and Unpredictable Outputs

LLMs exhibit stochastic behavior—running the same prompt multiple times can yield different results, and small phrasing changes can dramatically alter outputs ³⁷. This inconsistency frustrates users expecting deterministic software behavior and complicates quality assurance, especially when generating materials for high-stakes use like assessments or student feedback ³⁹.

Solution:

Educational content should normalize variability as an inherent characteristic of LLMs and teach strategies for managing it ³⁷. These include: (1) running prompts multiple times and selecting the best output, (2) using temperature and sampling parameters to control randomness when API access allows, (3) designing prompts with strong constraints and examples to narrow the output space, and (4) building review and revision steps into workflows ³⁷⁹. For example, a teacher generating quiz questions might prompt the model to produce 15 questions, then manually select and edit the best 10, treating the AI as a brainstorming partner rather than a final authority. Training exercises can include “prompt robustness testing,” where learners run the same prompt five times, compare outputs, and identify which prompt elements reduce variability ³⁷.

Challenge: Bias and Harmful Content Generation

LLMs can generate biased, stereotypical, or culturally insensitive content, reflecting patterns in their training data ²⁴⁹. In educational contexts, such outputs can harm students, reinforce inequities, and violate institutional values. Users without training in bias recognition may not notice problematic content, especially when outputs are fluent and plausible ²⁵.

Solution:

Educational materials must include explicit modules on bias, with concrete examples of how prompts can inadvertently elicit stereotypes and how to design prompts that promote inclusivity ²⁹. Training teaches users to include constraints like “Use diverse examples representing multiple cultures, genders, and abilities” and “Avoid stereotypes about [specific group]” ². Equally important is teaching critical evaluation: exercises where learners review AI-generated content for bias, discuss findings in groups, and revise prompts to mitigate issues ²⁵. Institutions should also establish review protocols requiring diverse reviewers to check materials before student use, and maintain reporting channels for problematic outputs. For example, a district might require that any AI-generated social studies content be reviewed by a committee including educators from underrepresented backgrounds, with a checklist covering representation, stereotypes, and historical accuracy ²⁹.

Challenge: Hallucinations and Factual Errors

LLMs can confidently generate false information—fabricated citations, incorrect facts, or nonsensical reasoning—that appears plausible ³⁴⁹. In educational settings, such errors can misinform students, undermine trust, and create legal or reputational risks if, for example, a teacher distributes a quiz with incorrect answers ²⁶⁹.

Solution:

Educational content must emphasize verification as a non-negotiable step in any workflow involving factual claims ²⁶⁹. Training includes exercises in fact-checking AI outputs against authoritative sources (textbooks, peer-reviewed articles, official standards documents) and teaches users to recognize hallucination patterns (e.g., overly specific but unverifiable claims, fabricated references) ³⁹. Prompts can be designed to reduce hallucinations by instructing the model to indicate uncertainty (“If you are not certain, say ‘I don’t know'”) or to cite sources when making factual claims, though users must still verify those sources exist and support the claims ³⁴. For high-stakes content like assessments, institutions implement multi-step review: the prompt designer, a subject-matter expert, and ideally a third party all check for accuracy before materials are used with students. Documentation of this review process protects both educators and institutions ²⁶⁹.

Challenge: Lack of Domain and Pedagogical Expertise

Effective prompt engineering in education requires both technical skill (crafting clear prompts) and domain expertise (knowing what good content looks like) ²⁶. Users without subject-matter knowledge may not recognize when AI outputs are incorrect, inappropriate for the grade level, or misaligned with learning objectives. Similarly, users without pedagogical training may generate technically accurate content that fails to support learning ¹⁶.

Solution:

Educational programs should integrate prompt-engineering training with domain-specific and pedagogical professional development ¹²⁶. For example, a workshop on using AI for math instruction would include both prompt techniques (few-shot examples, chain-of-thought) and math pedagogy (common misconceptions, problem-solving strategies, standards alignment). Participants practice generating math problems, then critique them using pedagogical criteria (Does this problem build on prior knowledge? Does it address a common error? Is the difficulty appropriate?) ¹⁶. Institutions can also foster collaboration between prompt-engineering experts and domain specialists, creating co-design teams that combine technical and content expertise. Prompt libraries should include not just the prompts but also pedagogical rationales and alignment notes, helping users understand why a prompt works and when to adapt it ¹²⁶.

See Also

• Prompt Design Patterns and Techniques
• Few-Shot and Zero-Shot Learning in Prompt Engineering

References

1. National Science Teaching Association. (2024). The Art and Science of Prompt Engineering. https://www.nsta.org/blog/art-and-science-prompt-engineering
2. Flint K12. (2024). AI Glossary: Prompt Engineering. https://flintk12.com/ai-glossary/prompt-engineering
3. Amazon Web Services. (2024). What is Prompt Engineering? https://aws.amazon.com/what-is/prompt-engineering/
4. Wikipedia. (2024). Prompt Engineering. https://en.wikipedia.org/wiki/Prompt_engineering
5. Georgia Institute of Technology, Ivan Allen College. (2024). AI Prompt Engineering with ChatGPT. https://iac.gatech.edu/featured-news/2024/02/AI-prompt-engineering-ChatGPT
6. Ohio State University, Center for Design and Learning Innovation. (2024). Prompt Engineering. https://cdli.ehe.osu.edu/ai-in-ehe/ai-101/prompt-engineering/
7. Stanford University Information Technology. (2024). AI Demystified: Prompt Engineering. https://uit.stanford.edu/service/techtraining/ai-demystified/prompt-engineering
8. Coursera. (2024). What Is Prompt Engineering? https://www.coursera.org/articles/what-is-prompt-engineering
9. IBM. (2024). What is Prompt Engineering? https://www.ibm.com/think/topics/prompt-engineering