AMERH: A Structured Framework for Writing Effective Prompts in Academia and Beyond

1. Introduction

GenAI tools such as Google Gemini and ChatGPT have sparked global attention since their release. For example, ChatGPT accumulated around 15.5 million subscribers within its first week, surging to 58 million in its second week (Carr, 2023). This increase highlights the importance of adopting GenAI tools, suggesting that higher education institutions must reassess the integration of these advancements within their educational curricula. According to (Lo, 2023), 89% of leaders ranked AI and GenAI as a priority for 2024. Similarly, (Hernholm, 2025) pointed out that companies are increasingly implementing AI to boost performance by up to 40% yet most students lack AI literacy upon graduation.

Generative AI (GenAI), a specific form of AI capable of generating multiple forms of media, including text, images, sounds, and video (Huang, Wang, & Zhang, 2024,Vartiainen & Tedre, 2024), has led to the creation of a diverse array of educational opportunities, from providing personalized feedback on writing to creating language learning partners and aiding in brainstorming creative projects. ChatGPT by OpenAI and Gemini by Google are examples of multimodal AI large language models (MLLMs) that illustrate how GenAI can engage in a natural, human-like conversation (Baidoo-Anu & Owusu Ansah, 2023,Pichai & Hassabis, 2023).

While GenAI presents a transformative potential for personalized learning experiences, concerns regarding its integration into the classroom environment necessitate exploration. Several challenges were introduced, including the potential for increased plagiarism (Cotton, Cotton, & Shipway, 2024,Meyer et al., 2023). Educators fear that students may become overly reliant on AI-generated content, leading to a decline in original thought and an increase in unattributed work. Furthermore, the digital divide poses another challenge of unequal access to reliable technology, which could exacerbate existing educational disparities (Ong et al., 2024,Wach et al., 2023). Therefore, equitable access to GenAI and the necessary infrastructure needs to be addressed to ensure inclusive learning environments.

Finally, other studies pointed out that overreliance on GenAI tools could hinder the development of critical thinking skills (Perera & Lankathilake, 2023,Spennemann, 2024). In other words, if students passively become accustomed to receiving information, their ability to independently evaluate information and form judgments may be compromised. As GenAI continues to expand, its role in shaping the future of higher education will likely be significant. It is expected that these new tools will become integral to the curricula, allowing students to engage with complex concepts through interactive experiences, while the human contributions will remain critical. The future classroom will continue to blend human expertise and technological innovation, empowering students to become self-directed learners and critical thinkers. Therefore, there is a need to examine the dynamic and immersive learning environments facilitated by GenAI. This paper introduces the AMERH framework, a structured and iterative approach to prompt writing, thereby enhancing students’ ability to interact with GenAI tools such as ChatGPT and Google Gemini. By focusing on the five principles—Ask, Museful, Evaluate, React, and Chain/Refine—students are not only equipped to generate more effective prompts but also to evaluate AI-generated responses critically.

2. Literature Review and Theoretical Framework

2.1. The Landscape of Prompt Engineering

The content generated by a GenAI model is strongly influenced by the user’s prompt, directly affecting its relevance and quality (Dinkevych, 2023). Prompt engineering is the practice of designing and enhancing inputs to help a GenAI model generate the most accurate and coherent results possible (Lo, 2023).

Several key techniques have been developed to enhance the performance of large language models (LLMs). These strategies include zero-shot prompting, where the model is given a task description without any examples; one-shot prompting, where a single example is provided; and few-shot prompting, which provides multiple examples to help the model recognize patterns (Brown et al., 2020). A more advanced technique, chain-of-thought prompting, enables the model to generate intermediate reasoning steps before arriving at a final answer, which significantly enhances performance on tasks requiring arithmetic or logical reasoning (Wei et al., 2022). These techniques represent the foundational technical approaches to interacting with LLMs.

2.2. Existing Pedagogical Frameworks and the Research Gap

As prompt engineering has become more critical, several pedagogical frameworks have emerged to help users, particularly learners, structure their prompts, such as CREATE (Birss, 2023), CLEAR (Lo, 2023), and CRISPE (Dinkevych, 2023), which offer structured approaches to ensure clarity and relevance in GenAI responses. These frameworks provide valuable heuristics for formulating initial queries. Despite the findings, a review of these frameworks uncovers a significant gap. These frameworks primarily emphasize how to create prompts, with less focus on the iterative process of how learners should interact with and critically evaluate the AI-generated responses. This gap highlights a need for a framework that positions learners as active participants in a dynamic dialogue with the AI models, empowering them to ask, refine, react to, and co-construct knowledge.

2.3. Theoretical Grounding: The Zone of Proximal Development (ZPD)

To address this gap, the AMERH framework, which is grounded in Lev Vygotsky’s sociocultural theory of learning, specifically the Zone of Proximal Development (ZPD), was introduced to help learners develop prompt writing skills. The ZPD describes the gap between what a learner can accomplish independently and what they can achieve with guidance and collaborative support (Vygotsky, 1978). Scaffolding involves providing temporary support that is gradually removed as the learner develops competence (Ness, 2023,Wood, Bruner, & Ross, 1976). Scaffolding in the context of prompt writing can be used to develop these skills. Instructors can model effective prompts, then gradually reduce support as students gain confidence and transition from guided to independent practice. This approach is particularly relevant because current AI models interactions are primarily one-way; the model responds but does not explicitly teach the user how to improve their prompts. This approach aligns with recent findings, such as those by (Song & Song, 2023), who demonstrated that AI-assisted writing instruction grounded in Vygotsky’s principles significantly improved the writing proficiency and motivation of English language learners. Therefore, a pedagogical framework for prompt writing should be designed as a scaffolding tool that supports learners within their ZPD, guiding them toward independent and critical use of GenAI.

3. The AMERH Framework: A Structured Approach to Prompt Writing

3.1. The Jigsaw Puzzle Analogy: A Conceptual Model

The process of designing an effective prompt can be compared to building a jigsaw puzzle. A puzzle consists of many unique, interlocking pieces that must be carefully fitted together to reveal a complete picture. A person solving the puzzle gathers pieces based on patterns or colors, pays close attention to detail, and exercises patience and persistence. Similarly, to generate a high-quality response from a GenAI model, a learner must thoughtfully select and arrange the elements of their prompt—the context, instructions, and constraints—paying close attention to the task’s details. A poorly constructed prompt element can lead to an irrelevant response, as a single lost puzzle piece can distort the final appearance. This analogy highlights the importance of a structured, iterative approach to help learners fit the pieces together correctly.

3.2. Redefining the Prompt and Prompt Engineering for a Multimodal Era

As discussed in Section 2.1, a prompt has traditionally been defined as an input or query given to an AI model to elicit a specific response (Dinkevych, 2023,Lo, 2023). However, with the rapid advancement of AI models capable of processing more than just words, this definition is becoming insufficient. To properly scaffold learning for today’s tools, I aim to provide a broader definition of a prompt and prompt engineering.

Therefore, a prompt is defined as a sequence of data that encompasses various modalities (text, images, audio, or video), allowing users to combine one or more modalities to form content and structure, and provide specific instructions or cues to elicit desired outputs from AI models.

Prompt engineering is a technique that enables users to incorporate texts, images, audio, or videos and select carefully structured phrases to elicit the most relevant and accurate responses from AI models, which are unique to each prompt. A well-crafted prompt in this multimodal context provides explicit directions to the AI model on what to generate and how to generate it, thereby enhancing the accuracy and relevance of the response to align with the user’s needs. As a result, there is a growing need for a structured approach to prompt writing. The next section outlines the core principles of the AMERH framework and provides various prompt structures and examples to help design effective prompts.

3.3. Introducing the AMERH Framework

The AMERH framework emerged from extensive experimentation with over 50 GenAI tools, including various versions of Gemini and ChatGPT, between 2023 and 2025, across diverse domains. It distinguishes itself from earlier frameworks (e.g., CREATE, CLEAR) in three main areas: First, it fosters a dynamic, iterative dialogue between the learner and the model, moving beyond a one-shot query. Second, it explicitly incorporates steps for evaluation and refinement, enabling learners to assess AI responses critically. Third, it is grounded in Vygotsky’s Zone of Proximal Development (ZPD), serving as a scaffold to help novices communicate effectively with GenAI models and gradually become self-directed, expert users.

The framework consists of five key principles, represented by the acronym AMERH: A (Ask), M (Museful), E (Evaluate), R (React), and H (Chain/Refine). Each principle guides the learner through a stage of the interaction, from initial query formulation to evaluation, refinement, and co-creation of knowledge. The following sections will detail each of these principles.

3.4. The Five Principles

3.4.1. Ask

This foundational principle involves the learner’s ability to pose the primary prompt or question. In this step, the learner sets the scope and direction for the response, which then serves as a starting point for formulating clear, precise, and effective prompts.

3.4.2. Museful

The Museful principle emphasizes the importance of deep, reflective thinking before writing a prompt. To guide this process, I have identified 16 key elements that can be incorporated into a prompt. These are broken down into two tiers: Essential and advanced components. This tiered approach provides a scaffolded method for prompt design.

3.4.2.1. The Essential Components

It is recommended that an effective prompt should be built upon seven essential components, encapsulated in the mnemonic PRACTIS, that establish its primary context and goal. These are: Profession, Restrictions, Audience, Concept, Task, Intent, and Situation. Mastering these seven components is the foundational step to crafting clear and effective prompts (see Appendix A).

3.4.2.2. The Advanced Components

Once the core prompt is established, it can be significantly enhanced by incorporating nine optional components. These components help to add nuance, specify format, and control the AI’s output with greater precision. These elements are encapsulated in the mnemonic SAFE ACTON: Scope, Amount, Format, Example, Accessibility, Communication, Tone, Other Options, and Nuance. I recommend that learners become comfortable with the essential components before integrating the advanced ones (see Appendix B).

3.4.3. Evaluate

This principle exemplifies a two-fold process of critical assessment: first, evaluating the user’s own prompt for inherent biases, and second, evaluating the AI-generated response for accuracy and objectivity.

In the first assessment, the learners should be able to evaluate the prompt before asking the model any questions. This approach will not steer the model toward a one-sided response. For example, asking "What are the proven benefits of Vaccine Pfizer over others?" presupposes that such benefits are established. Instead, learners should reframe the prompt and ask a balanced question. The second assessment emphasizes the learner’s ability to assess and verify the information received from the model critically. Students should actively cross-check facts using their existing knowledge or conventional research methods. It is crucial to identify potential misinformation or biased language, as AI models may reflect the political or social biases present in their training data. For instance, if a response presents a one-sided argument without acknowledging counterpoints, the student should recognize this as a form of bias and a signal that the initial prompt or the AI’s output needs further improvement.

3.4.4. React

This principle emphasizes the learner’s role in providing constructive feedback to the AI model, which can involve accepting a good response or rejecting a poor one with a clear justification. By engaging with the model’s output—especially by challenging its arguments or correcting inaccuracies—learners engage in a critical act of metacognition. This step is not about immediately refining the AI model itself, but about compelling students to articulate and solidify their own expert knowledge and moral judgment. Moral judgment, or "a decision about what one should do in a morally problematic situation" (Myyry, 2022, p. 332), is key to navigating the ethical complexities of AI-generated content. This process enables developers to use this data to fine-tune and train future versions of the model. This process is often part of what is called Reinforcement Learning from Human Feedback (RLHF). By analyzing the types of responses that yield "good responses" and "bad responses," developers can gradually steer the next generation of AI models to be more helpful, accurate, and safe.

3.4.5. Chain or Refine

This final principle involves the ability of learners to engage in two forms of iteration: refining a query (Refining) to improve a poor response or asking follow-up questions (Chaining) to explore a topic more deeply and address gaps in the model’s initial answer.

4. Classroom Implementation

Instructors can integrate the AMERH framework by scaffolding the learning process. Initially, they can model effective prompt writing and guide students through the five principles. As students grow more confident, they can assume greater responsibility for crafting, evaluating, and refining their prompts. To promote transparency and discourage over-reliance, educators can require students to submit their chatlogs, demonstrating how the GenAI tool was used as a partner in the learning process. The following Figure 1 illustrates the five principles.

5. Applying the Principles: Practical Examples

5.1. Ask

The following Table 1 demonstrates the application of the Ask principle.

5.2. Museful

The following Table 2 demonstrates the application of the Museful principle.

5.3. Evaluate

The following Table 3 demonstrates the application of the Evaluate principle.

5.4. React

The following Table 4 demonstrates the application of the React principle and illustrates how a user might react to model’s response based on their unique professional expertise.

Initial Prompt:

What is the best way to lose weight?

AI Model Response:

The best way to lose weight is to follow a calorie-deficient diet, where you consume fewer calories than you burn. This diet can be achieved by reducing calorie intake or increasing physical activity.

5.5. Chain/Refine

The following Table 5 demonstrates the application of the Chain/Refine principle and has two parts: Refining a flawed prompt and chaining follow-up prompt(s) to a good response.

6. Limitation

While the AMERH framework offers a robust and structured approach to prompt writing, it does have some limitations. The framework may be challenging for beginners due to its complexity. For example, the Museful principle, which is comprised of 16 different elements, can feel overwhelming for students with limited prior knowledge of AI models. It is therefore recommended to teach these elements in a step-by-step fashion. A gradual approach could be the right way to go, with students first mastering a few key elements, such as profession, concept, audience, and situation, before moving on to others, such as task, intent, and restrictions.

Additionally, this framework was developed and tested using the free versions of ChatGPT and Gemini. While the core principles should apply to other platforms, including paid versions and other models like Claude or Co-Pilot, further studies are recommended to examine the differences between free and paid versions.

7. Future Research

While this paper establishes the theoretical and practical foundation of the AMERH framework, it is crucial to test the framework with empirical studies. Future research should involve implementing the framework in various educational settings to assess its effectiveness. A proposed study could involve a pre-test/post-test design where the quality of students’ prompts is evaluated before and after receiving instruction on the AMERH principles. Additionally, qualitative data, such as student and instructor feedback, would be valuable for refining the pedagogical application of the framework.

8. Conclusion

The AMERH framework provides a structured and iterative approach to prompt writing, thereby enhancing students’ ability to interact with generative AI tools such as ChatGPT and Google Gemini. By focusing on the five principles—Ask, Museful, Evaluate, React, and Chain/Refine—students are not only equipped to generate more effective prompts but also to evaluate AI-generated responses critically. This approach fosters the development of essential skills, such as critical thinking, creativity, and ethical decision-making, preparing students for a future where these tools might become integral to both professional and personal life. Educators can implement the AMERH framework gradually, starting with guided instruction and moving toward independent use, mirroring the process of ZPD. As GenAI tools continue to evolve, educators must guide students to use these tools responsibly, ensuring they can navigate the ethical challenges of the digital age. Educators can encourage students to submit their chatlogs and drafts, demonstrating how the AI model was used in the process. This transparency promotes the ethical use of these tools and discourages over-reliance. GenAI tools, such as ChatGPT or Gemini, are similar to other technological innovations throughout history; they do not inherently possess harm. Instead, their influence relies entirely on how humans choose to use them.

We live in an era of AI expansion, and we witness cases of AI misuse that have caused harm to some victims. As educators, it is our responsibility to guide students in developing the ethical awareness and moral judgment necessary to navigate the age of AI responsibly, as well as to teach them how to interact effectively with these GenAI tools. Just as a reckless driver cannot blame the car for a deadly accident, we cannot blame GenAI tools for misuse caused by human decisions. The current advancements in AI call all educators to guide students in using GenAI tools responsibly and ethically, and help them develop the skills needed to navigate and benefit from these powerful tools.