Prompt Engineering for Language Teachers Using ChatGPT: A Brief Guide to Techniques, Applications, and Effective Practices

1. Introduction

Generative artificial intelligence (GenAI) – a subset of AI systems capable of producing novel content such as text, images, explanations, feedback, and learning materials – has moved from novelty to classroom reality. In foreign language settings, learners increasingly use GenAI tools to translate ideas, generate drafts, correct grammar, summarize readings, rehearse presentations, and prepare for speaking tasks (Vũ & Hà, 2026; Zaim et al., 2025). Teachers use the same technologies to create lesson plans, adapt texts, develop quizzes, produce model dialogues, generate rubrics, and provide feedback. This rapid adoption creates both opportunity and risk. More specifically, GenAI can expand access to language input and individualized support, but it can also weaken the learning processes through which languages are acquired: effortful production, negotiation of meaning, noticing, feedback uptake, revision, and reflection (Yan et al., 2024).

The central problem is that GenAI tools can generate language without guaranteeing language learning. A learner who submits an AI-generated essay may appear more proficient than they are. A teacher who accepts AI-generated feedback without review may unintentionally provide inaccurate, culturally inappropriate, or developmentally unsuitable guidance. A classroom that uses GenAI primarily to produce answers may reduce opportunities for productive struggle. Therefore, the important question is not whether foreign language teachers should use GenAI, but how they can design AI-supported learning in ways that preserve communicative purpose, learner agency, teacher judgment, and ethical responsibility.

Prompt engineering is often presented as the solution. In technical contexts, prompt engineering refers to the design and refinement of instructions that guide large language models toward useful outputs (Liu et al., 2023; Sahoo et al., 2024; Schulhoff et al., 2024). Practical guides commonly emphasize clarity, specificity, constraints, role assignment, examples, prompt chaining, and iterative refinement. These principles are valuable, and applied guides in other fields show how examples, tables, workflows, and case studies can make prompt engineering accessible to practitioners (Ekin, 2026). However, foreign language teachers need more than generic prompt tips because the success of a classroom prompt depends not only on the quality of AI output, but also on the quality of the learning activity it creates.

This article argues that prompt engineering for language teachers should be treated as pedagogical design. Here, engineering is used in the applied sense of intentional prompt design, testing, refinement, and constraint-setting. A prompt used in a foreign language classroom is not merely an instruction to GenAI. It is a compact lesson design, which defines the communicative goal, the learner role, the AI role, the linguistic focus, the expected output, the feedback process, the ethical boundary, and the evidence of learning. A weak prompt asks GenAI to complete the work, while a strong pedagogical prompt creates conditions for learners to think, compare, speak, write, revise, evaluate, and reflect.

Although many of the principles discussed in this article may apply to other GenAI tools, the paper focuses on ChatGPT (OpenAI, 2022) as the main practical example because of its widespread use in educational settings and its relevance to everyday foreign language classroom tasks. The paper makes three contributions. First, it reframes prompt engineering as a dimension of language-teacher expertise. Second, it organizes key prompt-design decisions into a practical checklist that teachers can use when preparing ChatGPT-supported foreign language tasks. Third, it provides examples, prompt templates, figures, tables, model lessons, and safeguards what teachers can adapt across writing, speaking, vocabulary, reading, pragmatics, and assessment. The core argument is that the strongest use of ChatGPT in foreign language classrooms is not answer generation, but guided participation in meaningful language tasks.

2. Prompt engineering as pedagogical design

2.1. Beyond generic prompt tips

Generic prompt engineering begins with a practical insight: better prompts produce better outputs. A vague prompt such as “Create a lesson” may generate a broad and unfocused response, while a more specific prompt such as “Create a 45-minute B1 lesson on polite requests in emails, including pre-task noticing, pair work, language focus, and a short reflection” is more likely to produce usable material. In classroom practice, specificity matters because teachers need outputs aligned with level, time, curriculum, and learner needs.

However, output quality is not the same as learning quality. A prompt that produces a polished essay may be technically successful but pedagogically weak if it replaces learner writing. A prompt that produces three revision questions may be less impressive as a product but stronger as a learning tool. Language learning depends on input, interaction, output, feedback, noticing, and reflection (Long, 1996; Schmidt, 1990; Swain, 2005). Prompts that bypass these processes risk improving performance while reducing development.

Prompt comparison: output generation vs learning support Weak prompt: Rewrite this student paragraph in perfect [language]. Stronger prompt: The learner is at a B1 level. Do not rewrite the paragraph. Identify two strengths, two areas for improvement, and two questions that help the learner revise independently. Focus only on organization and clarity. The stronger prompt preserves learner authorship and supports formative revision rather than substitution.

2.2. Prompt engineering as teacher professional knowledge

Prompt engineering should be located within the teacher’s professional knowledge. Teachers already design tasks, sequence instruction, select input, scaffold interaction, adapt materials, and assess performance. Prompt engineering extends these responsibilities into human-AI interaction. The teacher must decide what GenAI should do, what it should not do, how much support is appropriate, how output will be checked, and how learners will demonstrate ownership.

This view is consistent with responsible AI guidance, which emphasizes human oversight, transparency, privacy, fairness, and institutional accountability (European Commission, 2022; Miao & Holmes, 2023). In foreign language classrooms, these principles are concrete. Teachers must avoid entering identifiable student data into public tools, prevent GenAI from replacing learner production, check for inaccurate explanations, and ensure that GenAI feedback does not erase learner voice or impose narrow native-speaker norms. If prompt engineering is understood as part of teacher professional knowledge, then effective classroom prompting requires more than technical fluency with a GenAI tool. Table 1 summarizes the knowledge domains and their role in pedagogical prompt design.

3. Theoretical foundations

3.1. Task-based language teaching

Task-based language teaching (TBLT) offers a strong foundation for responsible AI use because it begins with meaningful language action rather than isolated practice. In TBLT, learners use language to complete communicative tasks such as solving a problem, planning an event, comparing options, reaching a decision, giving advice, or producing a text for a real or simulated audience (Ellis, 2003; Long, 2015; Nunan, 2004; Willis & Willis, 2007). Accuracy remains important, but it is integrated into meaningful use.

GenAI can support TBLT by generating role cards, producing input texts, simulating audiences, offering feedback, suggesting alternative expressions, and helping learners prepare for interaction. Yet GenAI can also undermine TBLT if it completes the communicative work for learners. The teacher’s task is to position GenAI as a resource within the task cycle, not as the performer of the task.

3.2. CEFR and the action-oriented classroom

The CEFR Companion Volume frames learners as social agents who mobilize linguistic, sociolinguistic, pragmatic, plurilingual, and intercultural resources to act in meaningful contexts (Council of Europe, 2020). This action-oriented view aligns closely with task-based prompt engineering and with earlier communicative approaches that emphasize grammatical, sociolinguistic, and strategic dimensions of language ability (Canale & Swain, 1980). It encourages teachers to ask not only what language learners know, but what they can do with language.

A CEFR-informed classroom prompt should therefore specify communicative action. For example, instead of asking ChatGPT to create a grammar exercise, a teacher might ask it to design a B1 task in which learners advise a classmate, use modal verbs, respond to follow-up questions, and self-assess clarity and politeness. The prompt connects language form with communicative function.

3.3. Formative feedback and learner agency

GenAI feedback is attractive because it is immediate, detailed, and scalable. Yet feedback is only effective when learners understand it and can act on it (Hattie & Timperley, 2007; Hyland & Hyland, 2019). If GenAI provides too many corrections, rewrites the learner’s text, or uses language beyond the learner’s level, it may reduce rather than increase learning.

For this reason, GenAI feedback prompts should be constrained. They should focus on selected criteria, avoid complete rewriting, use accessible language, and require learner decisions. The learner should remain the author of the text and the evaluator of AI suggestions.

4. Key decisions in foreign language prompt design

Prompt design in foreign language teaching should begin with a communicative learning purpose, not with an AI tool itself. Figure 1 presents a task-based cycle for integrating ChatGPT into foreign language classrooms. Effective foreign language prompt design begins with the communicative outcome, learner profile, and task stage. Teachers then write a constrained prompt that defines AI’s role, output format, and boundaries. The resulting output should not be treated as automatically correct or pedagogically appropriate. Learners should evaluate it against task criteria, revise or perform the task themselves, and document how ChatGPT was used. Finally, the teacher validates whether learners can demonstrate the target language or communicative outcome independently. Table 2 translates this process into a practical checklist for planning AI-supported foreign language tasks.

5. Core prompting techniques for foreign language teachers

The following prompting techniques are not presented as technical tricks. They are pedagogical design moves that help teachers control the learning process, protect learner agency, align ChatGPT use with communicative outcomes, and ensure that AI support remains connected to classroom aims.

5.1. Clear and specific instructions

Clarity and specificity are foundational. In foreign language education, a strong prompt should usually include level, topic, task purpose, time, interaction pattern, language focus, and desired output format. Without these details, ChatGPT may produce material that is too advanced, too long, culturally irrelevant, or disconnected from the lesson’s aim.

Example: improving specificity Weak prompt: Make a speaking activity. Stronger prompt: Create a 15-minute B1 pair-speaking activity on planning a weekend trip. Learners must make suggestions, agree, disagree politely, and reach a decision. Include role cards, useful phrases, and a self-assessment checklist.

5.2. Explicit constraints

Constraints help teachers control AI output. Useful constraints include word count, CEFR level, number of questions, type of questions, feedback scope, tone, genre, and prohibited actions. Constraints are particularly important when ChatGPT is used with younger or lower-proficiency learners.

Example: constrained reading prompt Create a 220-word B1 reading text about reducing plastic waste. Use short paragraphs, avoid idioms, include the words reduce, recycle, waste, protect, and environment, and write five comprehension questions: two factual, two inference-based, and one personal-response question.

5.3. Role prompting

Role prompting assigns ChatGPT a function, such as conversation partner, feedback assistant, vocabulary coach, text simplifier, or rubric checker. In foreign language contexts, the role should be bounded. Asking ChatGPT to act as “teacher” may be too broad. A safer approach is to define a limited role and specify what ChatGPT should not do. Table 3 illustrates how common AI roles can be bounded through explicit safeguards so that ChatGPT supports classroom learning without taking over learner production or teacher judgment.

5.4. Few-shot prompting

Few-shot prompting gives the model examples of the desired output, drawing on the capacity of large language models to perform tasks from examples provided in the prompt (Brown et al., 2020). In foreign language teaching, it is useful for controlling feedback tone, response length, proficiency level, and task format. Examples can prevent AI tools from producing overly technical, discouraging, or advanced language.

Example: few-shot feedback style Use this feedback style: - Your opinion is clear. Add one example to support it. - Your sentence is understandable. Check the verb tense in the second clause. Now give feedback on the learner paragraph below. Use the same simple and supportive style. Do not rewrite the paragraph.

5.5. Prompt chaining

Prompt chaining breaks complex tasks into stages. It is especially useful for writing, project work, problem-solving, and speaking preparation. Instead of asking ChatGPT to produce a complete final answer, the teacher guides learners through a sequence: brainstorm, select, organize, draft, receive feedback, revise, and reflect. Table 4 situates prompt chaining alongside other prompting techniques that can be adapted for foreign language teaching.

6. Reusable prompt templates for teachers

The following templates are designed for adaptation rather than direct copying. Teachers can complete the bracketed fields by adding the learner level, lesson goal, target language, task criteria, text, vocabulary set, or communicative purpose. The templates are intentionally structured to prevent common problems such as over-automation, level mismatch, vague feedback, and lack of learner reflection. They also keep ChatGPT use connected to classroom decision-making by specifying what ChatGPT should produce, what it should not do, and how learners will use the output. Table 5 provides reusable prompt templates for common foreign language teaching purposes.

7. Classroom applications

The following section illustrates how ChatGPT-supported prompt engineering can be applied across common areas of foreign language teaching. As shown in Figure 2, each classroom application should connect a specific AI function with a learner process and a safeguard that preserves learner agency, teacher judgment, and responsible classroom use.

7.1. Writing feedback

Writing feedback is one of the most immediate applications of GenAI in foreign language classrooms. To preserve authorship, GenAI feedback should be limited to comments, priorities, and revision questions rather than full rewriting. For example, students can first write a 150-word email requesting information about a school exchange programme. The teacher then uses a ChatGPT feedback prompt that focuses on clear purpose, organization, politeness, and grammar, while explicitly prohibiting full rewriting. Students revise their own email, highlight changes, and submit a reflection explaining which ChatGPT suggestions they accepted or rejected.

7.2. Speaking rehearsal

GenAI can simulate interlocutors and provide repeated practice, which is valuable in contexts where learners have limited opportunities to speak a foreign language outside class (Belda-Medina & Calvo-Ferrer, 2022). However, GenAI should support rather than replace human interaction (Meadan et al., 2026). A strong sequence is: ChatGPT rehearsal, peer interaction, teacher feedback, and learner reflection. For example, learners can rehearse asking for information at an airport desk with ChatGPT, then perform the same scenario with a classmate. ChatGPT provides preparation, but the communicative event remains social.

7.3. Vocabulary development

Vocabulary learning requires repeated exposure, retrieval, depth of processing, and meaningful use (Nation, 2013). GenAI is useful for generating contextualized examples, gap-fill tasks, sorting tasks, and communicative practice. However, prompts should avoid isolated word lists. A better prompt asks ChatGPT to create a problem-solving task in which learners must use target words to make decisions, negotiate, or produce a spoken or written outcome.

7.4. Reading adaptation

Teachers often need to adapt authentic texts to the learner’s level. GenAI can simplify vocabulary and sentence structure quickly, but simplification may remove nuance, argument structure, or cultural information (Aldamen et al., 2026). Therefore, reading adaptation prompts should require ChatGPT to list what it changed and what may have been lost. Learners can then compare the original and simplified texts, turning ChatGPT adaptation into a critical reading activity.

7.5. Intercultural pragmatics

Foreign language learners need to understand how language choices vary by relationship, power, distance, genre, and context (Pflaeging & Schleef, 2026; Taguchi, 2019). ChatGPT can generate multiple versions of requests, apologies, disagreements, and invitations. However, teachers must prevent cultural stereotyping. Prompts should explicitly state that appropriateness depends on situation and community, not fixed national stereotypes.

7.6. CEFR-informed assessment practice

GenAI can help learners understand assessment expectations by generating practice tasks and self-assessment checklists (Bearman et al., 2023). It should not replace teacher judgment or official CEFR assessment. A responsible prompt asks ChatGPT to create a CEFR-informed task, but not to assign an official level. This distinction preserves the formative role of ChatGPT and the professional role of teachers.

8. Model lessons

8.1. Model lesson 1: Writing a polite request email

Context: B1 foreign language learners are learning how to write semi-formal emails. The communicative outcome is to request information clearly and politely.

Stage 1 - Noticing: The teacher prompts ChatGPT to generate two short emails: one too direct and one appropriate. Students compare tone, greeting, request forms, and closing. This stage develops awareness before production.

Stage 2 - Planning: Students complete a planning table with a greeting, reason for writing, two questions, polite closing, and signature. ChatGPT can generate the planning table, but students fill it in.

Stage 3 - Drafting: Students write their first draft without ChatGPT completing it.

Stage 4 - Focused feedback: ChatGPT gives comments on purpose, politeness, organization, and grammar, but does not rewrite the text.

Stage 5 - Revision and reflection: Students revise, highlight changes, and explain one ChatGPT suggestion they accepted and one they rejected.

Teacher prompt for the feedback stage Read this B1 learner email. Do not rewrite it. Give feedback on purpose, politeness, organization, and grammar. For each category, give one short comment and one revision question. Use simple [language]. End with a four-item checklist for revision.

8.2. Model lesson 2: Giving advice in spoken interaction

Context: A2-B1 learners practice giving advice to classmates using “should”, “could”, “why don’t you”, and “maybe you can”.

Stage 1 - Input: ChatGPT generates problem cards about school-related issues while avoiding sensitive topics.

Stage 2 - Controlled practice: Students match advice expressions to problem cards and identify which expressions are more polite or supportive.

Stage 3 - ChatGPT rehearsal: Learners practice with ChatGPT, which asks one question at a time and gives delayed feedback after three exchanges.

Stage 4 - Peer role-play: Students perform the role-play with classmates. This maintains human interaction as the central communicative event.

Stage 5 - Reflection: Learners identify one phrase they used successfully and one communication strategy they want to improve.

Student-facing speaking prompt Act as a classmate with a simple school problem. Tell me the problem in two sentences. Wait for my advice. Ask one follow-up question. After three exchanges, give feedback on clarity, politeness, and use of advice phrases. Do not correct every grammar mistake.

8.3. Model lesson 3: Reading adaptation and critical AI literacy

Context: B1 learners read a short article about online learning. The teacher wants to make the text accessible without hiding the fact that simplification changes texts.

The prompt asks ChatGPT for a simplified version, a list of changed vocabulary, and an explanation of any removed details. Learners compare the original and simplified versions, identify meaning differences, and discuss whether simplification helped or weakened the text.

This lesson develops reading comprehension and critical AI literacy at the same time. Learners see that AI output is useful, but not neutral.

9. Ethical risks and safeguards

GenAI can support foreign language teaching, but its outputs should not be accepted uncritically. Before using ChatGPT-generated materials, feedback, or assessment support in class, teachers need a simple review process to check whether the output is accurate, level-appropriate, unbiased, privacy-safe, and pedagogically useful. Figure 3 presents a responsible prompt review loop for managing AI-related risks before classroom use.

9.1. Hallucination

GenAI may produce explanations that sound fluent and convincing but are factually wrong (Georgiou, 2026). In foreign language contexts, this may include false grammar rules, inaccurate cultural claims, unnatural examples, or invented academic references. For this reason, prompts should explicitly instruct ChatGPT not to invent facts, sources, or examples when it is uncertain. Teachers should also verify AI outputs before using them in class, especially when the content concerns grammar explanations, intercultural knowledge, or references (Shoufan & Esmaeil, 2026).

9.2. Overcorrection

GenAI can support feedback and revision, but it may also rewrite learner work too extensively. When this happens, the learner’s original meaning, voice, and developmental level may disappear. This is especially problematic in language learning because errors can show what learners are ready to notice and improve. Prompts should therefore ask ChatGPT to give feedback, identify patterns, and correct only errors linked to the lesson focus rather than fully rewriting the text (Schultz, 2026).

9.3. Bias and linguistic normativity

GenAI systems may reproduce linguistic and cultural biases from their training data (Georgiou, 2025a). In English language teaching, this can mean privileging dominant varieties such as Standard American or British English, treating other varieties as deficient, or producing culturally narrow examples. Teachers should therefore prompt ChatGPT to include international English contexts, avoid stereotypes, and focus on intelligibility, appropriateness, and communicative effectiveness rather than native-speaker norms alone (Bender et al., 2021).

9.4. Privacy

Teachers should avoid uploading identifiable learner data into GenAI tools because student writing, names, locations, classroom events, or personal experiences may contain sensitive information. Before using ChatGPT to analyse or adapt learner texts, teachers should anonymize the data, remove personal details, and follow institutional and national data-protection policies. This is particularly important when GenAI tools are hosted by external providers whose data storage and training practices may not be fully transparent (Miao & Holmes, 2023).

9.5. Learner dependency

GenAI can provide immediate explanations and answers, but excessive reliance may reduce learners’ independence, critical thinking, and willingness to struggle productively with language (Georgiou, 2025b). If students always receive complete answers, they may skip the cognitive work needed for noticing, hypothesis-testing, revision, and self-monitoring. Prompts should therefore ask ChatGPT to provide hints, guiding questions, examples, and staged feedback rather than final products. This keeps ChatGPT use supportive while preserving learner agency.

9.6. Academic integrity

Students need clear rules about when GenAI assistance is allowed, what kinds of support are acceptable, and how GenAI use should be acknowledged. Instead of relying only on AI-detection tools, teachers can make learning processes more visible through process logs, drafts, revision histories, oral explanations, and AI-use declarations. These practices help distinguish acceptable support from misconduct and encourage students to reflect on how models such as ChatGPT contributed to their work (Gonsalves, 2025).

AI-use declaration template I used AI to: brainstorm ideas / check vocabulary / receive feedback / practice speaking / improve organization. The most useful AI suggestion was: _____. I accepted it because: _____. I rejected this suggestion: _____. My final work is my own because: _____.

10. Prompt failure modes and repairs

Prompt failure should be treated as part of iterative instructional design rather than as a reason to abandon GenAI use. When AI output is unsuitable, the teacher can revise the prompt by adding context, constraints, examples, criteria, or safeguards. In foreign language classrooms, common failures include unclear instructions, level mismatch, over-automation, overcorrection, hallucination, cultural bias, feedback overload, lack of reflection, hidden GenAI use, and teacher overtrust. Table 6 summarizes these failure modes, shows how they may appear in classroom prompts, and suggests practical repairs.

11. Discussion

This article has argued that prompt engineering for foreign language teachers should be understood as a form of pedagogical design rather than as a purely technical skill. This shift is important because the value of GenAI in language education does not depend only on the fluency or speed of AI-generated output. In foreign language classrooms, the more important question is whether AI-supported tasks create opportunities for learners to notice language, produce meaning, receive feedback, revise their work, interact with others, and reflect on their learning. These processes are central to second language development, particularly in relation to noticing, interaction, output, and feedback uptake (Hattie & Timperley, 2007; Long, 1996; Schmidt, 1990; Swain, 2005).

The discussion also highlights the continuing importance of teacher judgment. ChatGPT can generate texts, dialogues, explanations, quizzes, feedback, and assessment-related materials, but it does not automatically know local curriculum goals, learner histories, classroom relationships, cultural expectations, institutional policies, or assessment standards. Teachers, therefore, remain responsible for deciding what ChatGPT should do, what it should not do, how much support is appropriate, and how AI output will be checked before use. This position is consistent with responsible AI guidance, which emphasizes human oversight, transparency, privacy, fairness, and accountability in educational settings (European Commission, 2022; Miao & Holmes, 2023).

A key implication is that ChatGPT use should be integrated into classroom workflows rather than treated as a shortcut to final products. The practical checklist proposed in this article helps teachers specify the learning target, learner level, communicative situation, input materials, process stage, ChatGPT role, output format, monitoring procedure, prompt refinement, and transparency expectations. These decisions make ChatGPT use more pedagogically controlled. For example, ChatGPT can support brainstorming without writing the final answer, provide feedback without erasing the learner’s voice, simulate speaking practice without replacing peer interaction, and generate assessment practice without assigning official proficiency levels. In this way, ChatGPT becomes a pedagogical support tool that encourages participation and reflection rather than a substitute for learner effort or teacher expertise (Council of Europe, 2020).

Implementation requires more than giving teachers lists of useful prompts. Professional development should help teachers evaluate AI output, identify hallucinations, avoid overcorrection, protect learner privacy, recognize bias, and design process-based assessments. Teachers need opportunities to compare weak and strong prompts, revise prompts collaboratively, build local prompt banks, and adapt templates to their own learners and curricula. They also need institutional guidance on acceptable AI use, AI-use declarations, data protection, and academic integrity. Without such support, ChatGPT integration may remain inconsistent, overly dependent on individual teacher confidence, or disconnected from curriculum and assessment policy (Bearman et al., 2023; Gonsalves, 2025; Miao & Holmes, 2023).

The article also suggests that responsible AI use in foreign language education should be evaluated through evidence of learning, not only through the quality of AI-generated output. A polished ChatGPT-generated paragraph, dialogue, or rubric is not necessarily evidence that learners have developed communicative competence. Stronger evidence may come from learner drafts, revision histories, oral explanations, peer interaction, reflection logs, self-assessment checklists, and teacher observations. This is especially important because ChatGPT can improve visible performance while reducing the productive struggle through which learners develop independence, accuracy, and communicative confidence (Yan et al., 2024; Zaim et al., 2025).

12. Conclusions

This article has presented a practical checklist, prompt templates, classroom applications, model lessons, ethical safeguards, and failure-mode repairs to show how ChatGPT can support lesson planning, writing feedback, speaking rehearsal, vocabulary learning, reading adaptation, intercultural pragmatics, CEFR-informed assessment practice, and AI literacy while preserving teacher judgment, learner agency, transparency, privacy, and academic integrity. However, the guidance remains conceptual and practice-oriented rather than empirically validated, and it should be tested across different foreign language contexts, learner ages, proficiency levels, institutions, and GenAI tools. Future research should examine whether pedagogically designed prompts improve learning outcomes such as writing revision quality, speaking confidence, vocabulary retention, reading comprehension, feedback uptake, and critical AI literacy.