Strategies for Designing Language Output Tasks in Generative AI-Assisted English Reading Instruction at the Tertiary Level ()
1. Introduction
College English teaching is required by national guidelines to develop learners’ ability to use English, yet the reading lesson, its most stable unit, remains organised around comprehension [1]: students read, answer questions and move on. Reading and writing are often treated as separate components rather than integrated activities [2]. Output can support learning from comprehension: producing language triggers noticing, hypothesis testing and metalinguistic reflection [3], and the production-oriented approach places it at both ends of the teaching sequence [4].
Recent discussions include some digitally empowered, output-oriented models for college English reading [5] and a review of ChatGPT’s potentials and pitfalls for second language writing process [6]. This paper proposes a three-tier task framework and four teaching strategies for designing output tasks in reading lessons and illustrates the frame with a lesson practice.
2. Theoretical Foundations
2.1. The Output Problem in College English Reading Instruction
Reading comprehension research describes understanding as a process in which textual information and prior knowledge are integrated with each other [7]. On this account, only answering a set of comprehension questions does not demonstrate that the reader can use the targeted language again. Production can help a lot. To retell, summarise or respond, students must decide what the text says, and the output shows teachers what they have learnt and corrected. Writing about what they have read has also been shown to improve comprehension [8].
The curricular context makes the point sharper. The national guidelines for college English name the ability to use English for communication as the goal of the course [1], and the production-oriented approach treats output as the driving force of the sequence: teaching begins from a productive task, and move on to the language input, and returns to the task in improved form [4]. Traditional classrooms attach importance to input but neglect output. They usually assume that students can produce language naturally with enough input.
2.2. Scaffolding the Move from Text to Production
Moving from reading to language production is challenging. Scaffolding theory explains why learners need support and how this support can fade. Support matched to learners’ current ability helps them do tasks they could not finish alone. Teachers should gradually remove this support so learners can build up their competence [9]. Research on reading-writing integration sees reading and writing as connected activities [2]. students will learn sentence structures, vocabulary and text organisation patterns from the reading text and then use them in their output. GenAI can help the process of scanffolding. It can produce frames, variants, model texts and draft criteria for students, but the decisions about what to scaffold towards are pedagogical and require teachers’ technological, pedagogical and ethical judgement [10]. Table 1 summarises this division of labour for the three task types used in this paper.
Table 1. Three tiers of output tasks and the distribution of work between GenAI and the teacher.
Task type |
Output function |
GenAI contribution |
Teacher responsibility |
Guided retelling |
Reconstruct the text’s content orally in structured form |
Generate retelling frames and prompt cards |
Select the frame; judge completeness and accuracy live |
Constrained summary |
Reduce the text to a fixed-length,
reader-ready summary |
Produce reference summaries at several lengths |
Set length and source constraints; verify coverage against the text |
Transfer and response |
Apply the text’s structure and language
to a new topic and defend a position |
Supply counter-arguments and alternative organisations |
Define the transfer target; own the evaluative stance |
3. A Three-Tier Framework of Output Tasks
3.1. Tier One: Guided Retelling
The first tier focuses on text reconstruction. Students retell the content following a teacher-provided frame. They are encouraged to keep the text’s organisational structure, and they can freely express ideas using their own words. For narrative text, students may retell the content following the pattern: situation, development, complication and resolution. For expositions, they can follow some structure such as phenomenon, mechanism and evidence. GenAI helps at this tier by producing different kinds of retelling frames. It can also prepare prompt cards for less confident students. Teachers will decide the details of the frame and check whether the frame is accurate.
3.2. Tier Two: Constrained Summary
In the second tier, students will be asked to produce a shortened version of the text. The summary has a fixed length, should stay within the original text, and has a clear intended audience. Students cannot copy phrases directly from the source text. They need to paraphrase texts’ ideas. With a clear audience in mind, students can select key content from the text that they think can cater to the audience.
GenAI can create reference summaries. These summaries work as samples to inspire students. They can also serve as comparison materials. After students finish their own drafts, teachers show these alternative versions for comparison activities. Teachers take on checking work that GenAI cannot do. They check if AI’s summaries are against the source. and they will make sure no invented content appears. Meanwhile, they will examine students’ summaries to spot some key information that is not from the original text.
3.3. Tier Three: Transfer and Response
The third tier asks students to transfer what they learn from the text to new situations. For example, they may use the original text’s organisational structure for a new topic and apply its key expressions to develop new arguments. This helps students learn to use English for practical purposes. During this process, GenAI can offer different text structures for students to compare. It can help students find suitable words and provide relevant arguments and phrases to support their writing. Teachers’ task is to set the goal of this transfer task and control the evaluation criteria.
4. Teaching Strategies
The four strategies below fix the decisions that determine whether an output task designed with GenAI recruits the learner’s own language or displaces it: what the task requires the learner to produce, when the machine may act on the learner’s text, who owns the criteria by which the work is judged, and how the design reaches the individual learner. Table 2 summarises them.
Table 2. The four strategies, the design decision each fixes, and the work each assigns to GenAI and to the teacher.
Strategy |
Design moment |
GenAI role |
Teacher role |
1. Output anchored in the text’s key vocabulary |
Specification |
Draft tasks and prompts keyed to the listed items; check coverage of the list |
Decide which listed items the task must require, and at what level of use |
2. Learner production first, AI revision immediately after |
Sequence |
Comment on and propose revisions to the learner’s own completed draft |
Protect the unaided first attempt; assess draft and revision separately |
3. Criteria by class covenant, not by machine fiat |
Assessment |
Apply the agreed criteria to a draft as a reader’s report only |
Draft the criteria with the class; make the final judgement |
4. Using GenAI for individualised help |
Provision |
Produce differentiated variants and
per-learner reports and practice material |
Set the common target; audit the individualised material |
4.1. Strategy 1: Producing from the Key Vocabulary of the Text
An output task is anchored when the language it requires has already been taught, and the one part of a reading lesson that the whole class demonstrably shares is the unit’s key vocabulary list. The list is the operational form of the principle that every task must be answerable with the language resources of the text: it is bounded, it has been selected and glossed by the course designers, and it is what the teacher can hold every learner accountable for. Strategy 1 therefore states the output target in the currency of that list. The task must require the learner to use a specified number of the listed items, in sentences the learner constructs, about content the learner supplies: six of the items for describing change and its effects at the first tier, four at the second and third. The requirement is one of use rather than of occurrence, since an item inserted where it does not belong has been displayed, not used. Stating the target in these terms does the same work for the machine as it does for the learner: it converts a general request for activities into a specification that can be checked, because a task such as ‘write a three-hundred-word exposition that uses at least four of the unit’s listed items for describing change’ names both the function and the resource, and both teacher and model can be held to it.
4.2. Strategy 2: Students Output Their Language First, AI Helps with Revision Immediately Afterwards
Students must complete their first draft independently, without using AI. If AI provides a text before students begin writing, they may simply edit or copy the AI-generated text. As soon as the draft is finished, students can use AI to receive feedback and revise their language output. They can compare the AI feedbacks with their original version, identify problems, and decide which suggestions to accept. This sequence keeps students responsible for their language usage while allowing AI to provide timely help. It also enables the teacher to see how students grow their language. Recent research on GenAI-assisted college English writing similarly recommends independent drafting before AI-supported language optimisation, feedback and critical revision [11].
4.3. Strategy 3: Criteria by Class Covenant, Not by Machine Fiat
Teachers and students should work together to agree on evaluation criteria, instead of letting GenAI take charge of setting them. Only criteria negotiated jointly by teachers and students fit the real teaching context well. Criteria produced by GenAI are usually general summaries drawn from large datasets and lack personalisation. Meanwhile, negotiating these criteria helps teachers and students build their own sense of judgement, rather than following others blindly.
4.4. Strategy 4: Using GenAI for Individualised Help
Providing individual support is a tough challenge in teaching. What GenAI can assist is personalised learning routes. First, differentiated materials: it produces frames, task rules and reference texts with different levels of support. Less confident learners can get more scaffolding generated by AI. Second, AI can provide individual feedback reports. They can apply the evaluation criteria agreed by the class to each student’s draft. The reports point out which key words are missing, which discourse moves are left out, and which sentences can be revised for better quality. Third, AI can produce targeted after-class practice. Practice tasks are designed based on the vocabulary and discourse that the student’s have not yet mastered well. The teacher audits a sample of the individualised material before it reaches learners, because AI-generated responses require critical evaluation and may lack adequate evidential grounding or qualification [12].
5. Illustration: Designing Output Tasks for a College English Reading Text
The text is Text A of Unit 3 of the first book of the current edition of a national college English reading-writing textbook series, an exposition on how the Internet has transformed university life [13].
5.1. The Text and Its Output Potential
The text describes how Internet access and digital devices have reshaped university life. It begins with students’ everyday use of laptops, smartphones and constant connectivity, considers both the educational benefits of these technologies and the need to regulate online time, and then turns to universities’ investment in networks, devices and support services. It closes by emphasizing the educational potential of computer technologies and the prospect of developing problem-solvers and intelligent thinkers. As a language resource, it offers recurring formulations for presenting technological change and its effects, numerous campus-based examples, and a key vocabulary list for the unit. As a content object it is double-edged in a useful way: its topic, technology transforming student life, is the topic on which students have the most recent first-hand evidence, since the text describes the Internet era and they live in the AI era.
For teaching purposes, the text can be reorganized into a three-part thematic frame: students’ everyday experience of digital campus life, universities’ technological provision, and support services together with the closing educational outlook. This frame is constructed by the teacher from the text’s content; it is not an explicitly marked three-part structure in the source. The output target is derived from these resources under Strategy 1, which states it in the currency of the unit’s key vocabulary list: across the three tiers learners must use a specified number of the listed items in sentences they construct themselves, while this teacher-constructed frame is rehearsed and finally transferred to a topic the text does not cover. Because the list is common to the whole class and has already been taught, the target is both anchored and checkable, which is what makes the word list the practical form of the principle that a task must be answerable with the language resources of the text. The transfer target almost proposes itself, and its aptness is not a contrivance: a text about one technological transformation of university life is the natural source text for a production task about another.
5.2. Tier One in Practice
Teachers use GenAI to generate retelling frames and key vocabulary based on the source text, and check whether these words are the key vocabulary from the passage. Students work in groups of three to finish the retelling task with support from the frame. Each student can retells one part of the text. Listeners evaluate their peers’ retelling according to the evaluation criteria agreed by teachers and students.
Notably, GenAI produces differentiated key expressions based on students’ proficiency levels, offering references for students’ language production. Still, teachers decide which scaffolds should be given to which level of students.
5.3. Tier Two in Practice
Students need to write a summary of around 300 words. They must not copy sentences directly from the source text, and they should imagine that their readers are classmates who want to learn about university life. For vocabulary use, students need to use at least four key words and phrases from the text accurately. After students finish their drafts, AI provides sample summaries for them to refer to and revise students’ works. Teachers shall check the AI summaries to see if they meet the agreed evaluation criteria. Then GenAI and the teacher assess students’ work together based on these criteria they have made together. Finally, the teacher guides students to reflect on their own writing: What content in the original text does the AI sample leave out? What main contents does it keep? What aspects should you pay attention to in your next writing task?
5.4. Tier Three in Practice
The transfer task asks students to apply the learned text organisation and vocabulary to a new topic. They write an exposition of around 300 words on how AI tools are transforming university life. Students develop their writing structure based on text frames. GenAI first generates these frames from the source text and offers differentiated scaffolding according to students’ langugage abilities. Besides, students need to use core vocabulary from the text in their original sentences. These language resources are provided by AI, and teachers need to check these language scaffolds. Finally, GenAI, the teacher and peers assess students’ work for creativity and accurate language use.
6. Conclusion
In summary, the three-tier framework creates a learning path from retelling to summary writing and to transfer tasks. The four strategies give teachers practical tools to tackle common problems in output-driven classes. In the design of output tasks in reading lessons, AI’s role is to assist students’ language output, not to replace them. Students are encouraged to write or speak independently, set assessment criteria proudly and negotiate with their classmates confidently. Of course, AI can give students and teachers lots of inspiration. But what is more important is that teachers’ and students’ independent thinking counts more. GenAI excels in preparing personalized learning materials. Teachers use their professional judgement abilities to set the tasks for students and judge whether students have met the learning goals.
Acknowledgements
The author thanks colleagues in the college English course team for discussions of the output problem in reading lessons and for comments on earlier versions of the framework.
This work was supported by the Teaching Reform Research Projects of Qilu Normal University (Grant Nos. XJG2024007).