Designing Inclusive Multimodal Learning Content with Generative AI for Migrant Adult Literacy: A Practice-Oriented Methodological Proposal ^†

Abstract

This article presents a practice-oriented methodological proposal for designing inclusive multimodal learning content with Generative AI (GAI) in migrant adult literacy. It does not report an experimental intervention or a statistical evaluation of learning outcomes. Its contribution lies in formalizing a context-sensitive design pathway for early preA1–A2 literacy and language-learning provision in Italian CPIA settings, where learner profiles are highly heterogeneous, attendance may be discontinuous, and written language is both a learning goal and a barrier to participation. Unlike generic AI-supported instructional design frameworks, the proposed approach starts from recurrent communicative needs in adult migrant education and translates them into short, modular and reusable learning artifacts that coordinate textual, visual, audio-oral and interactive layers. The framework distinguishes multimodal design, understood as the pedagogical coordination of different semiotic modes, from the mere use of multiple media. It also integrates accessibility as a set of concrete design criteria, including linguistic readability, visual clarity, audio quality, layout, font size, contrast, cognitive load and usability in print or mobile formats. The article outlines a sequence of design operations: mapping learner profiles, selecting situated communicative scenarios, generating and revising textual material, developing visual and audio scaffolds, structuring guided interaction, and applying pedagogical, cultural and ethical review. An illustrative micro-unit on asking for information at a municipal office shows how this pathway can support dialog, visual glossary, audio practice, role-play and formative assessment. The proposal is intended for CPIA educators, adult literacy professionals, instructional designers and researchers in multimedia learning and educational technology. Its educational implication is that GAI can support inclusive material design only when its outputs are treated as provisional resources to be selected, adapted and validated through human pedagogical judgment.

1. Introduction

Adult migrant literacy is among the most demanding areas of contemporary language education. It sits where second-language learning, social inclusion, interrupted educational biographies and immediate communicative needs meet. In the Italian context, this complexity is addressed primarily through the CPIA (Provincial Centers for Adult Education), which include literacy and Italian-language pathways for foreign adult learners and aim at the attainment of at least A2 level of the Common European Framework of References [1,2,3,4,5]. Yet the formal labels preA1–A2 only partially describe what happens in classrooms. In CPIA settings, progression rarely follows a neat sequence. Attendance may be discontinuous, prior competences are often difficult to assess, and the group may change from week to week. The term preA1–A2 is therefore used here not as a narrow measure of linguistic attainment, but as a practical reference to early-stage pathways that range from emerging literacy to the first levels of basic language use [2,6]. In such settings, even the word class requires caution. New arrivals, departures, work obligations, family responsibilities and administrative constraints continually reshape the learning group. Planning cannot rely only on a cumulative chain of ordered units. It must be flexible, recursive and modular [1,7]. Teachers often work with something closer to a linguistic multigrade class than to a homogeneous course: learners differ in orality, reading, writing, familiarity with the Latin alphabet, previous schooling and even in their expectations of what schooling is. A central pedagogical principle follows from this condition: materials may be linguistically simplified, but the adult learner must not be simplified. Reducing vocabulary, syntax or task complexity must not lead to childish content, paternalistic tone, infantilized images or activities detached from adult life. The aim is to produce materials that are accessible and usable, while remaining dignified, situated and socially meaningful [1,2,8,9,10,11,12]. Here, accessibility is understood in a concrete design sense. It concerns linguistic readability, visual clarity, font size, contrast, layout, audio pace and quality, cognitive load, printability and usability on mobile devices. Multimodal content becomes relevant at precisely this point. For learners at preA1–A2 level, written text is both a goal and a barrier. Words, images, audio, situated dialogs, visual glossaries and guided activities can help learners access meaning even when autonomous reading remains fragile. Properly designed, these resources distribute cognitive load and allow participation through more than one channel [13,14,15,16].

In this article, multimodality refers to the pedagogical coordination of different semiotic modes, such as written text, image, audio, layout, gesture and interaction. Multimedia refers more broadly to the presence of different media formats within the same artifact. The distinction matters because the simple accumulation of text, images and audio does not automatically produce a coherent learning resource.

GAI enters this discussion as a possible support for the design of flexible, modular and multimodal materials. Text, image and audio models can produce variants of the same content, grade short texts, draft situated dialogs, suggest visual glossaries, prepare scripts for oral practice and help build micro-units around everyday communicative situations [17,18,19]. This potential, however, is conditional. In education with socially and linguistically vulnerable adults, GAI cannot be treated as an automated instructional designer. Outputs that look clear and inclusive may still contain stereotypes, hidden cultural assumptions, problematic visual representations, inaccurate simplifications or tones that do not respect the adult status of learners [20,21]. Recent work on GAI in education has highlighted the potential of large language models for adaptation, feedback, content generation, inclusive design and customizable learning experiences [17,18,19]. Other studies have begun to examine AI literacy, digital divides and the ways in which adult learners from migrant and refugee backgrounds engage with AI-mediated language and literacy practices. In parallel, research on migrants’ digital literacies and technology-supported Italian L2 has shown the relevance of digital tools, speech technologies and conversational assistants for language learning and participation in everyday contexts [22,23,24,25]. Research on multimedia learning has also clarified how text, image, audio and layout may support comprehension and manage cognitive load [13,14,15,16,26,27].

These strands provide an important foundation, but they do not fully address the specific design problem considered in this article. The research gap does not lie in the absence of multimodal resources for language learning, nor in the novelty of AI-generated content as such. Both areas have already been widely discussed. The gap lies in the lack of a context-sensitive methodological pathway that connects GAI-supported production with the constraints of CPIA classrooms, the heterogeneity of adult migrant learners, the need for short and reusable materials, the risk of infantilization, and the requirement for pedagogical, cultural and ethical human review. Generic AI-supported instructional design frameworks usually address content generation or adaptation at a broad level. They rarely specify how a real communicative need in adult migrant literacy can become a coordinated artifact in which text, image, audio, interaction and assessment remain aligned with the learner’s level, adult status and context of use.

This article therefore does not offer a statistical study of learning gains. Its aim is methodological. It proposes a practice-oriented pathway for designing inclusive multimodal content for adult migrant literacy in early preA1–A2 CPIA contexts [1,2,6,28]. The pathway brings together linguistic simplification, visual support, audio-oral work, authentic communicative situations, accessibility, adult appropriateness, intercultural sensitivity and ethical human review. GAI is treated as an adaptive infrastructure for producing and reworking content, not as a replacement for pedagogical expertise. The contribution of the article is to shift attention from isolated generated outputs to a controlled design process through which those outputs are selected, revised, combined and validated. The specific objectives of the article are threefold. First, it aims to define the design conditions under which GAI can support the development of inclusive multimodal materials for adult migrant literacy. Second, it seeks to distinguish a coherent multimodal instructional artifact from the mere juxtaposition of multimedia components. Third, it proposes a set of quality criteria that can guide teachers, adult literacy professionals and instructional designers in reviewing GAI-supported materials before classroom use.

The article is guided by the following research questions:

▪

RQ1. How can recurrent communicative needs in CPIA preA1–A2 pathways be translated into inclusive multimodal learning artifacts with GAI support?

▪

RQ2. Which design criteria are needed to distinguish pedagogically coherent multimodal artifacts from merely generated multimedia outputs?

▪

RQ3. Under what conditions can GAI-supported materials remain linguistically accessible, adult-appropriate, culturally respectful and usable in heterogeneous adult migrant literacy classrooms?

Given the methodological and non-experimental nature of the article, the hypotheses are formulated as working design hypotheses rather than as statistical hypotheses to be tested through outcome measurement. H1: A design pathway that begins from recurrent communicative scenarios, rather than from tool functions, can produce materials more closely aligned with the real conditions of CPIA preA1–A2 classrooms. H2: GAI-supported materials become pedagogically meaningful when textual, visual, audio-oral and interactive layers are functionally coordinated around the same communicative goal. H3: Human review is a structural condition for the responsible use of GAI in adult migrant literacy, because linguistic accessibility, adult appropriateness, cultural sensitivity and ethical safety cannot be guaranteed by generated outputs alone.

On this basis, the article develops a context-sensitive design pathway, a set of quality criteria and an illustrative application tied to everyday communicative needs. The proposed framework is intended for CPIA educators, adult literacy professionals, instructional designers and researchers working at the intersection of multimedia learning, educational technology and migrant education.

2. From A1–A2 Literacy Needs to Multimodal Content Design

2.1. Adult Migrant Literacy Beyond Language Level

In literacy pathways for adult migrants, preA1–A2 levels should not be read as graded measures of language proficiency alone. They are useful for describing initial communicative abilities, but they do not capture the full profile of adult literacy learners. L2 competence intersects with prior literacy, familiarity with alphabetic writing, previous schooling, ability to use functional texts, knowledge of institutional practices, and the possibility of using language for everyday autonomy [3]. PreA1–A2 therefore marks more than limited vocabulary and grammar. It also marks a fragile access to the meaning-making systems of written language and public communication. A learner may produce recurring oral formulas and still struggle to decode a notice, complete a form, read a school message or interpret a short administrative text. At the same time, this learner may possess rich practical, occupational and relational experience that remains invisible in school because the linguistic and semiotic tools for expressing it are not yet available. This distinction has methodological consequences because literacy cannot be reduced to grammatical progression. In adult migrant education, becoming literate means gradually entering social practices mediated by language: reading, listening, naming, asking, answering, filling out, recognizing, choosing and finding one’s way. Language is not only an object of learning: it is also a means of participation. Teaching content must therefore respond to two demands: it must be linguistically accurate and it must activate practical competence, contextual recognition and growing communicative autonomy. Adult migrant literacy is a multidimensional process in which orality, reading, writing and pragmatic competence do not always develop together. A learner may speak with relative confidence while writing very little, recognize isolated words without understanding the function of a text, or memorize formulas without transferring them to a new situation. Heterogeneity is not an exception to be corrected. It is part of the learning environment itself. Recent research on digital and AI-mediated literacy reinforces this multidimensional view. Studies on migrants’ and refugees’ digital literacies show that language learning increasingly intersects with access to mobile devices, online services, platform-mediated communication and everyday digital participation [22]. In adult education contexts characterized by superdiversity, AI literacy has also been discussed as an emerging component of participation, since learners and teachers are increasingly exposed to automated translation, speech technologies, conversational agents and generative systems [29,30,31]. These studies do not replace the established literature on adult literacy; rather, they extend it by showing that early literacy today includes the ability to navigate written, oral, visual and digital signs across institutional and everyday environments. For CPIA preA1–A2 learners, this means that AI-supported materials should not be understood only as simplified texts generated by a tool, but as resources situated within broader practices of communication, recognition and mediated participation.

2.2. Functional and Situated Content for Early Literacy

If adult migrant literacy is about access to concrete social practices, early-level teaching content must begin from recognizable communicative situations. In preA1–A2 pathways, a material is not effective simply because the language is easy. It is effective when it connects language, action and context [10,12]. Accessible content helps the learner understand what is being said, who is saying it, where the exchange takes place, why it matters and which minimal linguistic forms are needed. These dimensions are especially important when the language being learned is directly tied to everyday life: health, school, work, transport, housing, public services, documents, appointments, safety and institutional communication. This is not a narrow utilitarian view of literacy, but it is a functional one. Reading and writing are situated practices that enable people to act within social environments. Materials for early literacy should therefore be short, self-contained, and reusable. They cannot assume that every learner has followed the previous lesson or that the group will remain the same next week. Each material needs an immediate threshold of access: recurring elements, partial participation, and the possibility of being used more than once without becoming empty repetition. This reshapes the logic of planning: repetition is not failure to progress but a consolidation. Modularity is not fragmentation, but a condition for inclusion. Brevity is not impoverishment but control of cognitive load. Situated content, in this sense, must not merely mention real-life domains. It must reproduce, in a controlled way, the essential communicative structure of a situation. A lesson about the doctor should include not only body vocabulary, but also greeting, giving one’s name, describing a problem, understanding a question and recognizing a time. A lesson on public services should allow learners to ask for information, identify a document, sign, read a sign or hand in a form [6]. This situated orientation also clarifies why the design pathway proposed in this article cannot be reduced to a generic AI-supported instructional design. In CPIA preA1–A2 contexts, the unit of design is not an abstract topic or a decontextualized skill, but a minimal communicative scenario that must remain usable despite discontinuous attendance, uneven literacy profiles and different degrees of familiarity with written, oral and digital mediation.

2.3. Simplification Without Infantilization

A recurring difficulty in adult migrant literacy concerns the boundary between simplification and infantilization. Texts, instructions and activities must be reduced in complexity, but the person addressed by the material must not be reduced with them. Simplification is a strategy of access, not a judgment on the learner’s cognitive, social or biographical complexity. The principle can be stated plainly: in adult migrant literacy, simplification should not mean infantilization. A text may be linguistically simple and still be adult, dignified and socially meaningful. Conversely, a grammatically correct material may be inappropriate if it relies on childish images, school-like scenarios, paternalistic instructions or representations of migrants as passive, poor, dependent or vulnerable by definition. The point matters because a beginning level in Italian is not a beginning level in life. Many adult migrant learners are workers, parents, carers, community members and people who deal with health, work, finances, documents and family responsibilities [1,2]. Materials must recognize this adulthood even when they use basic vocabulary, model sentences and guided tasks. In design terms, this means that images need to be functional and respectful; situations should belong to adult life; instructions should be clear without being patronizing; examples should avoid stereotyped depictions of migration; and activities should support autonomy rather than mechanical execution. Simplicity becomes a demanding form of mediation: careful selection, reduction and organization of content so that what would otherwise remain inaccessible can be approached without loss of dignity. This has direct consequences for GAI. Systems may quickly produce simplified texts, images or exercises, but they cannot reliably judge whether the result respects the adulthood of the learner [17,21].

2.4. Multimodality as a Design Requirement

Given these premises, multimodality is not an accessory feature of instructional design. Where written competence is fragile, meaning cannot depend on verbal sequence alone. It must be distributed across several semiotic modes: text, image, audio, gesture, layout, color, space, sequence and interaction [13,14]. An effective multimodal material is not a text with decorative images, but an instructional artifact in which each mode performs a specific function. Text introduces words and formulas; images anchor meaning to objects, actions and contexts; audio supports listening, rhythm, pronunciation and orality; layout guides the eye and reduces overload; dialog places language in a communicative relationship; interaction allows recognition to become use. This argument can be situated more explicitly within established theories of multimedia learning. The Cognitive Theory of Multimedia Learning suggests that learning is supported when verbal and pictorial information are meaningfully coordinated rather than presented as unrelated channels [15]. For preA1–A2 adult learners, this means that a written word, an image, an oral model and a guided action should not duplicate one another mechanically, but should converge on the same communicative meaning. Cognitive Load Theory further clarifies why early literacy materials need short sequences, limited information density, clear visual hierarchy and careful pacing [26,27]. When learners are still decoding written language, excessive text, crowded layouts, ambiguous images or rapid audio can impose extraneous cognitive load and reduce access to the task. A dual coding perspective adds another layer: connecting verbal forms with visual representations can support recognition and recall, provided that the association is clear, adult-appropriate and functionally tied to use [32].

For learners at preA1–A2 level, this mediation is decisive. A notice, a form or a short dialog becomes more intelligible when supported by icons, realistic images, highlighted keywords, audio, visual sequences and simulation activities [15,16]. What matters is not the addition of further stimuli, but the design of coherent redundancy: the same content returns through different channels, each time with a precise function. In this sense, multimodal design is not simply a matter of enrichment. It is a way of managing access to meaning, reducing avoidable overload and supporting the gradual movement from recognition to communicative use. This orientation aligns with the affordances of GAI, which can help produce textual, visual and audio variants of the same communicative scenario [19]. A single situation can become a short dialog, a visual glossary, an audio track, a worksheet, a role-play and an interactive activity. The question, however, is not simply what GAI can generate, but how these outputs can be organized into a coherent, accessible, adult and culturally responsible artifact. Recent studies on customizable learning experiences and AI literacy suggest that generative systems may support adaptation and participation, but only when their outputs are embedded in an explicit design logic and reviewed against learner needs, accessibility and context of use [29,30,31,33]. Multimodality is therefore the bridge between the needs of early adult literacy and the methodological proposal developed below.

3. GAI for Multimodal Content Design in Migrant Adult Literacy

Considering the requirements discussed above, GAI can support the design of modular, situated and multimodal content for adult migrant literacy [17,19]. Its relevance is not the automation of writing. It lies in the rapid production of variants of the same communicative core, redistributed across formats, levels and modes. This matters in CPIA preA1–A2 settings, where groups are heterogeneous and attendance may be irregular. A single request for information at a public office can be developed as a short dialog, a simplified text, a list of keywords, a visual glossary, an audio script, a role-play card, a matching activity or a brief formative check. The value of GAI is not any one of these outputs, but the possibility of building materials that remain linked to one another and oriented toward the same communicative task. Multimodal design, then, is not the accumulation of resources. It is the assignment of a clear instructional function to each mode. Every component of the artifact should support access to meaning and gradual use of language in context. Text, image, audio and interaction do not stand side by side as autonomous elements; they operate as a coordinated system of mediation in which each component is linked to a specific learning function and to principles of multimedia learning, including coherence, signaling, segmentation, verbal–visual association and cognitive load management (

Table 1. Multimodal components, learning functions and theoretical rationale in early adult migrant literacy.

Component	Learning Function	Link to Multimedia Learning Principles	Relevance for PreA1–A2 CPIA Learners
Short text	Introduces essential vocabulary, linguistic structures and functional phrases	Supports verbal processing and reduces linguistic complexity	Provides a controlled written entry point for learners with fragile autonomous reading
Image	Anchors meaning to objects, actions and contexts	Supports verbal–visual association and signaling	Helps learners recognize meaning before full written decoding is secure
Audio	Sustains listening, pronunciation, rhythm and orality	Activates the auditory channel and supports dual access to language	Connects written forms with sound, repetition and oral practice
Dialog	Places language within a real communicative situation	Supports coherence between language, action and context	Makes functional language usable in everyday institutional interactions
Visual glossary	Connects word, image, meaning and context of use	Reinforces verbal–visual mapping and recall	Stabilizes key vocabulary without relying only on written explanation
Role-play	Turns recognition into guided communicative practice	Supports transfer from recognition to use	Allows learners to practice formulas in a protected interactional setting
Interactive activity	Supports participation, reuse and formative assessment	Encourages retrieval, segmentation and guided practice	Allows short, reusable tasks suited to irregular attendance and mixed levels

) [13,14,15,16,26,27,32].

This distinction separates a set of materials placed together from a multimodal artifact designed as such. In the first case, resources are simply added. In the second, they converge on one communicative goal. For adult migrant literacy, convergence is crucial: it allows learners to encounter the same communicative content through coordinated verbal, visual, oral and interactive channels, while avoiding both unnecessary overload and empty repetition. The adaptive potential of GAI becomes useful when the same content can be modulated in relation to the actual class. A dialog may be drafted at A1 level and expanded for A2 learners; a text may be reduced to key sentences; a word list may become a visual glossary; an oral exchange may be turned into a simulation card; and an activity may be reformatted as a printable worksheet or a simple digital task. This flexibility answers a concrete need: materials that can be selected, shortened or reorganized without rebuilding the whole unit each time. The potential has limits. GAI can accelerate and diversify content production, but it cannot guarantee linguistic appropriateness, multimodal coherence, cultural sensitivity or instructional relevance [17,20]. A text labeled A1 may still be too complex; an image may be ambiguous or stereotyped; an audio file may be too fast; an instruction may presuppose school practices not yet acquired. Generated outputs should therefore be treated as provisional design material, not as classroom-ready resources. Human validation is part of the method: the teacher or designer defines the communicative goal, checks the language level against the real learners, selects the essential vocabulary, reviews visual and audio components, reduces cognitive load and ensures that the tone remains adult and appropriate to the CPIA context. GAI expands the repertoire of possible materials; turning that repertoire into inclusive multimodal content remains an act of pedagogical judgment. The relevant question is therefore not whether GAI can produce content for adult migrant literacy, but rather under what design conditions that content becomes meaningful in CPIA classrooms. What is at stake is a shift from an output logic to a design logic: from isolated generated materials to multimodal artifacts conceived, selected and reviewed in relation to real communicative needs.

3.1. GAI as an Adaptive Environment for Content Production

Within this framework, GAI can be understood as an environment for adaptive content production. From a multimedia perspective, its distinctive contribution is the transformation of one communicative core into several coordinated formats. Content no longer remains a stable textual unit; it becomes a configuration that can be distributed across expressive modes while retaining its communicative center [13,14]. In early literacy pathways, the same communicative goal may need different versions according to level, time available, group composition and conditions of use. A text can be shortened, expanded, reformulated or contextualized; keywords can anchor the lexical core; audio scripts can be drafted for recording; image descriptions can be prepared with attention to visual clarity. A communicative situation can become a self-contained micro-unit [18,19,33]. What matters is not the multiplication of outputs, but their derivation from a single semantic and pragmatic center. The scenario “Requesting information at a public office” may produce a short reading, a dialog, a visual glossary, an audio track, a worksheet, a role-play card and a formative check. Each element provides a different access point to the same situation. This logic is central to multimedia content production because it shifts attention from the single medium to the relationships among media. Text is not the main content with image or audio appended to it, but it is one component of a composite artifact. A generated text may prepare a dialog; the dialog may become an audio script; the script may guide oral practice; keywords may structure the visual glossary; and the glossary may support interaction. In this sense, GAI works as a device of pedagogical transcoding: it helps content move from one form to another while preserving a recognizable communicative core. The affordances proposed here are not derived from the technical capabilities of GAI alone. They are derived from the intersection between three dimensions: the recurrent communicative needs of CPIA preA1–A2 learners, the generative operations that AI systems can perform on textual, visual, audio and interactive material, and established principles of multimedia learning. From this perspective, variation, reformulation, segmentation, multimodal expansion, recombination, controlled personalization and preparation for review correspond to design operations that support coherence, signaling, redundancy control, cognitive load management, verbal–visual association and guided transfer from recognition to use [13,14,15,16,26,27,32,33]. In operational terms, the main affordances of GAI for multimodal educational content design can therefore be summarized as in

Table 2. Theoretical rationale for GAI affordances in multimodal educational content design.

GAI Affordance	Function in Content Production	Link to Multimedia Learning Constructs	Relevance for CPIA PreA1–A2 Learners
Variation	Producing A1/A2 versions of the same communicative content	Adaptation with semantic coherence	Allows the teacher to adjust the same scenario to uneven learner profiles
Reformulation	Rewriting texts, instructions and explanations in simpler or more functional forms	Reduction in unnecessary linguistic complexity and extraneous cognitive load	Makes institutional or everyday language more accessible without changing the communicative goal
Segmentation	Breaking content into keywords, model sentences, and micro-actions	Segmentation principle and gradual processing	Helps learners approach complex situations through manageable steps
Multimodal expansion	Turning a text into dialog, audio, image, glossary, activity or role-play	Coordination of verbal, visual, auditory and interactive channels	Provides multiple access routes when autonomous reading is fragile
Recombination	Building micro-units from separately generated components	Coherence across media and avoidance of isolated materials	Allows short, reusable units suited to irregular attendance
Controlled personalization	Adapting materials to recurring communicative needs without using personal data	Contextual relevance with privacy protection and controlled redundancy	Connects learning to real-life scenarios without exposing learner biographies
Preparation for review	Producing drafts to be assessed against linguistic, visual, pedagogical and ethical criteria	Quality control before instructional use	Keeps human judgment central in validating level, tone, accessibility and cultural appropriateness

.

These affordances make GAI an environment for instructional pre-production rather than a source of ready-made lessons. Their value depends on how generated outputs are selected and organized in relation to multimedia learning principles. Variation and reformulation support adaptation only when the communicative goal remains stable. Segmentation is useful when it reduces cognitive load rather than fragmenting the task. Multimodal expansion becomes pedagogically meaningful when text, image, audio and interaction converge on the same meaning instead of producing parallel and disconnected materials. Recombination is effective when it creates short, reusable micro-units that can function in fluid CPIA groups. Controlled personalization should refer to recurrent communicative situations, not to identifiable learner data. Preparation for review makes explicit that generated material remains provisional until it has been checked by a teacher or instructional designer. In this sense, the framework is grounded not in the novelty of AI generation itself, but in the alignment between generative operations, multimedia learning principles and the specific constraints of adult migrant literacy. The generated materials remain workable bases: drafts, variants, fragments, descriptions, scripts and activities that must be selected, ordered and reviewed before classroom use. Adaptivity shortens initial production time and widens design possibilities, but instructional value emerges only from intentional organization [33].

3.2. From Monomodal Output to Multimodal Instructional Artifact

In adult migrant literacy, GAI should not be used only to produce simplified texts. Its stronger contribution appears when the same linguistic content is transformed into a coordinated set of multimodal resources organized around one instructional goal [15]. For preA1–A2 learners, written text is a point of arrival as well as a means of access; comprehension is therefore often supported more effectively when content is distributed across text, image, audio, keywords, oral activity and guided assessment. A concrete communicative scenario can become the starting point for a complete micro-unit. Asking for information at a public counter, for instance, can be articulated into a short dialog, a list of keywords, a visual glossary, an audio track, an illustrated sequence, a role-play and a brief formative check. The function of GAI is not to produce many unrelated items, but to help generate resources that hang together around the same communicative scenario.

Table 3. From communicative scenario to coordinated multimodal artifact.

Scenario Element	GAI-Supported Output	Multimedia Learning Rationale	Pedagogical Function in the Micro-Unit
Communicative situation	Short user–officer dialog at a public counter	Coherence principle: all components refer to the same situation	Introduces functional formulas in context
Lexical core	Keywords such as document, form, signature, appointment	Segmentation and signaling	Isolates the essential vocabulary needed for the interaction
Verbal–visual support	Visual glossary with word, image and minimal sentence	Dual coding and verbal–visual association	Connects written form, object/action and meaning
Audio-oral layer	Slowly read dialog and repetition of key phrases	Auditory channel and pacing	Supports listening, pronunciation and oral rehearsal
Action sequence	Illustrated sequence of arrival, request, document, signature and appointment	Temporal organization and cognitive load management	Makes the order of the interaction visible
Guided interaction	Role-play between learner-user and officer	Transfer from recognition to communicative use	Turns comprehension into supported oral practice
Formative check	Word-image matching, sentence completion or guided choice	Retrieval practice and feedback	Checks whether the learner can recognize and reuse key elements

therefore focuses on the transformation of a single communicative situation into a coordinated artifact rather than on general modal functions.

The table clarifies that the artifact is not a collection of media products. Its coherence depends on the relationship among components. The dialog introduces formulas; the keywords identify the lexical core; the visual glossary links words to objects and actions; the audio layer supports pronunciation and listening; the illustrated sequence makes the order of the exchange visible; the role-play guides communicative use; and the formative check verifies recognition and reuse. In multimedia learning terms, the same communicative content is revisited through coordinated channels, with each mode reducing a specific barrier to access rather than simply adding stimulation. This is the point at which GAI-supported production becomes pedagogically relevant: not because it multiplies outputs, but because it helps the designer prepare aligned components that can be selected, simplified and validated for CPIA conditions.

3.3. Reliability and Coherence Checks in GAI-Supported Materials

The capacity to generate texts, images, audio and activities quickly should not be confused with instructional design. In adult migrant literacy, a material may look simple and orderly while still being too difficult, poorly aligned across modes or insufficiently respectful of the learners’ age and cultural plurality. Generated outputs should therefore be treated as provisional design material. The first check concerns ‘linguistic level’. A system may produce a text labeled A1 or A2 while introducing uncommon words, abstract instructions, long sentences or pragmatic structures beyond the learner’s reach [17,18]. The level must be tested against the actual learners, their oral and written competence, their familiarity with school language and the communicative purpose of the task. A further check concerns AI-specific reliability. GAI systems may generate fluent but inaccurate content, including invented procedural details, misleading institutional information, inappropriate examples or plausible but incorrect explanations. In adult migrant literacy, this risk is particularly relevant because learning materials often concern public services, health, school communication, work procedures, documents or appointments. A hallucinated office procedure, an inaccurate form requirement or an unrealistic institutional dialog may not only weaken the lesson, but also create confusion in situations that learners may later encounter outside the classroom. Reliability must therefore be assessed at three levels: factual accuracy, procedural plausibility and contextual appropriateness. Teachers should verify that names of services, documents, actions, times and institutional roles are realistic and, where necessary, locally adapted to the CPIA context. The second check concerns ‘multimodal quality’. The presence of several media does not guarantee accessibility because text, image, audio, layout and activity must converge on the same communicative goal. An ambiguous image, a fast audio track, a confusing sequence or a check that is harder than the material itself can weaken the unit. AI-generated multimodal materials may also contain cross-modal inconsistencies: an image may not correspond to the written word, an audio script may differ from the printed dialog, or an activity may assess vocabulary that has not been introduced. These inconsistencies are not minor formatting problems; they can break the coherence on which multimedia learning depends. A third check concerns the ‘cultural and social positioning of the material’. Simplification can become infantilization; generated images may represent migration through fragility, dependence or marginality [20,21]. Review must therefore ask whether the material is simple but adult, accessible but not paternalistic, and whether it represents the learner as an active subject rather than as a passive recipient of help. To avoid overlap with the broader review checklists presented later in the article,

Table 4. AI-specific limitations and verification checks in GAI-supported multimodal material design.

AI-Specific Limitation	Possible Manifestation in Learning Materials	Required Verification
Hallucinated or fabricated content	Invented procedures, unrealistic institutional details, incorrect document requirements	Check factual accuracy and local plausibility before classroom use
Cross-modal inconsistency	Image, text, audio or activity do not refer to the same word, action or situation	Verify alignment among all components of the artifact
Unreliable level classification	Texts labeled A1/A2 contain uncommon words, long sentences or abstract instructions	Review vocabulary, syntax, sentence length and pragmatic complexity
Content reliability	Explanations sound plausible but are incomplete, misleading or contextually inappropriate	Compare the output with teacher expertise and reliable institutional information
Stereotyped generation	Images or examples depict migrants through fragility, dependence or marginality	Review representations, roles, names and visual framing
Data and privacy risk	Personalized prompts include real biographies, documents or identifiable learner information	Use general scenarios and abstract learner profiles rather than personal data
Low instructional value	Activities are formally correct but do not support comprehension, participation or reuse	Check whether the task serves the communicative goal and the learner profile

focuses only on AI-specific limitations that require verification before GAI-generated outputs can be integrated into a multimodal artifact.

Human review is therefore not a final polish, but the point at which generated output becomes usable instructional content. In Section 3, review is understood primarily as reliability and coherence control: inaccurate details are corrected, cross-modal inconsistencies are removed, and the generated components are realigned with the communicative goal. Broader pedagogical, cultural, ethical and accessibility criteria are then developed in the methodological and ethical sections of the article, so that the review process does not remain a generic checklist but becomes part of the overall design pathway.

4. GAI Tool Categories for Multimodal Design

Designing multimodal content for adult migrant literacy requires a clear distinction between tool categories, semiotic modes and instructional functions. GAI is not a single technology, nor is it limited to conversational LLMs. In this article, it refers to systems used to generate or reorganize text, images, audio, layout and interactive activities [17,18,19]. In CPIA preA1–A2 settings, however, the choice of tool should follow the communicative and pedagogical function of the material, not the novelty of the technology. The designer should first identify the instructional need: simplifying a short text, introducing functional vocabulary, supporting oral comprehension, making a sequence visible, guiding interaction, checking recognition of keywords or preparing a reusable micro-unit. The appropriate tool category follows from that decision.

Table 5. GAI tool categories as function-oriented supports in CPIA preA1–A2 multimodal design.

Tool Category	Main Design Function	Example in a Cpia Micro-Unit	Minimum Human Check
LLMs	Draft short texts, dialogs, instructions, model sentences and guided questions	A short learner–officer dialog for requesting information at a municipal office	Check linguistic level, adult tone and procedural plausibility
Image and visual generators	Produce images, icons, visual sequences or glossary supports	Images for document, form, signature and appointment	Check visual clarity, cultural sensitivity and absence of stereotypes
Text-to-speech and voice-generation tools	Produce audio dialogs, repetition sentences or oral instructions	Slowly read dialog and key phrases for repetition	Check pace, pronunciation, naturalness and alignment with printed text
Presentation and design tools	Organize content into worksheets, cards, visual glossaries or mobile-friendly materials	Printable role-play card and visual glossary	Check layout, font size, contrast, readability and cognitive load
Quiz-generation and interactive tools	Create low-threshold recognition and reuse tasks	Word–image matching or guided choice on appointment time	Check that the activity assesses introduced content only
Assisted review tools	Support re-reading, simplification and consistency checks	Comparing A1/A2 versions or checking alignment across components	Treat feedback as advisory and validate through teacher judgment

Note: The tool-specific checks draw on guidance on Generative AI in education and instructional design [17,18,19], research on AI-generated representations of migrants [21], speech technologies in Italian L2 literacy [23], multimedia learning and Universal Design for Learning [15,16], and WCAG 2.2 [34].

should be read as an operational map rather than as a list of recommended tools. The same communicative scenario may require several tool categories, but each tool is selected for a specific design function. For example, the municipal office scenario may begin with an LLM-generated dialog, continue with a visual glossary, be supported by a slow audio version, and end with a role-play card or a guided recognition task. The quality of the micro-unit does not depend on the number of tools used but on the coherence among the outputs and on their suitability for CPIA preA1–A2 learners. For this reason, the table includes a minimum human check for each tool category.

Table 6. Multimodal variants of communicative scenarios in CPIA literacy contexts.

Scenario	Textual Variant	Visual Variant	Audio-Oral Variant	Interactive Variant
Asking for information at a counter	A1 dialog/A2 dialog	Glossary with images of document, form, signature	Slowly read dialog; sentences for repetition	Role-play; word-image matching; guided questions
Making an appointment	Short text with day and time	Simplified calendar	Listening to a short phone call	Sentence completion; choice among time slots
Describing a symptom	Model sentences: “I have a pain in…”	Body images or functional icons	Oral repetition of phrases	Doctor-patient simulation
Understanding a school notice	Simplified text	Highlighting of date, time and place	Audio recording of the notice	Questions with visual support

illustrates how the same communicative scenario can be developed through different but coordinated variants, without requiring all components to be used in every lesson.

This logic suits non-linear CPIA contexts, where attendance may be discontinuous and learner profiles may vary substantially over time [1,7]. The material is not a closed lesson but a set of selectable components. A teacher may use the dialog and glossary only, add audio if the group needs it, turn the activity into a role-play when oral competence allows, or reduce the unit to a minimal worksheet when time is short. Modularity is therefore not only an organizational necessity; it becomes a design property of the material itself. The caveat remains: generative production does not secure quality. A text may be formally simplified but not genuinely A1; an image may be clear but stereotyped; an audio file may be technically correct but unnatural; a check may be well structured but mismatched to the actual group [17,19,21]. Generative systems therefore provide draft components rather than completed lessons. Their instructional value depends on selection, reduction, integration and validation in relation to the actual group and communicative situation [17,19,21]. Through this mediation, generated outputs can become multimodal instructional artifacts that are linguistically accessible, visually clear, orally workable, culturally respectful and usable in CPIA conditions.

5. Methodological Positioning: A Context-Sensitive, Practice-Oriented Proposal

This article is a methodological contribution oriented to practice. It does not measure the effectiveness of GAI through a controlled experiment, and it does not offer a quantitative evaluation of learning outcomes. Its purpose is to define a context-sensitive design pathway that can be adapted across adult migrant literacy settings rather than mechanically applied, with particular attention to early preA1–A2 pathways in CPIA contexts. The choice follows from the problem itself. The quality of materials in adult migrant literacy cannot be judged only by linguistic accuracy or by impact measured after the fact. It also depends on whether the content is accessible, modular, situated, reusable, respectful of adults and suitable for groups marked by strong heterogeneity and irregular attendance [2,3,28]. Before asking whether AI-supported materials improve learning, it is necessary to ask what design conditions make such materials pedagogically usable. In CPIA settings, this question cannot be separated from the practical conditions in which materials are used. A resource may be linguistically adequate in isolation and still prove unsuitable if it presupposes continuous attendance, stable learner profiles, extended autonomous reading or familiarity with school-based routines. The proposal therefore differs from generic AI-supported instructional design frameworks in both its point of departure and its design constraints. It begins with a recurrent communicative situation and a working profile of the group, rather than with a platform, a tool category or a predefined content format. Its intended outcome is a short, modular and reusable artifact that can remain meaningful when attendance, prior schooling and literacy profiles vary.

The proposal integrates four axes. The first concerns adult migrant literacy and early language learning, especially the relation between language, autonomy and social participation [1,35]. The second concerns multimodal design as the intentional organization of text, image, audio, layout and interaction [13,15]. The third concerns inclusion, accessibility and adult appropriateness, which prevent simplification from becoming poor, childish or culturally reductive [16]. The fourth concerns the critical use of GAI as an adaptive production environment requiring pedagogical, cultural and ethical verification [17,19,20]. These axes are not treated as separate layers. Together, they guide a sequence of design decisions that links a communicative need to the conditions under which a material may become usable in a particular CPIA classroom. Generative tools change quickly and depend on versions, access conditions, supported languages and terms of use. For that reason, the article focuses on criteria and design steps that can be applied across categories of tools: language models, image generators, text-to-speech systems, design environments and interactive tools. The focus shifts from the tool to the ‘process’. More specifically, the contribution lies in making visible the decisions that connect an everyday communicative need to a viable instructional artifact: defining the scenario, identifying functional vocabulary and language acts, generating provisional textual, visual and audio materials, designing brief guided interaction, and reviewing the result for accessibility, adult appropriateness, contextual plausibility and ethical soundness.

The sequence is offered as an open methodological pathway, not as a definitive model. It does not prescribe a universal order of software actions. Instead, it provides a shared frame through which teachers and instructional designers can adapt design choices to actual classroom conditions. This practice-oriented stance addresses a frequent gap in discussions of AI in education: the passage from technical possibility to instructional sustainability. The proposal is not concerned with the capacity of a system to generate content in isolation. It concerns the conditions under which generated drafts can be transformed into materials that are usable in early adult migrant literacy. These conditions include a stable communicative goal across modes, the possibility of partial participation, and the capacity to reconfigure materials without rebuilding an entire sequence. The proposal is illustrated through an application rather than tested through an experiment. It shows how the pathway can be translated into a multimodal micro-unit connected to an everyday communicative need. Its purpose is to make the design logic transparent, not to claim evidence of effectiveness. The contribution of the present article rests on methodological transparency, coherence between the educational problem and the proposed design response, explicit quality criteria, and the possibility of adapting the pathway to comparable adult migrant literacy settings.

6. Methodological Pathway: From Communicative Need to Multimodal Learning Artifact

The proposal takes the form of a structured but non-linear design pathway. Its purpose is to transform a recurrent communicative need into a multimodal instructional artifact that can be used under the variable conditions of adult migrant literacy at preA1–A2 levels. The pathway combines principles of adult migrant literacy, multimodal design, inclusive design and the critical use of GAI. It is intended to produce materials that are short, modular, situated, linguistically accessible, appropriate for adult learners and culturally responsible [1,2,13,16,17,20]. GAI supports the production of drafts and variants; the pathway specifies how those materials are selected, adapted and prepared for use in a particular CPIA group. The sequence is not intended as a fixed instructional script. It is a set of connected design operations that begins with a communicative need rather than a technological function and proceeds through textual, visual, audio-oral and interactive layers. What makes the pathway specific to CPIA contexts is the need to preserve an immediate entry point for learners with uneven literacy profiles, discontinuous attendance and different degrees of familiarity with written, oral and digital mediation [2,3,6,22,28].

Table 7. CPIA-specific phases of multimodal instructional design with GAI support.

Phase	Design Operation	CPIA-Specific Design Question	Expected Output
1. Working learner profile	Identify oral communication, literacy and access conditions without relying on a level label alone	What forms of access, participation and support are realistic for this group?	Non-identifying working profile and access map
2. Communicative scenario	Select a narrow, recurrent situation from adult everyday life	Which situation is immediately recognizable and usable even by learners who have missed earlier lessons?	Situated communicative scenario
3. Linguistic core	Generate and simplify texts, dialogs, keywords and instructions	Which words, formulas and actions are indispensable to the scenario?	Short textual core of the micro-unit
4. Visual scaffolding	Add images, icons, visual glossaries and action sequences	Which visual elements clarify meaning rather than merely decorate the page?	Functional visual layer
5. Audio-oral support	Produce paced support for listening, pronunciation and oral rehearsal	What should be heard, repeated or recognized before learners are expected to read independently?	Audio-supported oral practice
6. Guided reuse	Design low-threshold tasks, simulations or role-plays	How can learners reuse the content through partial participation and guided interaction?	Matching activity, micro-task or role-play
7. CPIA readiness review	Assess the instructional usability of the assembled artifact	Can the unit work with mixed levels, irregular attendance and limited autonomous reading?	Classroom-ready multimodal artifact
8. Documentation, adaptation and reuse	Record the design choices and prepare adaptable components	What should be retained so that the material can be revised or reused in another CPIA group?	Traceable and reusable material set

outlines the pathway.

The first phase establishes a working learner profile. In CPIA settings, a formal A1 or A2 label is not enough, since learners placed at the same level may differ sharply in oral comprehension, reading, writing and pragmatic use of language [3,6]. Design should begin from a concise and non-identifying map of relevant conditions: familiarity with alphabetic writing, previous schooling, ability to use written materials, access to digital devices, need for printable supports, previous attendance and urgent communicative needs [2,3,22]. This is not a diagnosis and does not require the collection of personal biographies. It is a provisional working description that helps the teacher decide which modes, supports and activities are likely to offer a realistic point of entry. The second phase selects a real communicative scenario. Content should come from a recognizable situation rather than an abstract language category: asking for information at a public office, making a medical appointment, reading a notice, filling out a form, using public transport or understanding safety instructions [10,12]. The scenario should be narrow enough to work. “Public services” is too broad; “asking at the counter which document is needed” is usable. This decision fixes vocabulary, dialog, images, audio and the final activity. For CPIA preA1–A2 learners, the scenario should also permit partial participation. A learner who joins the class after an absence should still be able to recognize a keyword, follow a visual sequence, repeat a short formula or take part in a guided exchange without having completed every prior activity. The third phase builds the textual layer. A language model may draft a dialog, short text, keywords, useful phrases, instructions or comprehension questions [18]. The output is not final. It must be reduced and checked: short sentences, concrete vocabulary, few subordinate clauses, predictable sequencing, functional repetition and adult tone. The same situation may be written in an A1 version and in a slightly expanded A2 version. This textual core then supports the other modes. The fourth phase introduces the visual layer. Images, icons, visual glossaries, flashcards and illustrated sequences must clarify words, actions and steps. If the unit uses document, form, signature and appointment, the visual layer should help learners distinguish these concepts and connect them to use. Image selection requires care: visuals should be clear, appropriate for adult learners and free from stereotyped representations of migration or of learners as passive, fragile or dependent [16,21]. The fifth phase concerns audio and oral work. At preA1–A2 level, audio is not an optional add-on. Some learners understand more easily through listening than through autonomous reading [15,23]. Text-to-speech or voice-generation tools can provide slow dialogs, repetition sentences, keywords, oral instructions or short listening activities. The audio must be checked for pace, naturalness and coherence with the written material. Its function is to connect written form, sound and oral use. The sixth phase turns the material into an activity. A multimodal artifact should not only present content; it should guide the learner toward use. From text, image and audio, the designer can build word-image matching, sentence completion, guided multiple choice, pair work, simulations or role-plays. The aim is not immediate free production, but the controlled reuse of meaningful words and formulas. The task should remain workable even when some learners can only point, match, repeat or select a response, while others are ready to sustain a short oral exchange. The seventh phase is a CPIA readiness review. It begins only after the AI-specific checks presented in Table 4 have been completed. Table 4 concerns the dependability of generated components, including factual accuracy, cross-modal consistency, level classification, stereotyped generation and data risks.

Table 8. CPIA readiness review for a GAI-supported multimodal artifact.

Review Dimension	CPIA-Specific Control Question	Revision Action When Needed
Immediate entry and re-entry	Can a learner who missed previous lessons still recognize at least one word, image, action or model sentence?	Add a visual cue, keyword, model sentence or brief oral prompt
Reading and processing demand	Does the task require more autonomous reading, memory or inferencing than the target group can reasonably manage?	Shorten the text, reduce the number of steps, highlight keywords or introduce visual support
Guided participation	Can learners participate through pointing, matching, repeating or selecting a response before they are expected to sustain an oral exchange?	Add a low-threshold task, model dialog or structured response option
Instructional sequencing across modes	Can learners move from text, image or audio to the activity without losing the communicative thread?	Re-sequence the components and make the transition between modes more explicit
Adult relevance	Does the material address an adult social practice without becoming paternalistic, school-like or detached from everyday needs?	Revise examples, images, instructions or task framing
Practical accessibility	Can the material be used in print and, where relevant, on a mobile device with readable layout, adequate contrast and manageable audio pace?	Adjust layout, font size, spacing, contrast, audio speed or delivery format
Modular adaptability	Can selected components be used independently when time is short, attendance is irregular or the group is uneven?	Reorganize the unit into shorter, self-contained components

addresses a separate question: can verified components function as a usable learning artifact in a heterogeneous CPIA classroom? The focus is therefore on access to meaning, partial participation, adult relevance, practical accessibility and modular adaptation.

This review is not a second AI-risk checklist. It is a final check of classroom readiness. A dialog may be linguistically accurate but still be unusable if it requires learners to remember too many steps; an audio file may be technically clear but too fast for a first exposure; a role-play may be relevant but inaccessible to learners who cannot yet read the prompt independently. The review therefore identifies which component should be reduced, re-sequenced or supported before classroom use. The eighth phase concerns documentation, adaptation and reuse. A practice-oriented proposal requires traceability, not because every material must become a formal research dataset, but because teachers and designers need to know how a resource was produced, revised and adapted over time [19]. Documentation supports transparency, makes revision possible and helps distinguish generated drafts from the pedagogical decisions that shaped the final artifact (see

Table 9. Documentation for adaptation and reuse in GAI-supported CPIA materials.

Documentation Element	What Should Be Recorded	Why It Matters
Starting scenario	The selected communicative situation and its intended adult social practice	Clarifies the practical need addressed by the unit
Working group conditions	Relevant literacy, participation, print and device-access conditions	Supports adaptation without collecting unnecessary personal data
Tool context	Tool category, access conditions and, where relevant, version or language settings	Makes the production context transparent
Prompt and input choices	Prompt structure, constraints and key instructions used to generate drafts	Supports comparison and future revision
Generated components	Draft texts, images, audio scripts or activities considered during design	Preserves the relationship between initial outputs and final materials
Human revisions	Changes made to language, visuals, audio, sequencing or task design	Makes pedagogical judgment visible
CPIA adaptation notes	Components retained, omitted or modified for a particular group	Supports reuse under different attendance and literacy conditions
Identified limits	Uncertain local details, accessibility constraints or components requiring later verification	Encourages cautious reuse and further improvement

).

The pathway ends with a traceable and adaptable material set: for example, a printable worksheet, a dialog with audio, a visual glossary, role-play cards or a short unit composed of selectable components. In CPIA settings, reusability does not mean that the same material is delivered unchanged. It means that its communicative core can be retained while the level of support, the sequence of activities and the selected modes are adjusted to the group in front of the teacher. The expected result is not a finished output generated by AI, but a modular instructional artifact designed for use, revision and responsible reuse in adult migrant literacy work.

7. Illustrative Application: Designing a Multimodal Micro-Unit for preA1–A2 Adult Learners

This section provides an illustrative application of the proposed pathway. It shows how a recurrent communicative need can be developed into a small set of coordinated materials for adult migrant learners at preA1–A2 level. The selected scenario concerns requesting information at a municipal office, a recurring situation in adult migrant literacy in which language supports access to services and everyday autonomy [2,10]. The micro-unit is designed for learners whose oral and written competences may not align and who may enter or re-enter the sequence at different moments [6]. It therefore concentrates on a limited set of high-frequency words and formulas that can be encountered through text, image, audio and guided interaction. The aim is not to teach administrative language in general, but to support a first exchange at a counter: greeting, requesting information, showing a document, recognizing a form, signing and identifying an appointment.

Table 10. Design profile of the illustrative multimodal micro-unit.

Design Element	Operational Choice
Scenario	Asking for information at a municipal office
Target group	Adult migrant learners at preA1–A2 level
Communicative goal	Understand and take part in a brief exchange at a municipal office counter
Key vocabulary	‘Ufficio, documento, modulo, firma, appuntamento’ (office, document, form, signature, appointment)
Format	Printable worksheet, visual glossary, slow audio, role-play and brief formative check
Indicative duration	30–45 min, reusable across lessons
Re-entry option	Visual glossary, model dialog and audio can be used independently by learners returning after an absence

presents the design profile of the micro-unit.

The process begins with a short textual core. A language model may be used to prepare a draft dialog, which is subsequently reduced and reviewed for sentence length, vocabulary, pragmatic plausibility and adult tone [18]. Useful formulas include Vorrei informazioni, Ecco il documento, Devo firmare qui? and Quando è l’appuntamento? The same core can be expanded into a slightly more articulated A2 version without changing the communicative situation. The following materials show how this limited textual core can be made available through coordinated access points [13,15]. Rather than repeating the general functions of textual, visual, audio and interactive layers already discussed in Table 3,

Table 11. Illustrative sample materials for the municipal-office micro-unit.

Component	Illustrative Sample Material	Function in the Micro-Unit	Accessible Participation Route
Short dialog	Officer: Buongiorno. Learner: Buongiorno. Vorrei informazioni. Officer: Certo. Ha un documento? Learner: Sì, ecco il documento. Officer: Deve compilare il modulo e firmare qui. Learner: Devo firmare qui? Officer: Sì. L’appuntamento è lunedì alle dieci.	Introduces the communicative sequence and the core formulas	Listen to the dialog, point to a word card or repeat one formula
Visual glossary	Illustration of a generic identity document—Documento: Ho un documento. Illustration of a generic form—Modulo: Compilo il modulo. Illustration of a signature line—Firma: Firmo qui. Illustration of an appointment slip or calendar—Appuntamento: Ho un appuntamento lunedì.	Links written form, image, action and minimal sentence	Match a word to an image, point to the relevant card or repeat one word
Audio-oral support	Slow recording of the dialog, followed by short pauses after: Vorrei informazioni… Ecco il documento… Devo firmare qui…	Supports listening, pronunciation and oral rehearsal [24]	Listen, repeat one phrase or identify a word heard in the recording
Role-play card	Learner-user: greet; ask for information; show the document; ask where to sign. Learner-officer: greet; ask for the document; indicate the form; state the appointment.	Moves from recognition to guided oral use	Use word cards, images or model sentences while speaking
Brief formative check	Match documento, modulo, firma AND appuntamento to four images. Then choose the correct response to: Quando è l’appuntamento?	Checks recognition and limited reuse of the key vocabulary	Complete only the matching task or answer through a guided choice

presents a small set of sample materials that learners may actually encounter and use during the micro-unit.

The materials are designed as coordinated entry points rather than as a sequence that every learner must complete in the same order. The visual glossary supports recognition of the words that recur in the dialog and role-play. The audio component makes the same formulas available through paced listening and repetition. The role-play then allows learners to move from recognition to supported oral use. A learner who cannot yet read the whole dialog independently may still participate by identifying an image, selecting a word card, repeating a short formula or responding to a guided prompt.

The micro-unit is deliberately modular. In one lesson, the teacher may use only the dialog and visual glossary; in another, the audio and matching activity may be added; in a later session, the same content may support a short role-play. This arrangement allows learners returning after an absence to re-enter through a recognizable component rather than being excluded by a missed sequence. Before classroom use, the material should undergo the CPIA readiness review presented in Table 8, with particular attention to access to meaning, adult relevance, practical accessibility and the possibility of partial participation. The illustrative case therefore shows how a recurrent communicative situation can be developed into a coherent but adaptable set of materials. Its purpose is to make the operational consequences of the proposed pathway visible: a short textual core can be expanded into linked visual, audio-oral and interactive resources, then selected and adjusted in relation to the group present in the classroom.

8. Ethical, Cultural, and Pedagogical Considerations

The use of GAI in adult migrant literacy cannot be treated as a technical extension of material production. Generated content is not neutral. It can carry representations of the learner, implicit models of participation, cultural assumptions and symbolic hierarchies beneath the surface of the artifact [17,20]. In CPIA settings, quality must therefore be assessed not only by readability or formal correctness, but also by dignity, contextual relevance and pedagogical responsibility.

A first issue concerns linguistic simplification and symbolic reduction. Early literacy materials rightly use short sentences, limited vocabulary, visual support and guided activities. But simplification must avoid childish phrasing, paternalistic instructions and scenarios that make the adult learner appear socially or cognitively diminished. The learner may be a beginner in the language of schooling, but not in life experience, responsibility or practical reasoning [1,2].

The visual dimension requires similar attention. Image-generation systems may reproduce stereotyped imaginaries of migration, linking migrant subjects with vulnerability, poverty, passivity or institutional dependence [21]. The problem is not only openly offensive imagery but appears when migrants are consistently shown as disoriented users or recipients of aid, and rarely as workers, parents, citizens, competent adults or active social subjects. Graphic choices are therefore ethical choices, not merely esthetic ones.

Cultural assumptions also need review. Dialogs and scenarios generated by GAI may presuppose familiarity with institutional routines, health practices, school communication, administrative procedures or social norms that learners do not necessarily share [2,35]. What is obvious from the standpoint of the host context may be opaque to those who come from different educational or institutional systems. Responsible design must avoid both excessive simplification and excessive presupposition.

Accuracy becomes especially sensitive when materials deal with health, work, safety, documents, school or public services. In these domains, hallucinated or fabricated content is not merely a technical defect. An invented procedure, an incorrect document requirement or a misleading appointment instruction may compromise a learner’s autonomy and confidence in situations encountered beyond the classroom. The reliability controls outlined in Table 4 should therefore be applied before use, with attention to factual accuracy, local procedural plausibility and contextual appropriateness [17,20].

Privacy and data protection belong to the design itself. Personalization should not mean entering real names, recognizable stories, administrative conditions, health data, family information or classroom episodes into generative systems [17,20]. In migrant education, even apparently minor details may become sensitive when combined. Teachers and designers should therefore work with abstract profiles and recurrent communicative situations, minimize the information included in prompts, avoid uploading learner documents or identifiable images, and follow local institutional procedures when a tool involves accounts, cloud storage or shared workspaces.

Copyright, ownership and transparency require comparable care. A generated text, image or audio file should not be assumed to be free from contractual, licensing or copyright-related constraints. The status of an output may depend on the tool, its terms of use, the materials supplied as inputs and the applicable jurisdiction. Designers should avoid requesting reproductions of protected textbooks, worksheets, images or recognizable branded materials; check the relevant licenses and institutional policies; and record the tool category, prompt constraints, source materials and substantial human revisions for any output retained in the final artifact [17]. Within this framework, explainable AI practices do not require teachers or instructional designers to explain the internal architecture of a model. They require them to account for the educational choices surrounding an output. A teacher or designer should be able to explain why a tool was used, which inputs shaped the draft, which elements were accepted, altered or rejected, and why the final material is appropriate for a particular group.

The documentation process described in Table 9 supports this form of pedagogical explainability and makes AI-supported production traceable without presenting generated content as a neutral or authoritative source [19,20]. Multimodality does not guarantee accessibility; simplification does not guarantee comprehension; representation of diversity does not guarantee respect. Inclusion is not a surface property of the product, but the result of controlled design and ethical validation. It must be pedagogically designed, culturally reviewed and made accountable throughout the production process. Recent work on GAI practices, literacy and digital divides in the Italian context suggests that these risks are not merely theoretical [29]. Uneven access to tools and unequal capacity to assess outputs can intersect with social vulnerabilities already present in learners’ everyday environments [30]. The ethical task is therefore not limited to correcting isolated errors at the end of production. It is to maintain dignity, autonomy, fairness, privacy and accountability throughout the design process, so that GAI-supported materials remain accessible, respectful and meaningful within CPIA settings [19,20,29].

9. Discussion

The proposal developed in this article addresses a question broader than the production of teaching materials. In adult migrant literacy, the issue is not only whether content is available, but whether it can function in educational environments characterized by heterogeneous literacy profiles, discontinuous attendance and variable access to written, oral and digital mediation [2,3,6,22,34]. The framework therefore treats GAI as a resource for preparing draft components within a context-sensitive design process.

Its first contribution concerns the status of instructional content. In CPIA settings, material cannot be treated as a closed unit intended for a stable group progressing through a predictable sequence. It needs to remain modular, so that it can be reactivated, reduced, expanded or reorganized in relation to the group present on a given day [1,7]. A dialog, for example, may be developed into a worksheet, a visual glossary, an audio track, a role-play card or a brief formative check. The point is not to multiply materials, but to construct coherent sets that can be adapted without losing their communicative core. This position distinguishes the proposed pathway from generic AI-supported instructional design approaches. Existing work on GAI-supported education has explored adaptation, inclusive design and customizable learning experiences [17,18,19,33]. The present framework differs in that it begins from a recurrent communicative need and from the practical conditions under which learners can access and reuse a material. The present framework begins instead from a recurrent communicative need and from the practical conditions under which learners can access and reuse a material. Its distinctive features are therefore not the use of GAI alone, but the attention to partial participation, re-entry after absence, modular adaptation and the relationship between literacy, communicative action and adult social life. Multimedia learning principles help explain how modes should be coordinated [13,14,15,16,27,28,35], while the CPIA perspective specifies why this coordination must remain adaptable to uneven literacy profiles and non-linear participation.

A second implication concerns multimodality. In adult migrant literacy, coordinated textual, visual, audio-oral and interactive resources can create different routes into the same communicative situation. Learners are developing oral, written, pragmatic and functional competences at the same time; meaning cannot rest on written words alone [11,13]. Images, audio, layout, visual sequences, keywords and interaction redistribute the interpretive load and open access even when decoding remains partial [15,16]. This redistribution is cognitive, but also social. A learner may recognize a document through an image, hear a formula, repeat it in a guided dialog, connect it to a written word and use it in a simulation. Language is learned here not as an abstract object, but as participation in recognizable social situations [9,10,12]. Multimodality mediates between literacy, communication and autonomy.

A third implication concerns simplification. For adult learners, simplifying does not mean reducing experience or social identity. It means building a more precise relationship between language, context and action. GAI may rapidly generate short and formally accessible texts, but their educational value depends on whether they remain realistic in situation, appropriate in tone and usable for the intended group [17,21]. The relevant question is therefore not whether a generated output appears fluent or simple, but whether it supports an adult learner in understanding and acting within a recognizable social situation.

This has direct implications for teachers. The framework encourages teachers to begin from a narrow communicative scenario and to select only those components that support the group present in the classroom. A visual glossary, a short model dialog, a slow audio track or a guided matching task can provide an entry point for learners who return after an absence or who are not yet ready for extended reading or oral production. Materials can therefore be used selectively, without requiring every learner to complete the same sequence in the same order.

For instructional designers, the framework suggests moving from closed learning packages to adaptable material sets. Rather than preparing a single linear lesson, designers can develop related components around one stable communicative core: a simplified dialog, a visual glossary, an audio script, a role-play card and a brief formative check. This approach preserves coherence while allowing changes in the level of support, the order of activities and the selected modes.

There are also institutional implications for CPIA coordinators and policymakers. Responsible GAI-supported design requires more than access to digital tools. It requires guidance on tool selection, data minimization, documentation, copyright awareness, review procedures and the practical conditions under which generated materials can be adapted and reused. In this respect, documentation is not only an administrative task. It is part of the pedagogical and ethical infrastructure that makes the design process transparent, revisable and accountable over time [19,20]. The contribution of the proposed framework therefore lies in connecting a situated communicative need with a modular process of multimodal design. Its value is not the rapid generation of content as such, but the possibility of producing coherent and adaptable materials for adult migrant literacy under real CPIA conditions. Adaptivity, multimodal coordination and responsible review are not separate objectives. Together, they support access to language as a social practice, rather than as a purely school-based subject [9,11].

10. Opportunities, Limitations, and Future Research

The framework proposed in this article identifies several opportunities for GAI-supported design in adult migrant literacy. Its main potential lies in supporting the preparation of short, modular and multimodal materials that can be adapted to recurring communicative needs and to the variable conditions of CPIA classrooms. In particular, the pathway may help teachers and instructional designers prepare related versions of the same communicative core, such as a simplified dialog, a visual glossary, an audio-supported activity, a role-play card and a brief formative task. This may reduce some of the initial workload involved in adapting materials for groups characterized by uneven literacy profiles, discontinuous attendance and different forms of participation.

These opportunities should be interpreted in light of important limitations. First, the present article is a methodological and practice-oriented proposal. It does not report an empirical intervention, a controlled comparison or evidence of learning gains. The illustrative application demonstrates how the pathway can be operationalized, but it does not establish whether GAI-supported micro-units improve comprehension, participation, oral use, learner confidence or access to services. Nor does it provide direct evidence of how adult migrant learners perceive the materials, the modes of participation they prefer, or the difficulties they may encounter when using them.

A second limitation concerns context. The framework was developed with particular attention to Italian CPIA settings and to adult migrant learners at preA1–A2 level. Although several of the proposed principles may be relevant to other adult education environments, their transferability cannot be assumed. Institutional arrangements, language policies, digital access, learner trajectories and the social meaning of everyday communicative situations may differ substantially across countries and educational systems. The pathway should therefore be adapted rather than transferred unchanged to other contexts.

A further limitation concerns the rapid evolution of GAI technologies. Tool capabilities, interface design, supported languages, pricing models, data practices and terms of use may change quickly. The article therefore focuses on design principles and review procedures rather than recommending specific platforms or treating current technical features as stable. It also does not offer a comparative evaluation of individual tools, models or commercial services. This choice supports the durability of the methodological proposal, but it necessarily limits the level of technical detail that can be provided.

Future research should test the framework in real CPIA classrooms through multi-site and mixed-methods studies. Relevant evidence could include classroom observations, teacher design logs, interviews with teachers and learners, analysis of produced materials, and measures of usability, participation, comprehension and guided oral interaction. Studies could examine whether modular materials support re-entry after absence, whether visual and audio components offer meaningful access for learners with different literacy profiles, and whether the pathway remains manageable for teachers working under ordinary time constraints.

Comparative research would also be valuable. Future studies could compare micro-units designed through the proposed pathway with materials developed through more conventional processes, while avoiding simplistic assumptions that any difference is attributable to GAI alone. Attention should be given to the quality of the communicative scenario, the appropriateness of the materials for adult learners, the accessibility of individual modes, the cultural plausibility of generated content and the role of teacher mediation. Longitudinal research could further investigate whether documentation, adaptation and reuse make the pathway sustainable across groups, learning cycles and institutional settings.

Finally, future work should examine the ethical and institutional conditions required for responsible implementation. This includes the practical effectiveness of data-minimization procedures, copyright and licensing awareness, transparency in the use of generated outputs, and the extent to which teachers and learners can understand and challenge the choices embedded in AI-supported materials. Empirical research in these areas would help refine the framework and clarify under which conditions GAI can contribute to adult migrant literacy without reproducing inequalities, cultural simplifications or forms of technological dependence.

11. Conclusions

This article has proposed a practice-oriented pathway for designing GAI-supported multimodal content for adult migrant literacy, with particular attention to preA1–A2 learners in Italian CPIA settings. The proposal responds to a recurring design problem: how to prepare materials that remain usable when literacy profiles are uneven, attendance is discontinuous and access to written, oral and digital mediation varies within the same group. The contribution of the framework lies in shifting attention from isolated generated outputs to modular instructional artifacts organized around recurrent communicative situations. A short dialog, for example, can be developed into a visual glossary, an audio-supported activity, a role-play card and a brief formative task, provided that these components remain aligned with the same communicative purpose. This makes it possible to offer different entry points into the material without reducing the learner to a fixed level label or requiring every participant to follow the same sequence. Within this perspective, GAI can support the preparation of drafts, variants and multimodal components, but educational value emerges through the design process that shapes them. The pathway makes this process visible through the definition of a working learner profile, the selection of a situated communicative scenario, the coordination of textual, visual, audio-oral and interactive resources, the review of classroom readiness and the documentation of revisions for adaptation and reuse. The framework is not intended as a fixed model or as a substitute for pedagogical expertise. It offers a set of criteria through which teachers and instructional designers can transform generated material into resources that are linguistically accessible, appropriate for adult learners, culturally responsible and adaptable to real CPIA conditions. In this sense, the value of GAI-supported design lies not in the volume of content produced, but in the possibility of constructing coherent materials that support adult migrants’ access to language as a social practice.