As generative AI continues to expand its presence in education, designing for equity has never been more urgent. At WGU Labs, we believe inclusion begins not just with access to content, but with how that content is delivered. In previous research, we have explored aspects of student belonging within the online learner context. More specifically, we have found that online learner belonging is often derived from institutional outreach and interventions, rather than peer interactions, which often inform belonging in traditional in-person environments.
One area of inquiry we have recently explored related to belonging is the role of language within the student experience, and how language must be considered when leveraging AI to enhance learning. Language is not neutral. It encodes power, identity, and belonging. In the hands of AI, it can either reinforce dominant norms or radically expand who feels they belong in learning environments.
That’s why we’ve been experimenting with how AI can recognize the different ways we use language based on region, or regional variants. We hypothesize that a method is needed to prompt generative AI to accurately interpret and produce responses tailored to the language styles, expressions, and educational realities of specific communities. Further, this exploration seeks to build upon Labs’ existing research on belonging within the online adult learner environment, to better understand how regionally-specific language might support learner success. What began as an exploration of U.S. regional differences has evolved into a deeper reflection on how AI can support linguistic diversity as a lever for educational access.
Why Sociolinguistics Matter: Language, Power, and Belonging
Sociolects (language varieties associated with social groups), regional dialects, and even generational communication styles all play a role in shaping identity and perceptions. Yet education systems have historically privileged certain forms of speech, largely Standard American English, professional tone, and middle-class academic speech, while marginalizing others.
This is not a coincidence; it is structural. The ability to code-switch (navigate multiple linguistic codes) has long been a hidden curriculum of access to power. Students who speak in non-dominant codes are often implicitly or explicitly told to “speak properly” to succeed, even if non-dominant ways of speaking are entirely valid, rule-governed, and expressive of rich cultural heritages.
This linguistic conformity exacts a toll, not just academically, but psychologically. It reinforces the idea that to belong, one must assimilate.
In order to explore the challenges this structure creates within the learning system for marginalized students, we explored this question: What if our system met learners where they are to ensure they feel a sense of belonging, rather than requiring them to question their identities through language?
The Sociolects Approach
To conduct this inquiry, we developed a structured prompt for generative AI that asks it to localize responses across U.S. regions. The prompt can decode and reflect the learner’s sociolect and either mirror or respond in a scholarly tone based on user preferences.
The outputs were then stored in standardized JSON files, allowing educators and learning experience designers to access and implement them at scale across various platforms. These files can be used for a variety of things, including:
- Localizing curriculum content
- Training faculty on linguistic diversity
- Developing inclusive, responsive chatbots
- Offering regionally resonant student support services
But the potential goes beyond regions.
We also explored how this same method can surface differences across myriad applications:
- sociolects (e.g., casual vs. academic English),
- generational styles (e.g., Gen Z slang vs. Boomer formality),
- international English variants (e.g., used in the U.S., U.K., India, Nigeria).
In each case, the goal is the same: to affirm linguistic diversity as a strength, not a deficit.
The Process: From Prompt to JSON
The experiment began with a structured prompt delivered to a generative AI (in this case, ChatGPT), asking it to localize responses across a range of regions. Once the AI had generated the region-specific responses, each variant was saved in a standardized JSON format, making the content machine-readable and easy to sort and scale. Once the variants were uploaded to the agent as support files, the AI agent guided users to help with:
Student Onboarding
"Would you like to set your communication preferences?"
- Tone: Chill, Formal, Flexible
- Language Mode: Native / Fluent / Learning English
- Cultural Setting: Global | U.S.-Based | UK-Based | Other
- Generational Style: Boomer | Gen X | Millennial | Gen Z
Real-Time Input Analysis
- Detect language complexity, sociolect, slang
- Detect potential regional English markers
- Detect generational tone if not set manually
Tonal + Cultural + Generational Adaptive Response
- Tailor what is said (content and context)
- Adapt to how it is said (formality, emojis, directness)
Scaffold Academic Moves Transparently
"In U.S. academic essays, it's normal to start with a clear thesis. Want me to help you build one?"
Ongoing Flexibility
- Always offer a "Change Style" button in the UI.
- Allow students to toggle between Casual, Academic, Gen Z, Formal, etc.
In addition to prompting the user to provide language-informing information to the bot, and files automatically instructed the agent on how to understand regional differences in language used by the respondent, and how to respond in kind.
What We’ve Learned So Far
A few early takeaways stand out:
- The process is replicable. Because the system produces standardized JSON files, it can be scaled to include additional regions, languages, or even thematic variants (e.g., for rural vs. urban districts).
- Bias still requires human oversight. While generative AI is powerful, its responses must be audited by experts to ensure accuracy and avoid reinforcing stereotypes. The JSON format enables easy flagging and editing of entries, supporting continuous refinement, but it is only a tool, not the end result. And the design can make mistakes.
As the use of AI within the learning experience becomes more widespread, we have an opportunity to rethink how we are creating belonging and inclusion so that all learners can thrive. Beginning with the way we communicate, and the language we use, is an important first step to creating new systems that prioritize the needs of all learners.
Discover how WGU Labs is advancing AI and design to create more inclusive, learner-centered solutions.
