Generative Artificial Intelligence as an Educational Collaborator: A Next-Generation Resource

Ms. Smith is a busy special education teacher in her first year at a high-needs elementary school with limited resources. The school’s existing reading resources for the general and special education classrooms do not provide enough practice opportunities for her students. This causes her to spend hours outside of school creating or finding additional activities, such as additional practice word lists, and adapting reading passages to students’ reading level. Although Ms. Smith enjoys teaching her students and creating the activities, she wishes there was a time-saving way to do so that also aligns to her teaching pedagogy, students’ instructional needs, and their learning objectives. As Ms. Smith considers how to develop activities to provide additional practice reading consonant blends for her ongoing unit, she remembers another teacher, Ms. Martinez, describing how they are using artificial intelligence (AI) to create math instructional materials. Ms. Smith considers how AI can also be applied to developing materials for reading.

“Much like collaborators at school, such as another teacher, AI can be used to generate an additional perspective.

“For special education teachers, AI can act as an electronic collaborator to generate tailored content for more personalized learning for their students with disabilities.

The application of AI is rapidly becoming commonplace in our daily lives and in education. AI refers to the use of a technology-based tool to “automate reasoning based on associations in the data (or, associations deducted from expert knowledge)” (Office of Educational Technology [OET], 2023b, p. 1). AI is a wide variety of technology-based tools that use algorithms to recognize patterns in how two concepts are related and then responds to those in an efficient, quick manner. For example, AI might be virtual assistants—tools to support searching for and synthesizing information online, automated chatbots, and speech recognition tools (OET, 2023b). Alexa and Siri are common virtual assistants that answer in response to a verbal question or prompt, such as reporting the weather, giving driving directions, sending a text message, playing a video, adding an item to a shopping list, or other functions. AI is on many devices and accessible through different means (e.g., embedded function in a smartphone, tablet, or computer or standalone devices like an Amazon Echo).

In education, AI offers tools to K–12 students and their teachers that can support learning and teaching (Hopcan et al., 2022; Zafari et al., 2022). AI can be used to develop instructional tasks, such as those to assess students’ learning, support grammar or word choice when writing, or allow students to participate in personalized or intelligent learning (i.e., instruction tailored to students based on their performance; Chen et al., 2022). However, AI will not replace the teacher. Humans, such as special education teachers, must remain central to education systems (OET, 2023b). However, AI could add to the teacher’s array of resources. Much like collaborators at school, such as another teacher, AI can be used to generate an additional perspective.

Teachers often turn to colleagues for novel ideas or confirmation of one’s thought processes or plans. Engaging in collaboration is inherent in special education teachers’ ability to follow legal mandates that require access to the general education curriculum in the least restrictive environment (Cook & Friend, 2010; Every Student Succeeds Act, 2015; Individuals with Disabilities Education Act, 2004). Collaboration is also a daily function of successful special education teachers (Quinn & Machalicek, 2025) and is a high-leverage practice (McLesky et al., 2017). Similarly, teachers can use AI as another collaborative resource to seek input on a situation. For special education teachers, AI can act as an electronic collaborator to generate tailored content for more personalized learning for their students with disabilities. AI can reduce time and “increase, effect, automatic processes” (Marino et al., 2023, p. 404) by developing and delivering efficient instruction and other supports for their students with disabilities.

AI: Responsive or Generative?

The type of AI teachers may be most familiar with is responsive and not generative. Responsive AI has been used in several different types of software, but it is typically limited to functioning within a discrete system (i.e., software or other function with a primary purpose; Ertel, 2017), for example, using Grammarly as an extension on an internet browser. Grammarly takes the input of typed text and compares it to its own knowledge base to provide feedback. Another example is iXL, which analyzes students’ errors against its knowledge base of errors and provides corrective feedback. In this curriculum software, a student’s assessments results create a learning path based on content already in the software—again, the AI uses the input and compares it to its knowledge base. Although our computer science colleagues use the term “discriminative,” we use the more general term “responsive” for ease of understanding. Responsive AI has been shown to be effective for students with disabilities (Hopcan et al., 2022) and is mostly well received by the general public because it is safely viewed as a user’s partner as opposed to a competitor or rival.

Comparatively, generative AI is not limited to a discrete knowledge base (see Table 1 ). It can learn from multiple different data sources beyond the scope of a single corpus of information, such as using the entirety of published research and gray literature along with other resources (e.g., historical documents, commercial products), and is not bound to text. Generative AI learns to create something entirely new; for example, a generative AI image generator continually creates new images based on entered prompts. Image creation was an original intended use of generative AI during its inception (Ertel, 2017). Now, it has expanded to other formats, such as audio, video, and text, for instance, prompting a text-based generative AI (e.g., Google Gemini, ChatGPT) to summarize information from multiple sources based on a user’s prompt (e.g., “summarize the research from the last 10 years on reading”) or generating unique texts (e.g., “write me a story about two friends”). Recent advances in cloud-based computing have helped generative AI to be readily available (Narayanan et al., 2021) to the point that people may fear they no longer have agency over the technology. Table 1 contains additional ideas of generative AI for the classroom and links to sample AI tools.

Table 1

Sample Generative Artificial Intelligence (AI) Tools and Purposes

Example AI purposes	Descriptions and sample AI tools
Adapting reading passages to a lower reading level, simplified text, or summarized (e.g., copy and pasting a passage from an eighth-grade science passage into an AI tool to reword the passage to a fourth-grade reading level) Developing lesson plans and generating assessments or other instructional resources; for example, using AI to generate a word list of 10 CVC words with /a/ as the vowel, a social story on walking down the hall, or 10 story problems using addition without carrying	AI tools include: • ChatGPT (https://chatgpt.com/; free, paid features for the Plus version) • Google Gemini (https://gemini.google.com/; free, paid additional functions) • Perplexity (https://www.perplexity.ai/; free, paid version for additional responses and files); includes external links for more information based on the information generated Instructional planning-specific: • Curiopod (https://curipod.com/); limited free lessons, subscription-based. Allows users to generate a variety of documents, including lessons, slides, and activities. • MagicSchool AI (https://www.magicschool.ai/); limited free functions, subscription-based. MagicSchool AI has many AI options, including those that are generative, such as their chatbot to ask specific questions. AI options include creating presentations, generating lesson plans, and rewriting text to adapt it, for example, by length or reading level.
Creating or editing images to use with readings, social stories, and vocabulary; illustrate a concept; or other instructional needs	For all AI image options, it is important to be as detailed as possible to generate an image; AI image tools use text-based prompts. Some also include options to upload or edit images or allow a user to include “negative” prompts for aspects not to be included in an image. AI tools include: • Dall-E (https://openai.com/dall-e-2; cost per image) • DreamStudio (https://beta.dreamstudio.ai/; limited free images, cost per image) • FireFly from Adobe (https://firefly.adobe.com/; limited free images, yearly cost)
Data collection and progress monitoring	ChartGPT (https://chartgpt.io/talk/) and GraphMaker (https://www.graphmaker.ai/); limited free charts or requests, subscription-based, allows you to enter in data, such as scores on a science test, to graph it

Note. For more information on AI in education, see the AI resources from the International Society for Technology in Education at https://iste.org/ai and the Center for Innovation, Design, and Digital Learning at https://ciddl.org/ciddl-artificial-intelligence/. AI = artificial intelligence; CVC = consonant, vowel, consonant.

Combining powerful computational capacity, natural language processing, and machine learning has brought generative AI to the forefront of technology usage as a disruption to the status quo. As with any other major disruptive change, many questions arise. For generative AI in a special education context, we examine how teachers can use AI as a collaboration resource and how they can instruct their students to be good consumers of AI.

AI Instruction and Instruction With AI

AI is not a new phenomenon (e.g., predictive words while typing or highlighted grammatical errors in a word processor; see Chen et al., 2022; Hopcan et al., 2022). However, the latest iteration is generative AI, made popular by cloud-based programs such as ChatGPT, Google Gemini, and Microsoft Copilot. Generative AI is remarkably different than previous technology-based tools in its ability to produce something new with each use.

Generative AI combines multiple facets of computer science, such as natural language processing, large language modeling, and neural networks/machine learning. These technical terms are often used in conjunction with AI discussions. Combined, these fundamentally create the output of AI that is so different from previous technology. Understanding the underpinnings of AI and how to generate useful content will help special education teachers to appreciate generative AI as an additional collaborative and efficient resource. These underpinnings are rooted in the use of language.

Natural language processing allows computers to understand human language as it is typically spoken. People can input requests as they would typically inquire if they were in conversation with another person (e.g., “What are good strategies for reading comprehension?”). It has iteratively progressed from the early 1940s through today; as computer hardware and engineering have advanced, so has the ability to understand and improve computers’ natural language processing capabilities (Jones, 1994). Large language models have also become possible with advances in computing technology.

“With AI, it all starts with the prompt.

In large language models, hundreds of statements are used to train an enormous neural network to infer the likelihood of a particular word in an already existing word sequence. For example, if the phrase “special education” is stated, what is the likelihood that the next word would be “teacher,” “classroom,” “context,” or one of countless other possibilities? With a neural network, the AI is trained (i.e., learns) using an observed data set to estimate the future likelihood of a particular word occurring as the very next word in the sequence. The learning requires large data sets, which have only recently become computationally available with the advances in software, hardware, and cloud computing (Xu et al., 2020). The crux of applied generative AI is understanding where and how to begin. Teachers need instruction to develop their own understanding and application of effective AI interactions. With AI, it all starts with the prompt. Computer scientists call it “prompt engineering,” and this term is applicable to teachers as well (White et al., 2023).

Prompt Engineering: Tips for Teachers

Effective use of the human-facing component of generative AI requires understanding the iterative, adaptive, and interactive nature of exchanging ideas through prompt engineering. Much like a conversation with a human evolves based on communication components, AI establishes and retains context and coherency though multiple rounds of interaction. This communication requires human input and subsequent reactions to the AI’s response. Each time the user prompts the AI, the output is tailored and is built on from previous interactions, creating a kind of ongoing relationship between the user and the AI, similar in fashion to a discussion one would have with a human collaborator.

Prompt engineering can be compared to performing a Google search with terms: The first iteration contains keywords generated based on the user’s initial query. Then, depending on the results yielded, the user engages in an iterative process of refining search terms until obtaining the answers to their initial query. For example, a teacher might search “inclusion” on their initial query; due to the large quantity and contextual variety of results yielded, the teacher might search “inclusion AND learning disabilities” or other information needed to narrow down their desired results and rerun their search. However, generative AI is much more sophisticated in its interaction. Unlike a Google search, generative AI retains the context of previous inputs and is nuanced to understand more conversational language. A user’s prompting builds on previous prompts and becomes more of a connected chain that refines and shapes the AI output, much like human responses provide feedback during a dialogue.

Ms. Martinez demonstrates to Ms. Smith how they used prompt engineering in ChatGPT to create addition story problems to meet their instructional needs. Ms. Martinez described using prompt engineering much like when teachers are helping a student with a new task by providing specific directions and then adapting their response and follow-up directions based on how the student responds. Ms. Martinez also explains that through doing this, Ms. Smith can reflect on the generated output, keep it, or make changes aligned to her goals and students’ needs.

PROMPTing AI

Teachers will have a general idea of what they would like to create (e.g., addition story problems, rubric for a paragraph writing task) when initially prompting the AI. For efficient and effective prompt engineering, teachers can use the PROMPT acronym (see Figure 1 ) when electronically collaborating with generative AI:

Pick your terms.

Run it.

Optimize/Once again.

Make notes.

Pause and reflect.

Take, Tailor, or Try again.

Figure 1

PROMPT infographic reference

Specifically, PROMPT reminds users of the steps needed to prompt generative AI (see Table 2 ; for a more complex example, see Figure 2 ). First, in pick your terms, the teacher starts with relative terms based on their inquiry and uses the appropriate button or click to initiate the AI’s first yielded response. When picking terms, it is important to be specific. For example, instead of a broad prompt such as “what are math strategies,” a better input might be, “what are math strategies for teaching two-digit addition without carrying” (for additional tips, see Figure 3 ).

Table 2

PROMPT Detailed Example

Steps	Description	Explanation	User input	Chat GPT output
Pick your terms	Pick initial terms/vocabulary designed to yield a desired response to this inquiry.	Teacher desires a list of “bl” words that can be paired with pictures and start with those letters.	Tell me words that start with bl.
Run it	Initiate the query.	Starts the interaction/conversation	Press enter (or click on appropriate button to initiate).	Certainly! Here are more words that start with “bl”:Words starting with “bl”:BlastBlinkBlitheBlotBlunderBlurBlazingBlurtBlamelessBlightThese additional words should expand your vocabulary starting with bl.
Optimize/once again	Based on what the AI output does or does not produce, the teacher potentially needs to make adjustments to their terms.	In this case, the teacher’s students can consistently read words with short vowel patterns, such as a word ending in “-ink” or “-ot,” and short words with a long vowel that end in “-e” (e.g., “slide”). But the teacher reflects that the students would not be able to read many of the words generated due to their length, difficulty, or mismatch with student’s background experiences, such as “blight” and “blameless.”The teacher optimizes the words produced with two to three additional letters after /bl/ to capture word endings and use of a long vowel and adjusts the next requests for words that are four and five letters long.	Tell me 4 and 5 letter words	Certainly! Here are some 4 and 5-letter words that start with bl:4-letter words:BlabBlipBlobBlueBlurBlotBlowBled5-letter words:BlackBladeBlazeBlendBlinkBlushBlockBlondBloodBloomThese are examples of 4 and 5-letter words that begin with “bl.”
Make notes	Note the useful portions of the AI’s response.	In this case, some of these words are useful to the teacher for creating the word list, but some are not due to the students’ individual needs, including cultural relevancy, vocabulary, current levels of achievement, and IEP goals. Some of these words also will not be easy to clearly depict in a picture.	None
Pause and reflect	Examine what has been done up to this point and how to proceed with next steps.	In this case, the teacher reflects on their notes and the generated word lists. They recognize that AI has provided useful lists of words but that they need to tailor the words lists to their students’ needs.	None	Remains the same
Take, Tailor, or Try again	Based on what the AI has generated thus far, users decide to end the interaction or make adjustments.	This ends the session or leads the user to subsequent iterations using the P, R, O, M, and/or P steps again.In this case, from the word lists generated, the teacher tailors it to her students’ reading needs and the activity. Because the activity includes matching a picture to the printed word, the teacher selects the following 10 words:BlobBlueBlurBlowBlockBlinkBlackBloomBlastBlondThe teacher didn’t include words unfamiliar to the student, too difficult to read, and/or too difficult to clearly illustrate, such as “blab” or ‘blip.”Out of curiosity, the teacher repeats this process again with other blends.	None	Remains the same

Note. In this scenario, ChatGPT (OpenAI, 2024) is used as an educational collaborator by a special education teacher seeking assistance with consonant blends for a multiweek decoding unit. The teacher recently introduced students to the /bl/ consonant blend at the beginning of a word and realized some of her students need additional opportunities to practice. She decides to try out ChatGPT with the /bl/ blend to see what the results yield. Her goal is to create a word list that can be paired with pictures of each word for an upcoming practice activity where the students will match the words to a picture. AI = artificial intelligence; IEP = individualized education program.

Figure 2

Additional PROMPT example: A middle school teacher uses ChatGPT (OpenAI, 2024) to generate ideas on teaching science, specifically, the water cycle, to students with reading and math instructional needs

Figure 3

Example prompting statements

Next, PROMPT reminds teachers to run it, or run their query in the AI. Based on what the AI output does or does not produce, the user needs to make corrections to optimize the vocabulary used as the next input to the AI and start the next iteration.

From this iteration of output, the user can make notes about useful parts of what the AI has produced, for instance, identifying the generated aspects that match the users’ needs and/or what might need to be changed. Next, the user pauses and reflects to decide to either take what the AI has created or continue. Last, with take, tailor, or try again, the user can opt to take and use the output if it meets their needs (i.e., take), ending the session. If not, they can initiate another iteration by tailoring additional input terms (i.e., tailor) and initiating another iteration. Users can do this as many times as needed to get an acceptable output or start over (i.e., try again).

Using PROMPT, Ms. Smith is eager to develop reading materials for her students for an upcoming lesson on reading words with the /bl/ consonant blend. She reviews Figures 1 and 2 to familiarize herself with possible interactions with AI. She prints Figure 1, places it by her laptop, and refers to it often as she tries out PROMPT when using ChatGPT for the first time. She wants to create word lists for a reading game to provide students with additional practice (for her prompts and output, see Table 2 ). Ms. Smith starts with a broad request, “tell me words that start with bl,” and realizes that it is important to be specific on the word types and length. As she makes modifications, she reflects on her students’ specific needs and the objective: The words should be decodable, should be in the students’ vocabulary, should not have more than one meaning, and should be able to be paired with a picture for the game. Using this information and the generated word list, she picks 10 words for the game. She opts to not use words such as “blip.” Although the students might be able to decode it, that may not know what the word “blip” means or may not have heard it, and it would be difficult to pair with a picture. Instead, she uses words such as “block” and “blue,” which can be illustrated and match her students’ reading abilities.

Teachers can use AI to generate information to help them be more efficient, such as identifying evidence-based practices, developing lesson plans, modifying content, and finding other resources (see Table 1 ). AI has the ability to generate any number of time-saving resources that are much more sophisticated than what can be typically garnered from an internet search, including those that teachers may not have access to without the use of AI (e.g., quickly create AI-generated images for a social story unique to a student and the topic). Although an internet search can generate things such as lesson plans and lists of evidence-based practices, AI allows the teacher to continually modify the generated output by prompting AI to provide more or different information that is increasingly more specific. For example, a teacher could search “strategies to teach science.” This would bring up many results from a search engine and typically require the teacher to make multiple clicks into different pages on an assortment of websites. But through AI, a teacher can input follow-up prompts specific to their students, their instructional practices, and in response to what the AI generates (e.g., asking for more information on a science strategy generated and then how it applies to students’ specific needs; for an example, see Figure 2 ).

The usefulness of AI-generated output is that it is more closely aligned to collaborating with another colleague who has a better understanding of students’ needs and the learning environment. Using PROMPT, Ms. Smith engages generative AI as another collaborative source for her reading game, much like she would from a colleague, coach, or mentor.

Ms. Smith’s reading game with the AI-generated wordlist went off without a hitch! The students were engaged and consistently read the words with the consonant blend. She used the steps in PROMPT to get to the needed word lists to deliver instruction and create activities to meet her students’ needs quickly. Since then, she has considered other, more complex ways to use AI for her instruction. Each time, she uses PROMPT as a guide during her interactions with AI. She shared her success with other colleagues who are also learning to use AI as an effective collaborator.

AI Is Not the Teacher

Despite the potential value of AI in special education, using AI will not replace instruction, pedagogy, or teachers’ knowledge of special education processes and regulations. For those apprehensive about AI, consider that AI is another tool for the instructional toolbox to augment and support, not replace, instruction or the teacher. Much like how other types of special education technology did not replace instruction or the teacher, AI acts in collaboration with teachers and their instruction (OET, 2023b). For example, using text-to-speech assistive technology has not replaced reading instruction or actual reading, but it has reduced barriers for students with disabilities accessing text.

Additionally, the OET (2023a) specifically stated that “teachers and other people must be ‘in the loop’ whenever AI is applied to notice patterns and automate educational processes.” Although AI might generate a word list, lesson plan, reading passage, or other instructional materials, teachers must evaluate the content generated, including the accuracy and the applicability to their students’ and their instructional objectives. Teachers must apply their pedagogical and content-area knowledge to the content generated to create the best possible instruction for their students. These steps are represented in the pause and reflect and take, tailor, or try again components of PROMPT (see Figure 1).

AI acts as a technology-based collaborator; a teacher can use AI in instruction for delivery, monitoring progress, and other tasks, but AI cannot do those without the teacher designing and overseeing its function. Just as a teacher might consider feedback from other colleagues or stakeholders and then make decisions based on their own experiences, knowledge, and students, they must do the same when using generative AI. Collaboration, even technology-generated, benefits their students by providing ideas and instruction that originate from multiple perspectives. This added perspective just so happens to be through AI. Therefore, it is important that teachers are critical consumers of AI and the information that implementing PROMPT generates. Just as teachers would consider any other collaborators’ suggestions and keep what they understand to be in the best interest of their students, teachers can critically consume information created with AI.

“Prompt engineering is fundamental to AI collaboration and obtaining useful responses for developing instructional materials and other resources.

Conclusion

The recent proliferation of AI that can create something completely new is causing a swift and dramatic paradigm shift in how we approach many aspects of our lives, including how we learn and teach. Much in the same way teachers collaborate with one another, AI can be considered an additional collaborative resource. Prompt engineering is fundamental to AI collaboration and obtaining useful responses for developing instructional materials and other resources. For special education teachers, learning to design quality and efficient prompts can be guided by the steps in the PROMPT acronym. These steps can be implemented in many different types of requests every time a teacher seeks to collaborate with AI to support instruction. Successful collaboration benefits special education teachers and the students they serve.