logo image

TD Magazine Article

Brainstorming With AI

With the proper prompts, integrate AI dialogues into your content development toolkit.


Fri Jun 30 2023

Brainstorming With AI

Artificial intelligence has become a ubiquitous technology in daily life. From chatbots to speech recognition to content generation, AI is everywhere. One area where it is poised to make a significant impact in L&D is in content design and development.

AI-powered programs with sophisticated algorithms, such as ChatGPT, are on the verge of revolutionizing the way we create and deliver learning content. However, it's important to understand that AI is not a panacea. Effective integration of generative AI into the development process requires careful consideration and planning.


How generative AI works

To fully leverage the potential of AI-powered programs, you must have a foundational knowledge of how they work. At a high level, ChatGPT is a chatbot interface paired with complex AI algorithms referred to as a large language model; LLMs are a type of AI designed to understand and generate human language.

The term large refers to the size of the model's neural network, which is the computational system that processes information and makes predictions using a technique called autoregression. Basically, the model predicts the most likely next word in a sequence based on the previous words, which start with a user's prompt. The larger the neural network, the more data the model can process and the more accurate its responses.

LLMs are trained on vast amounts of text data from sources such as books, articles, and websites, and they can generate human-like responses to a wide range of language-related tasks. When you hear people referring to GPT 3.5 or GPT 4, they are referring to ChatGPT's current LLM version. With each new version of the LLM, the AI tool becomes more powerful.

Working with AI as a brainstorming activity

To obtain the most accurate response from ChatGPT, it is essential to comprehend its generation process and tailor the question accordingly in a process often referred to as prompt engineering. A good way to begin is to treat the AI tool as a brainstorming partner because its foundation is in language, so it is great at holding conversations.

Traditional brainstorming is often described as a funnel-shaped process, starting with broad, open dialogue and transitioning to more and more focused concepts. When working with AI, the same brainstorming dialogue principle applies. The process begins with a well-crafted prompt from a human. That prompt elicits a response from the AI tool, which the human can then modify or redirect to refine the output. Such iterative dialogue between a human and AI can result in improved productivity and output quality.


To optimize the effectiveness of the approach, start with a prompt that allows for a broad range of potential responses. As the brainstorming process continues, gradually refine the output. View the AI as a partner rather than as a replacement for human creativity and intuition. The key is to use the prompts to encourage the AI to generate responses aligned with your learning project goals.

In the context of content creation, initially focus the AI on generating storyboards or training outlines for a specific topic. Analyze the resulting outputs for overlapping topics and accuracy, filtering through the suggestions to determine what will resonate with your target audience. Such an iterative process often involves multiple rounds of refinement and evaluation, with each round building on the previous one as you gradually narrow the focus toward the most promising ideas or topics.

By using the AI output as a starting point, you can produce increasingly useful and relevant content. Once the outline or storyboard is complete, you can repeat the process for each specific objective in a course.

It's all about the prompts

Because so much of working successfully with generative AI is dependent on the prompts, focus time and effort on getting them right. I remember taking a course that introduced me to the five Cs of effective written communication; I wondered if there were similar best practices for writing prompts, so I asked ChatGPT.

First, it responded that prompts should be clear, concise, relevant, specific, and creative. Upon asking a second time, it said that prompts only need to be clear, concise, and focused on the desired outcome. Last, it presented recommendations that included providing context, using an iterative process, and specifying the output format.


That is a great example of why working with AI to create content should be an iterative process—one that requires human input and validation. Each iteration can provide new ideas as well as repeat ideas or themes from prior iterations. Using your knowledge and expertise, pick and choose from all the AI responses to select the best suggestions and filter out anything that is incorrect.

A brainstorming process that works for me begins by giving ChatGPT a role. You can start the prompt with "Role:"—for example, "Role: You are an instructional designer creating a course on individual development planning (or IDP)."

Next, give the AI any background information that it may need, including information about the learners: "You work in the tech industry for a startup social media company experiencing explosive growth. Your learners are all new managers with little or no knowledge about the IDP process."

Last, provide the AI with a clearly defined task for it to perform, including rough suggestions for direction if appropriate: "Draft a storyboard for an IDP course."

As you draft your prompts, strike a balance between providing specific guidance and allowing the AI to generate creative and relevant responses. While keywords can be useful for indicating certain aspects or constraints of a task, overreliance on them can limit the AI's ability to produce diverse and meaningful outputs.

For example, the prompt "Design a comprehensive online course that empowers leaders to navigate complex challenges and to inspire high-performance teams" has emphasis on the course's broader context and purpose. In contrast, "Create an e-learning course on time management skills for middle managers" is a prompt whose focus is primarily on the specific keywords "time management skills" and the target audience of "middle managers," which restrict the AI's creativity and limit the scope of content generation, ultimately leading to a course that is too narrow in focus.

Another prompt-writing guideline is to use inclusive, bias-free language. AI systems can be biased based on the data they have consumed, which means prompts that reinforce or perpetuate biases can lead to problematic content generation (see sidebar).

AI in action

When brainstorming, start broadly and then refine by chaining your prompts together. The initial output may not be perfect, but by using it as a starting point and modifying the prompts based on the output, you can steer the AI toward producing increasingly useful and relevant content.

For example, I gave ChatGPT this prompt:

Role: You are developing a course on individual development plans (IDPs) to be delivered to managers who are unfamiliar with the process.

Provide me with potential course objectives for this course.

ChatGPT returned 10 objectives, including:

Understand the purpose and benefits of individual development plans (IDPs) and how they can support employee growth and development.

Do you notice anything wrong with that objective? I followed up the results by writing:

Please add "so that you can" statements to the end of each of my objectives.

ChatGPT recognized that as a chained prompt and applied the new request to its previous results. Here is the rewritten objective:

Understand the purpose and benefits of individual development plans (IDPs) and how they can support employee growth and development so that you can effectively implement and communicate the value of IDPs to your employees.

ChatGPT similarly strengthened all 10 objectives as a result of my addition. Next, I chained another modifier onto the request by asking ChatGPT to:

Make these objectives more measurable.

It returned 10 measurable objectives such as:

Define the purpose and benefits of individual development plans (IDPs) and how they can support employee growth and development, and develop a communication plan to effectively communicate the value of IDPs to employees.

With that result, I both gained something (measurement) and lost something (the learner's activity), so I next asked ChatGPT to:

Turn these objectives into learner-facing "you will" statements.

ChatGPT reworded the objective:

You will be able to define the purpose and benefits of individual development plans (IDPs) and develop a communication plan to effectively communicate the value of IDPs to employees.

Finally, I wondered whether ChatGPT could apply verbs from Bloom's taxonomy to the results, so I tested it with this prompt:

Rewrite the learning objectives using Bloom's taxonomy verbs.

ChatGPT changed all my objectives, often splitting each into multiple new objectives. My first objective, for example, suddenly morphed into three new ones.

By the end of this course, the learner will be able to:

  • Explain the purpose and benefits of individual development plans (IDPs) using appropriate examples.

  • Evaluate the effectiveness of different communication methods for conveying the value of IDPs to employees.

  • Develop a comprehensive communication plan for promoting the benefits of IDPs to employees, using appropriate channels and language.

Notice the evolution of the results as I chained more modifiers onto the original request. While I chose to add the modifiers on a step-by-step basis, you can chain requests in a single prompt. The value in breaking it down into multiple prompts is that it enables me to check results along the way to ensure that the AI is returning accurate results.

At this point, you may be asking yourself whether it is worth it to incorporate AI into your content development process. With so much time spent on crafting prompts and validating the output, is AI really a time saver? The answer is that it can be. AI can generate large amounts of content quickly, and as you improve your prompt-writing skills, the output likewise will improve.

The time savings will also depend on the complexity of the course structure, topic, or format. I recently created a microlearning course for which AI generated roughly 60 percent of the material. The development time was about 50 percent of what it would have taken me previously for similar courses.


While AI technologies have made significant progress in recent years, remember that they still have their limitations. One of the most significant ones is AI's reliance on incomplete data. As previously mentioned, ChatGPT and other LLM programs rely on data sourced from the internet, which is not always reliable.

To see how unreliable ChatGPT can be, ask it to provide reference sources. If it can't provide them, it will make them up. You read that right—ChatGPT will prioritize filling your prompted request over the truth.

Further, the significant amount of misinformation and bias on the internet can lead to AI models perpetuating the inaccuracies. Another limitation is their lack of common sense and context. Although AI models can generate text and language that is coherent (and even sophisticated), they cannot understand the nuances of human communication. For example, AI technologies can't comprehend sarcasm, irony, or metaphor, which are all essential components of human language. That can lead to misinterpretations and inaccuracies in the output.

AI technologies also struggle with understanding ambiguity and uncertainty. When developing learning content, that can be particularly problematic because the subject matter is often complex and multifaceted. AI models may have trouble comprehending the context of a specific topic and thus provide inaccurate or incomplete information. Another learning-specific limitation is AI's difficulty in recognizing the emotional and psychological aspects of learning.

Ethical and moral considerations are yet more matters AI models are unable to understand. Humans are responsible for making ethical and moral judgments about the content they create and distribute. AI technologies, on the other hand, lack the ability to comprehend those considerations and may generate content that is harmful or inappropriate.

See AI's potential

Despite the limitations, there is significant potential for AI to be a valuable partner in the content creation process. With its ability to generate content quickly and efficiently, ChatGPT can create large amounts of material in a short amount of time. Additionally, it can help with identifying knowledge gaps and can provide personalized learning experiences for individuals.

As AI models become more sophisticated, they will be able to generate even more complex and nuanced content, making it increasingly difficult to distinguish between human- and machine-generated content. Although AI can generate informative and accurate content, it is best to use AI in conjunction with human instructional designers rather than as a replacement for them. AI has the potential to revolutionize the way we approach L&D, but we must use it in a way that complements human expertise rather than replacing it entirely.

Ensure AI Prompts Are Inclusive and Free of Implicit Biases

When working with artificial intelligence systems like ChatGPT, it is crucial to recognize that they can be biased based on the data they have consumed. As a result, prompts that reinforce or perpetuate biases can lead to problematic content generation. Reduce the likelihood of problems with these guidelines.

Start with inclusive language. Use language that avoids assumptions about gender, race, or other characteristics. For example, instead of "he" or "she," use gender-neutral pronouns such as "they" or "them." That helps to ensure that your prompts are welcoming to a diverse range of learners. You can even tell ChatGPT to return results using gender-neutral language.

Focus on the learning objectives. When crafting prompts, home in on the learning objectives and the task's broader context. Avoid prompts that are overly specific or based on stereotypes or assumptions about certain groups of people. For instance, instead of "Write an essay on the challenges working mothers face by balancing their careers and family responsibilities," a better prompt would be: "Explore the evolving dynamics of work-life balance and the factors that influence individuals' ability to integrate their personal and professional lives."

Use diverse examples. Incorporate diverse examples and perspectives into prompts to keep the generated content inclusive and relevant to a wide range of learners. Rather than "Provide information about the history of music," be more specific with: "Provide information on the diverse origins and influences of music, including contributions from various cultures, regions, and historically marginalized communities. Highlight lesser-known genres, musicians, and musical traditions that are often underrepresented in mainstream narratives."

Check for bias in the results. After generating content using the AI, carefully review the results for any bias, stereotypes, or incorrect information. Make any necessary modifications or adjustments.

You've Reached ATD Member-only Content

Become an ATD member to continue

Already a member?Sign In



July 2023 - TD Magazine

View Articles

Copyright © 2024 ATD

ASTD changed its name to ATD to meet the growing needs of a dynamic, global profession.

Terms of UsePrivacy NoticeCookie Policy