I gave a presentation at the 38th national USDLA National Conference on Tuesday June 24th. On this page you will find additional resources to support that presentation.

You can download a PDF of the presentation handout here.

Just as a sommelier understands the nuances of wine, I encourage everyone to appreciate the provenance, pairing, and “price” of AI. It’s about selecting the right tool for the right task while being mindful of its capabilities and risks.

When most people hear "artificial intelligence," they immediately think of ChatGPT or similar chatbots. But AI is actually an umbrella term covering several distinct technologies, each with unique applications in educational settings, for example.

Generative AI creates new content (text, images, music, and video) that resembles its training data. This is the technology behind tools like ChatGPT, Claude, and Gemini that can write essays, create lesson plans, or generate quiz questions.

Predictive AI analyzes existing patterns to forecast future outcomes. In education, this might help predict which students are at risk of dropping out or which teaching methods will be most effective for different learning styles.

Machine Learning encompasses the broader field of systems that can analyze data, identify patterns, and make decisions. This includes supervised learning (where the system learns from labeled examples), unsupervised learning (finding hidden patterns), and reinforcement learning (learning through trial and error).

At the cutting edge are Foundation Models (also called "Base Models) like GPT, Claude, and LLaMA. These are trained on vast amounts of data and can be fine-tuned for specific educational tasks.

The Hallucination Problem: When AI Gets Creative with Facts

One of the biggest challenges with Generative AI that educators encounter is the phenomenon of "hallucinations," where AI generates information that sounds plausible but is completely fabricated. It might hallucinate information that was not in the source material (closed domain), or inject completely false information that sounds authoritative (open domain).

This is not a fatal flaw. However, It is a characteristic that educators need to understand and work with. There are several strategies for reducing hallucinations, such as:

• Retrieval-Augmented Generation (RAG): Providing the AI with specific, relevant documents to reference
• Self-consistency: Generating multiple responses and choosing the most consistent one
• Chain of Thought prompting: Asking the AI to think step-by-step through problems

NotebookLM is one tool that can be helpful in reducing hallucinations, via source grounding, where the LLM provides answers based on your content only. NotebookLM has multi-document synthesis, where it draws data from multiple files to compose an answer. Additionally, it provides inline citations.

Currently, I categorize AI models into three distinct levels, each requiring different interaction strategies:

1. General Purpose Models: Older, simpler models that need explicit instructions
2. Reasoning Models: Advanced systems that think before responding, using longer internal chains of thought
3. Agentic Models: Independent systems that can work autonomously for extended periods, conducting research and completing complex tasks

Understanding these distinctions helps educators choose the right tool for each educational task.

The quality of AI output depends heavily on how you communicate with it. Here are two primary frameworks for effective prompting:

The PREPARE Method (by Dan Fitzpatrick)

• Prompt: Start with a clear question that sets the stage
• Role: Give the AI a specific role and context
• Explicit: Be specific to avoid misunderstandings
• Parameters: Set frameworks for tone and format
• Ask: Request clarification questions before proceeding
• Rate: Ask the AI to evaluate its own output
• Emotion: Add emotional stimulus to increase quality

The CREATE Method (by Dave Birss)

• Character: Describe the role the AI should assume
• Request: Clearly define what you want
• Examples: Provide examples when possible
• Additions: Refine the task with additional considerations
• Type of Output: Specify the format and length
• Extras: Include any reference materials

You can use these frameworks to create a syllabus statement for AI usage in courses.

Using the PREPARE method, you might prompt:

“Write a syllabus statement for AI usage in a DePaul course. You’re an education expert and skilled teacher. In the syllabus statement, clearly articulate how or how not Generative AI should be used in this course. Use an informative tone and keep the syllabus statement under 300 words. Ask me some clarification questions first and then answer. Give the summary a rating based on 0-10 points and indicate what could be improved. Breathe in and breathe out. Try to really do your best. It’s important to me.”

Using the CREATE method, you might prompt:

“You are an expert educator with 20 years’ experience and numerous teaching awards. You can create and teach amazing classes that incentivize students to think critically. I want you to generate a simple, straightforward syllabus statement that sets parameters for if/when generative AI usage is permitted in my class. Start by asking me what type of assignments I will use in class and my expectations of student AI usage. Draw your language from examples like the AI Assessment Scale, DePaul University’s Academic Integrity Policy, and DePaul University’s AI Teaching Recommendations. Use an informative tone and keep the syllabus statement under 300 words. Do not suggest a syllabus statement until I give you my expectations of student AI usage and information on the type of assignments I will use. Ask for my expectations and types of assignments then wait for my response before answering. Write it in plain English without jargon.Explain your thinking.”


Automated Quiz Generation:
I routinely convert my handouts and resources into D2L Brightspace quizzes. Here is one of the prompts that I use:

Based on the PDF, generate 20 multiple-choice questions formatted specifically for use in D2L
Brightspace. Please export as a Word document. Follow these guidelines to ensure clarity and compliance
with formatting rules:

1. Question Format: Each question must be a single line and start a new paragraph.
Example: Which of the following is a planet?

2. Answer Options: Provide four answers, each on a separate line.
Randomize the placement of the correct answer among the four options.
Mark the correct answer with an asterisk (*) placed at the beginning.
Example:
Sun.
*Earth.
Proxima Centauri.
Mars.

3. Feedback Text: Include a feedback line immediately after the last answer option.
Begin the feedback line with an at sign (@) and provide an explanation quoting the relevant part of the
reading. Insert a paragraph break after the feedback.
Example:
@Earth is the only planet in this list.

4. Avoid Clues from Order: Ensure the placement of the correct answer is randomized in each question to
prevent patterns or clues from the order.

5. Reworded Variants: It’s acceptable to ask multiple questions about the same content, but ensure the
phrasing differs significantly.

6. Special Answer Types: No more than 15% of the questions should include “all of the above” or “none of
the above” as an answer choice. For these cases, do not randomize the answer order.

7. Reference Naming: Always refer to the source document as “this week’s slides” in the feedback or when
quoting content.

8. Correct Format Example: Ensure each question is unnumbered and unbulleted and follows this format:
Correct Format Example:
Which of the following is a planet?
Mars.
Sun.
*Earth.
Proxima Centauri.
@Earth is the only planet in this list.

Which of the following statements is true about the market for foundation models?
It is highly competitive with many small players.
It has low fixed costs.
*It tends towards concentration due to economies of scale.
It is unaffected by data availability.
@The market for foundation models tends towards concentration due to economies of scale.

9. Incorrect Format Example: Ensure each question is not in a numbered or bulleted list. Ensure each
question, answer, and feedback is on a new line:

Incorrect Format Example:
What is the estimated percentage of the global GDP that foundation models may underpin within a
decade? 1-2%. 3-5%. *7-10%. 15-20%. @Foundation models may underpin 7-10% of the global GDP
within a decade.


You can use AI to convert Zoom course transcripts into engaging Kahoot quizzes, complete with randomized answer placement and proper formatting for easy import.

# Kahoot Quiz
## Instructions
Create 20 Kahoot quiz questions from the transcript of the AI in Marketing course. Format the quiz following the requirements posted here: https://kahoot.com/library/quiz-spreadsheet-template/
Export the quiz as an Excel file.
## Additional Requirements
* Ignore questions that were previously asked in the Kahoot quiz at the start of class.
* Provide four answers for each question.
* Randomize the placement of the correct answer among the four options.
* Avoid Clues from Order: Ensure the placement of the correct answer is randomized in each question to prevent patterns or clues from the order.
* In Excel, the order of the columns is as follows:
* B: Question - max 120 characters.
* C: Answer 1 - max 75 characters.
* D: Answer 2 - max 75 characters.
* E: Answer 3 - max 75 characters.
* F: Answer 4 - max 75 characters.
* G: Time limit (sec) – Time limit for each question is 20 seconds.
* H: Correct answer - This should be a number from 1 to 4.
<>
<>

Deep Research

Advanced AI research tools from ChatGPT, Gemini, and Perplexity can conduct comprehensive investigations, generating detailed reports with proper citations and structured analysis.

You can use the template below to create your deep research prompt:

1. Copy and paste the template into a Word document or equivalent.
2. Edit the italicized Instruction, using the Example as a guide.
3. Delete the Example content and the Instruction heading
4. Save your prompt.
5. Use your prompt with one of the deep research services.

Title:
• Instruction:
• Insert a clear, concise title that reflects the focus of your study.
• Example:
• Exploring Emerging Trends in [Your Topic]

Background & Context:
• Instruction:
• Summarize the historical background and current trends related to your topic. Discuss prior research or initiatives that have led to the present inquiry.
• Example:
• In recent years, [Your Topic] has evolved significantly due to advancements in technology and changes in societal needs. Early research in the field laid the groundwork for understanding these dynamics, while recent studies have provided fresh insights into emerging patterns and practices.

Brief Description:
• Instruction:
• Provide a summary that explains the purpose of your research, the core ideas you will examine, and the approaches or perspectives under consideration.
• Example:
• This project investigates the key factors driving [Your Topic] by examining current trends, challenges, and opportunities. The research will compare different methodologies and assess their impact on the field.

Purpose & Value Proposition:
• Instruction:
• Explain why the research is critical. Describe how the insights will influence decision-making, strategy, or further study.
• Example:
• The analysis aims to uncover critical insights into [Your Topic] that can inform policy development, strategic planning, and future research initiatives. These insights will be valuable for stakeholders seeking to understand the implications of ongoing changes and for guiding effective decision-making.

Scope & Assumptions:
• Instruction:
• Define the boundaries of your research, including the aspects you will focus on, the time frame, and any key assumptions or limitations (e.g., data availability, methodological constraints).
• Example:
• This study focuses on the analysis of [Your Topic] over the past decade and assumes that the available data and literature accurately reflect the current state of affairs. Limitations include potential biases in source materials and the evolving nature of the subject matter.

Recommendations:
• Instruction:
• Provide actionable recommendations based on your analysis. Identify areas for further research or potential strategic initiatives.
• Example: It is recommended that future studies focus on hybrid models that leverage the strengths of both approaches. Additionally, stakeholders should consider investing in research that addresses the identified limitations to maximize overall impact.

Documentation:
• List all research studies, expert reports, and credible sources used in your analysis. Ensure that your citations follow proper formatting guidelines and that all sources are reliable and current.
• Example:
References: - [Author, Year]. Title of the Study/Report. - [Organization, Year]. Title of the Publication.

Clarity:
• Edit your text to ensure clarity and conciseness. Avoid overly complex sentences.

Academic Tone:
• Maintain a formal, scholarly tone. Avoid colloquial language or informal expressions.

Evidence-Based Language:
• Support every claim with credible evidence or citations. Clearly indicate the sources of your information. Focus on providing a concise yet informative background. Include only the most relevant historical details to maintain focus.

Structure:
• Use clear subheadings (e.g., Key Features, Performance, Challenges) to organize your analysis. Ensure logical flow and coherence between sections.

Citation Integrity:
• Double-check all references for accuracy and ensure that every factual statement is properly supported by evidence.

Example Prompt 1

Title: AI and Academic Plagiarism in Online Asynchronous Courses

Background & Context: Some undergraduate and graduate students in online asynchronous courses do not do their assigned reading, or review required materials, but instead ask Generative Artificial Intelligence services (like ChatGPT, Claude, or Gemini) to write their assignments and respond to discussion posts on their behalf. These students do not review or edit what the generative Artificial Intelligence services produce. They are not doing the work they are expected to or engaging in critical thinking.

Brief Description: Investigate the key factors driving students to cheat with AI in online asynchronous courses. Identify proven strategies that faculty can employ to detect AI plagiarism, mitigate AI plagiarism, and to motivate students into actively engaging with the material and assignments.

Purpose & Value Proposition: The analysis aims to uncover critical insights into why and how students cheat with Generative Artificial Intelligence, that can inform policy development, instructional design, faculty training, and future research initiatives. These insights will be valuable for professors teaching College of Business online asynchronous courses at DePaul University

Scope & Assumptions: This study focuses on the analysis of the past five years and assumes that the available data and literature accurately reflect the current state of affairs. Limitations include potential biases in source materials and the evolving nature of the subject matter.

Recommendations: Provide recommendations that are generalizable across academic disciplines within a university.

Documentation:
• List all research studies, expert reports, and credible sources used in your analysis. Ensure that your citations follow proper formatting guidelines and that all sources are reliable and current.
• Example:
References: - [Author, Year]. Title of the Study/Report. - [Organization, Year]. Title of the Publication.

Clarity:
• Edit your text to ensure clarity and conciseness. Avoid overly complex sentences.

Academic Tone:
• Maintain a formal, scholarly tone. Avoid colloquial language or informal expressions.

Evidence-Based Language:
• Support every claim with credible evidence or citations. Clearly indicate the sources of your information. Focus on providing a concise yet informative background. Include only the most relevant historical details to maintain focus.

Structure:
• Use clear subheadings (e.g., Key Features, Performance, Challenges) to organize your analysis. Ensure logical flow and coherence between sections.

Citation Integrity:
• Double-check all references for accuracy and ensure that every factual statement is properly supported by evidence.

Example Prompt 2

# Best Practices for LLM Optimization (LLMO) of Website Content
## Background & Context In an era where AI-powered assistants (e.g., ChatGPT, Claude, Gemini) increasingly mediate users’ access to information, it is no longer sufficient to optimize web pages solely for search-engine algorithms. LLM Optimization (LLMO) focuses on structuring and enriching your site’s content so that it is accurately parsed, understood, and surfaced by large language models when they generate answers in response to user queries.

## Brief Description Investigate the principal techniques, content patterns, and technical considerations that enable websites to rank well not only in search engines but also in AI-generated outputs. Identify actionable best practices in content architecture, metadata tagging, semantic markup, and conversational framing that drive high-quality, factual, and context-rich AI answers—thereby maximizing visibility and authority in AI-mediated environments.

## Purpose & Value Proposition This analysis will:
1. Illuminate how content must evolve to meet the interpretive needs of generative AI.
2. Provide a framework that digital marketers, SEO specialists, and web developers can apply to future-proof their online presence against the shift toward AI-first information retrieval.
3. Offer measurable criteria by which organizations can audit and improve AI-readiness of their content.

By adopting these LLMO best practices, stakeholders can ensure their expertise is accurately represented in AI assistant responses—boosting trust, engagement, and conversions across AI-driven channels.
## Scope & Assumptions - **Timeframe:** Focused on developments and studies from the past five years (2020–2025).
- **Content Types:** Applies to informational articles, FAQs, product/service pages, and knowledge-base entries.
- **Assumptions:**
- AI models principally rely on publicly accessible text and structured data (e.g., `JSON-LD`, `schema.org`).
- Recommended practices are compatible with standard web-publishing platforms.
- Source materials accurately reflect current LLM behaviors, acknowledging that model architectures and training corpora continually evolve.

## Recommendations 1. **Semantic Structuring**
- Implement clear hierarchies (`H1–H4`) and use descriptive headings.
- Embed machine-readable schema (`FAQPage`, `HowTo`, `Article`) to signal content type.

2. **Concise, Self-Contained Answers**
- Provide standalone, context-rich answer blocks near the top of pages to facilitate snippet extraction.

3. **Conversational Framing**
- Incorporate Q&A pairs and natural-language prompts that mirror how users phrase questions to AI.

4. **Robust Citation & Attribution**
- Include inline citations and source lists to enable AI to verify and reference authoritative materials.

5. **Metadata & Embeddings**
- Supply rich meta descriptions and consider generating and exposing document embeddings for advanced retrieval.

6. **Continuous Monitoring & Feedback Loops**
- Use AI-driven analytics to test which content formats yield the most accurate, highest-ranked AI responses, and iterate accordingly.

## Documentation - List every research paper, industry report, and authoritative blog post reviewed.
- Ensure citations adhere to a consistent style (e.g., APA, IEEE).

**Example References:** - Smith, J., & Lee, R. (2023). *Optimizing for Generative AI: A New SEO Frontier*. Journal of Digital Marketing, 12(1), 45–62.
- OpenAI. (2024). *Best Practices for Prompt Engineering*. OpenAI Technical Report.

## Clarity - Use short paragraphs and bullet lists.
- Avoid jargon; define specialized terms (e.g., “tokenization,” “knowledge graphs”).
- Prioritize plain English and active voice.

## Academic Tone - Maintain formality and objectivity.
- Avoid contractions and colloquial phrases.
- Use third-person and passive constructions judiciously to emphasize evidence over opinion.

## Evidence-Based Language - Preface claims with qualifiers when necessary (e.g., “Recent studies suggest…,” “Empirical results indicate…”).
- Attribute all data points and model behaviors to specific sources.
- Present counterpoints where the literature diverges.

## Structure Organize the investigation under these subheadings for logical flow:
1. Introduction to LLMO
2. Content Architecture & Semantic Markup
3. Conversational Prompt Patterns
4. Citation and Source Signaling
5. Technical Integrations (Embeddings, APIs)
6. Monitoring & Evaluation Methods
7. Case Studies & Comparative Analyses
8. Synthesis & Best Practice Framework

## Citation Integrity - Verify every reference’s publication date, authorship, and URL/DOI.
- Cross-check facts against at least two independent sources when possible.
- Ensure bibliography entries exactly match in-text citations

Custom Educational Bots

Creating specialized AI assistants tailored to specific courses or subjects, loaded with relevant knowledge and configured to respond appropriately to student questions.

Agents

Manus (manus.im): An agentic AI system that can work independently, using multiple models and tools to complete complex tasks like creating presentations or analyzing data.

Perhaps the most crucial approach is an emphasis on ethical AI use and academic integrity. One tool to consider is the AI Assessment Scale, a five-level framework for integrating AI into coursework:

1. No AI: Assessments completed entirely without AI assistance
2. AI Planning: AI used for brainstorming and initial research only
3. AI Collaboration: AI assists with tasks, but students must critically evaluate outputs
4. Full AI: Extensive AI use with focus on directing AI effectively
5. AI Exploration: Creative AI use to solve complex problems

In teaching, you want to emphasize the importance of transparency, teaching students to provide acknowledgment statements that include:

• Tool identification and version
• Purpose and role of AI assistance
• Extent of use
• Confirmation of human oversight

Which could be converted into a sentence like so:

Portions of Sections 2 and 3 were drafted with the assistance of OpenAI’s ChatGPT-4, used to generate initial outlines and paraphrase complex passages. The author verified all facts, revised the text for coherence, and takes full responsibility for the final content.

Legal and Security Considerations

For educators working in institutional settings, there are critical compliance issues:

• FERPA requirements mandate careful handling of student data, with specific protocols for cloud storage and access controls.
• HIPAA considerations apply when dealing with health-related information, requiring encryption and audit controls.
• Data security involves understanding institutional policies about where sensitive information can be stored and processed.

At DePaul, we recommend that faculty use institutional tools like Microsoft Copilot that offer encrypted chats and do not use conversations for model training.

1. Experiment with prompting frameworks using free and paid tools like ChatGPT, Claude, and Gemini.
2. Try document analysis with tools like NotebookLM for research or lesson planning.
3. Develop institutional policies for AI use that balance innovation with integrity.
4. Create acknowledgment protocols that promote transparency in AI-assisted work.
5. Build AI literacy through hands-on experimentation and continuous learning.

These four free MOOCs (Massive Online Open Courses) are highly recommended for beginners:

1. Generative AI for Educators (Google)
2. Elements of AI
3. Building AI
4. Ethics of AI