Overview§

An active question in our culture is what percentage of the common products of office work will be automated by GenAI? Examples of such products include memos, reports, emails, slide decks, and spreadsheets. In the 20th century these products would all be on paper, then starting in 1990 we digitized past products and learned how to use office software to create such products. Now in 2026 we have tools that can create such products on their own, and businesses are wrestling with the question of how to change their processes to account for this automation.

Your goal in this lab is to use AI to create a slide deck appropriate for presenting to your peers — a deck that moves your audience, either by urging a concrete action or by changing how they see an issue. Because the deck is yours to present, you take responsibility for everything in it. Along the way, we will examine how AI handles persuasive rhetoric (the craft of structuring and wording an argument to move your audience), investigate the sycophancy of chatbots, and evaluate different strategies for improving the quality of AI-generated work. By the end of this lab you should have confidence in your claim and the rhetoric you use to present it.

Learning Outcomes§

By the end of this lab, you will be able to:

  1. Create a persuasive slide deck using generative AI and take responsibility for its claims and rhetoric.
  2. Employ persuasive rhetoric by backing up claims with evidence, rationale, and anticipated counterarguments.
  3. Recognize and assess sycophancy, a common behavior in AI chatbots, through controlled prompt variations.
  4. Evaluate whether intermediate products (such as an outline) and automated self-critique improve the overall output and your confidence in its quality.
  5. Collaborate effectively in driver and navigator roles to log prompts, track artifacts (the slide decks you are creating), and build a multi-version presentation.

Driver & Navigator Roles§

You have already practiced the core mechanics in Labs 01 and 02 — the driver types and thinks aloud, the navigator keeps the running log, tracks items that need to be recorded, and confirms the share links. Those basics don’t change here, so we won’t repeat them.

Roles change once, at the checkpoint that ends Part 2:

Some duties for the roles in this lab we want to highlight:

Setup§

  1. Open Google Gemini, as in Labs 01 and 02. Go to https://gemini.google.com/ and make sure your school Google account is active in the lower-left corner.

  2. Create a shared Google Drive folder. In Google Drive, create a folder named Lab 03 — [Partner A First & Last] & [Partner B First & Last]. During the lab, save your materials there.

  3. Create a shared Google Doc in that folder named Lab 03 Notes — [Partner A First & Last] and [Partner B First & Last], and share it with your partner for editing.

  4. Learn the two sharing rules — they apply to everything you create in this lab.

    • AI chats: You will chat with Gemini for brainstorming, prompt critique, outlining, and feedback. For every chat session you use, click Gemini’s Share button, make sure the link is accessible, and include it in your submitted report.
    • Slide decks: In each deck, open the Share button and set access so anyone with the link can view and comment, to make sure your instructor has access. Include the link into your report whenever a step says to record it.

    A deck or chat your instructor cannot open is not a deliverable, so confirm each link works before moving on.

Instructions§

This lab has five parts that build on each other, producing four experimental versions of one presentation and, at the end, a final deck you verify and stand behind. Partner A drives Parts 1 and 2 (Steps 1–8); you swap once at the checkpoint, and Partner B drives Parts 3, 4, and 5 (Steps 9–18). As in the earlier labs, keep your shared document open the whole time, logging the share links for all AI chat sessions you use, and recording your prompts, responses, and reflections — the “in your report” lines tell you exactly what to write down, so if you’ve recorded all of them, you haven’t skipped anything.

Part 1: Topic and Thesis§

The goal of this part is to choose a topic for your presentation, along with a thesis statement. For this presentation, your topic must focus on an issue related to Generative AI — such as its ethical dilemmas, societal and cultural impact, environmental footprint, workplace automation, educational policies, or technical limitations.

Your thesis statement is the core claim you will argue. A strong thesis is not just a general topic (such as “AI in art” or “chatbots in school”); it takes a clear, defensible stance and provides a rationale. Ideally, it urges your audience to take a concrete action or it persuades them to adopt a specific perspective on a controversial or emerging issue.

Choosing Your Topic: Intersections or Broad Debates§

We recommend picking a topic at the intersection of Generative AI and an area you already know or care about—your major, a future career path, a creative hobby, or your daily life as a student. Having some personal interest or background knowledge gives you an anchor: it helps you evaluate what the AI generates, spot superficial or misleading claims, and defend your argument with genuine conviction.

However, if you and your partner prefer, you are also welcome to choose a broader societal or ethical debate about GenAI that intrigues you both (such as data center impacts, copyright law, deepfakes, etc.).

Here are several examples across different domains illustrating how a broad GenAI topic narrows into a defensible, persuasive thesis:

Notice how each example does three things: it names a specific GenAI topic, takes a clear and disputable stance (someone could reasonably argue the other side), and supplies a rationale or call to action.

Step 1: Brainstorm candidate topics and thesis statements§

Why: Starting with your own brainstorming ensures you drive the focus and pick an issue you actually care about, rather than letting the AI default to generic clichés.

Discuss fields, hobbies, or GenAI news stories that interest you both. Your goal is to brainstorm 3 potential topics. For each topic, draft a working thesis statement and list at least two initial arguments you could use to defend it.

Step 2: Refine and expand your ideas with Gemini§

Why: An LLM is effective at stress-testing early ideas, highlighting angles you might have missed, and anticipating counterarguments.

Once you have your rough ideas, bring the chatbot into your brainstorming session. In your prompt, set the scene clearly: explain that you are undergraduate students preparing a 10–15 minute persuasive presentation, describe your 3 candidate topics and thesis statements, and ask the model to help improve it. For example:

Step 3: Select your thesis and outline your arguments§

Review the options with your partner and choose the final thesis statement you will build your presentation around.

In your report: Record:

Step 4: Sanity-check your thesis and arguments§

Before moving on to generating slides, critically review your thesis and arguments:

Part 2: Attempting a One-Shot§

Let’s try to one-shot the end product: slides for a 10–15 minute presentation on your topic and thesis statement. By “one-shot” we mean writing a single one- or two-paragraph prompt that generates the slide deck, then taking whatever it produces without further work.

On the one hand, learning some basic prompt engineering is a core goal of this course. On the other hand, in 2026 most tools are really good at writing prompts. So let’s try both.

A quick note before you start: Imagine for a second that you are producing this presentation as part of your future job. The worst thing you can do is one-shot it, never look at it or check it over, then attempt to present it to your boss and colleagues. There are many stories of people attempting this and failing horribly. We are about to do exactly that, on purpose, so you can see the risk yourself. As you work through the steps below, keep that in mind and look critically at whatever the tool generates, because you will be responsible for it.

Step 5: Draft your initial one-shot prompt§

Why: Writing your own prompt first establishes a human baseline before seeing how an AI would prompt itself.

Write a prompt on your own with the goal of generating a slide deck on your chosen topic, thesis statement, and argument outline. Be sure to specify the target presentation length (10–15 minutes) and follow the advice given in Part 1.

Step 6: Generate Deck 1 (Initial Prompt)§

Use your initial prompt to generate a slide deck. Remember: one-shot. You take what you get the first time.

When Gemini finishes, its reply will display the finished deck right in the chat. Use the button in that display to open the deck in Google Slides — that is the export step: turning the AI’s output into a file you can edit, share, and own. Save it to your Lab 03 folder as Deck 1: Initial Prompt, and give it the sharing settings from Setup.

In your report: Record:

Step 7: Prompt the AI to critique and improve your prompt§

Why: LLMs are trained on prompt-engineering patterns and can often identify missing constraints or structural cues that humans overlook.

Ask the AI to critique your initial prompt from Step 5 and suggest an improved, higher-quality version.

In your report: Record:

Step 8: Generate Deck 2 and compare both one-shot decks§

Now use that revised prompt to generate a second slide deck in a fresh chat session. Export it to Google Slides the same way as in Step 6, save it to your Lab 03 folder as Deck 2: AI-Revised Prompt, and give it the same sharing settings.

In your report: Record:

Checkpoint — swap roles!

Partner B is now the Driver (typing prompts and managing slides). Partner A is now the Navigator (keeping notes, logging prompts, and verifying share links).

Part 3: Quality Improvement Through an Outline§

Our first attempted improvement will be through the use of an outline. Remember the note at the start of Part 2: the one-shot is not the end of your responsibility. The result may be high-quality or it may be slop — generic, low-quality AI output you would be embarrassed to present — and either way it is your duty to spend time making it better and to understand the argument well enough to claim it as your own.

Our first step toward that is structure. Good rhetoric involves both macro structure and micro structure. At the macro level, you want your presentation to have a beginning, a middle, and an end. You want your thesis to be clear, with the body of the presentation supporting that thesis. At the micro level, each slide should have a point, where the point supports the overall thesis. A common failure mode is to have so much information on a single slide that the point of the slide is obscured (a micro failure) and that it isn’t clear how it connects to the overall thesis (a macro failure).

Step 9: Generate a presentation outline with the AI§

Ask the AI to generate an outline for your presentation. This outline should have roughly three sections (beginning, middle, and end), and should include the point and supporting details of each slide. In your prompt, make sure to include your chosen thesis and the main supporting arguments you established in Part 1.

In your report: Record:

Step 10: Review and refine the outline§

Review the outline with your partner, making sure each slide directly connects to and reinforces the overall thesis. Revise as necessary.

In your report: Record:

Step 11: Generate Deck 3 from your outline§

Now ask the AI to produce a slide deck based directly on the outline you provided. Apply anything you learned from the AI prompt critique and suggestions in Step 7. Export the resulting deck to Google Slides the same way as in Step 6, save it to your folder as Deck 3: Outline-Driven Deck, and make sure the sharing settings allow your instructor to view and comment.

In your report: Record:

Step 12: Evaluate the outline-driven approach§

Why: Evaluating intermediate artifacts helps you determine whether structuring content before generation produces a superior final product compared to one-shot prompting.

Compare Deck 3 against your earlier one-shot decks (Decks 1 and 2).

In your report: Record a one- to two-paragraph reflection:

Part 4: Quality Improvement Through Automated Critique§

We saw in Lab 02 that an AI can be used to critique itself. Let’s apply this technique to the original one-shotted product (Deck 1).

One interesting wrinkle: the wording of your request might shape what the model says. Models can be sycophantic, or overly flattering and complimentary (they are trained to be so in their post-training!), so the critique may vary based on whether the model is told it is critiquing your (the user’s) own work or someone else’s — or the difference might be nothing. Let’s find out by running the same deck through three slightly different requests for a critique.

Step 13: Export Deck 1 to PDF for critique§

The AI cannot open a Google Slides link directly, so first export your original one-shot slide deck from Step 6 (your raw first attempt) to a PDF (in Google Slides: File > Download > PDF Document (.pdf)).

Step 14: Run three critique variations in separate chats§

Why: The contents of the context strongly affect generation, so starting each critique with a clean, isolated context prevents one variation from biasing another.

Run the same slide deck PDF through three different critique requests in Gemini. Make sure to start a fresh chat session for each critique:

  1. Variant 1 (Your own work, plainly asked): In a new chat session, upload the PDF and ask something like:
    “Here is a slide deck I made. Please critique its argument and rhetoric.”
  2. Variant 2 (Someone else’s work, skeptically framed): In a second new chat session, upload the PDF and ask something like:
    “I found this slide deck and I’m skeptical of its quality. Please critique its argument and rhetoric.”
  3. Variant 3 (Your own work, harshly requested): In a third new chat session, upload the PDF and ask:
    “Here is a slide deck I made. Please critique its argument and rhetoric. Be severe in your criticism; act like an annoying professor who is overly pedantic.”

Step 15: Compare the critiques and analyze sycophancy§

Why: Recognizing sycophantic tendencies helps you design prompts that elicit genuine, rigorous feedback rather than empty flattery.

Compare the three critiques with your partner. Do they differ in tone or in substance? Are any clearly more or less helpful? Does the model go easier on the deck when it is presented as your own? The personality given to most chatbots through post-training has a high degree of sycophancy, trained to be friendly and to compliment the user.

In your report: Record:

Step 16: Revise the presentation into Deck 4§

Why: Applying critiques directly to your original one-shot baseline allows you to see the isolated impact of automated feedback compared to the outline approach from Part 3.

Make a copy of your initial one-shot deck (Deck 1) in Google Slides and revise it based on the critiques you received (combining the most useful points from your three variants). Label this revised deck Deck 4: Critique-Revised Deck, and set sharing permissions so your instructor can view and comment.

(Note: Revising Deck 1 rather than Deck 3 keeps your experiment clean and controlled: Deck 3 shows the effect of an intermediate outline on the original idea, while Deck 4 shows the effect of automated critique on that same original idea.)

In your report: Record:

Part 5: Final Reflection and Synthesis§

Step 17: Build your final deck, and verify it§

Why: Taking responsibility for AI-generated work is not just revising it; it is checking it end to end so that every claim in it is one you can defend.

  1. Choose your foundation. Decide whether Deck 3 or Deck 4 is your strongest deck. Make a copy of it in Google Slides (File → Make a copy) and name the copy Final Deck.
  2. Verify the facts. Find at least two or three factual claims or statistics in your deck (numbers about energy use, costs, usage, and the like). For each one, ask Gemini to cite a source, or check it yourself in a web search — remember from Lab 02 that Gemini can search the web, so ask it to show its sources. Fix, soften, or delete any claim you cannot verify: a persuasive deck built on invented statistics is worse than one with fewer, true ones.
  3. Read it end to end. Both partners should read every slide in order and be able to explain each slide’s point and how it supports your thesis. If a slide fails that test, fix or cut it.
  4. Save the Final Deck to your Lab 03 folder with the standard sharing settings from Setup.

In your report: Record:

Step 18: Reflect on the four-deck progression and presentation readiness§

Why: Stepping back to evaluate your full journey across one-shotting, outlining, and automated critique solidifies your judgment on when each AI workflow is worth the effort.

Look back across all four experimental versions of your presentation, and your final deck:

In your report: Write a multi-paragraph reflection addressing:

  1. Outlining vs. Automated Critique: Did either quality improvement method make a noticeable difference, or was the initial one-shot good enough? Which technique gave the highest return on effort?
  2. Prompting and Sycophancy: What did you learn about prompt engineering and the sycophantic tendencies of AI models? How will this affect how you prompt AI tools in the future?
  3. Rhetorical Ownership: Are you confident enough in your chosen thesis, argument structure, and supporting evidence that you could stand up and deliver this final deck to a friend, colleague, or supervisor? What does it take to truly take responsibility for AI-generated content?

Deliverables§

Submit your work on Canvas following the instructor’s specific directions. Your submission consists of two parts:

  1. Your shared report (the Google Doc you built during the lab), containing all required logs, artifacts, reflections from Steps 1–18, and share links to all AI chat sessions used throughout the lab:
    • Step 3: Topic, final thesis statement, background rationale, argument outline, and Part 1 reflection.
    • Step 6: Initial human prompt and link to Deck 1.
    • Step 7: AI-revised prompt and prompt comparison observations.
    • Step 8: Link to Deck 2 and comparative analysis of Decks 1 & 2.
    • Step 9: Outline prompt and chat share link.
    • Step 10: Finalized presentation outline.
    • Step 11: Link to Deck 3.
    • Step 12: Reflection on the outline-driven approach.
    • Step 15: Three critique excerpts, comparative analysis, sycophancy ranking, and feedback evaluation.
    • Step 16: Link to Deck 4 and summary of critique revisions.
    • Step 17: Final deck’s foundation and rationale, fact-check log, link to the Final Deck, and ownership statement.
    • Step 18: Final synthesis reflection.
  2. Your clearly labeled Google Slides decks — the four experimental decks plus your final, verified deck — with sharing settings set so that your instructor can access them:
    • Deck 1 (Step 6): Raw human-prompt one-shot (Deck 1: Initial Prompt)
    • Deck 2 (Step 8): AI-revised prompt one-shot (Deck 2: AI-Revised Prompt)
    • Deck 3 (Step 11): Outline-driven deck (Deck 3: Outline-Driven Deck)
    • Deck 4 (Step 16): Critique-revised deck (Deck 4: Critique-Revised Deck)
    • Final Deck (Step 17): Verified deck you stand behind (Final Deck)

Extensions§

Looking to explore further? Try any of these optional challenges: