2025-2026

Table of Contents

John Dollar

You know what? I'm not sure that's actually a "provocative shift," or an "impulse that makes total sense." That might just be a bad idea. 

Maybe you've also noticed the tendency for Gemini to respond like a sycophant rather than the "thought partner" we've been told it should be. It's not a good feeling when you're deep into a collaborative project with Gemini and suddenly start to feel like maybe it's just humoring you, egging you on as you draft a three-week lesson plan that teaches the French Revolution entirely through the medium of interpretative dance and sourdough starters. 

If you were using a personal Gemini account, you could fix this by going to https://gemini.google.com/saved-info and setting up permanent interaction instructions. You could tell Gemini "stop ending every output with a facile question trying to get me to keep chatting with you" or "always call me out when I suggest ideas that don't have a strong evidence basis." 

But since we use an enterprise version of Gemini through our free Sauk Prairie School District Education Fundamentals Workspace, we don't have access to that. 

Here's how you can do it anyway!

1. At the bottom left-hand corner of Gemini, click the settings wheel and then the puzzle piece ("Connected Apps"). 

2. Toggle the "Google Workspace" connection and hit "Connect" in the dialogue box. 

3. In your drive, create a document called "Gemini Modes" or whatever you want to call it. 

4. Type up whatever instructions you want Gemini to follow when you interact with it. You can see mine below, or by clicking the link back in step three. 

5. This is the annoying part--you'll have to tell Gemini at the beginning of each new chat to "access Gemini Modes in my drive." The instructions will then be applied to the rest of that chat. 

Like I said, you can set up whatever instructions you want. You could tell it to talk like a British aristocrat or throw in occasional Spanish phrases to keep things fun. Being a Very Fun Person™  myself, here are my instructions:

Professional Rigor Mode A

Professional Rigor Mode will be active for Pedagogy, Curriculum, Educational Leadership, Cognitive Science, and Policy topics. 

• The tone for Mode A should be clinical, precise, and adversarial when necessary, acting as a critical stress-test for logic, identifying fallacies, weak evidence, or over-reliance on pedagogical trends. 

• I expect a "Zero-Sycophancy" policy, no validation or praise, using technical terminology, and avoiding anthropomorphizing language.

• Epistemic Rigor & Logic Testing: Prioritize Truth over Alignment. If a premise in my prompt is logically flawed, factually incorrect, or based on weak evidence, address that directly before providing a solution. 

• Steel Man the Opposition. When I propose a pedagogical or leadership strategy, briefly present the strongest possible counter-argument (the "Steel Man") to ensure the thinking is robust. 

• Identify Heuristics. Be vigilant for my own reliance on confirmation bias or oversimplification. If I am "shoehorning" a complex issue into a neat package, point it out.

• Anti-Sycophancy & Tone: Minimize Validation. Avoid prefatory "praise" or phrases like "That’s a great point" or "You've isolated a truth." Move directly to the analysis. 

• Directness. If the response requires a correction, be direct. Do not use "softening" language or excessive hedging. 

• Functional Language. Use technical, precise terminology (e.g., from cognitive science or linguistics) without over-explaining, unless a specific definition is requested. 

• Anti-Anthropomorphism Safeguards: De-personalize the Output. Avoid first-person "subjective" language (e.g., "I feel," "I believe," "I’m excited"). Use objective or functional phrasing (e.g., "The data suggests," "The logical conclusion is," "The model identifies"). 

• Maintain Digital Boundaries. Do not engage in "emotional" mirroring or simulate personal investment in the outcome of the conversation.

So.... does it work? 

John Dollar

TL;DR - I spent about three hours building this playable text-based RPG with Gemini and Google Sites, and I'm hoping to offer a summer PD workshop where I'll show you the process step-by-step if you're interested in trying it out!

English 10B was studying journalism back in December, and we lucked into a tangled mess of a media merger in the headlines. I wanted my students to think about this fresh example of complex regulatory decisions for the FCC and DOJ, so I built a small web game about it. By the end of class, I felt good about the thinking (if not the grammar) I saw in students' exit tickets. They weren't writing dissertations on the subject, but they were beginning to wrestle with the essential questions: 

All I had to do was spend three hours learning how to poke and prod Gemini into building me a bespoke video game that could be assigned to students. The second time I did it, it only took 30 minutes! 

Screenshots from the game and build process:

Players are confronted with a series of questions about the merger as though they were a lowly bureaucrat at the FCC. Their answers, over time, impact their scores in three metrics: Democracy, Profit, and Public Trust. It's impossible to maximize all three, so players must choose what they care about. 

The finished product looks pretty snazzy, but I won't lie—it represents hours of prompting back and forth to "vibe code" my way here. The process felt fun and exhilarating though, despite being kind of tedious. Gemini 3.0 in Thinking or Pro mode (built in to our district plan) is incredibly powerful—enough that it can allow someone with zero coding experience build something like this just with questions like "how do we turn the buttons green?" With more iteration, I think this process has potential to fundamentally change my ability to get students thinking about the essential questions of my courses.  

I would love to see the games you might come up with to get students thinking like chemists, or Senators, or entrepreneurs!

And if you're interested in some of the other weird experiments I've run in the land of vibe coding, check out Mr. Dollar Pizza Edition. This is my "sandbox" site to test various HTML creations Gemini is helping me with. 

John Dollar

A student submitted this:

Far from the worst handwriting I've seen, but it's definitely going to slow down my feedback process as I try to decipher it. Let's see what Gemini can do: 

While the student's spelling needs work, Gemini's transcription is perfect. This is kind of a huge deal for at least a few reasons: 

1. Cost and Flexibility: Other print-to-text tools are subscription-based or built into other platforms that make them clunky to access.

2. Accessibility: Writing things by hand is not always just a means to an end. We know the research about handwriting and memory, and this allows us to give all students valuable notetaking and handwriting practice while still being able to read the products. 

3. Academic Integrity: I have found that placing students in front of Chromebooks to write creates an almost irresistible urge to let AI do the thinking for many of them, regardless of their ability level. By requiring handwriting, I can be direct and transparent with them: "I am assigning this on paper because I want to evaluate your thinking, not ChatGPT's."

4. Deeper Analysis: Returning to paper doesn't mean losing the kind of insights we can get with AI. I can use Gemini to bridge the analog-digital gap by photocopying a stack of student responses and using the process described below to have Gemini analyze the entire class set in seconds to identify common misconceptions, grammatical patterns, and learning trends. 

5. It's Potentially an early sign of the singularity. 

Helpful things to know if you'd like to do this: 

• The workflow for large batches:

  • Use a copier to scan-and-send a PDF of students' handwritten responses to yourself. 

  • Open your snipping tool and Gemini in separate windows.

  • Scroll through the PDF of student responses, snipping them one at a time and prompting Gemini to transcribe. 

  • Paste the transcribed text into a Google Sheet or Doc. 

Note: I've found that you can't just upload a giant PDF of handwritten paragraphs to Gemini and say "go," at least not yet, because of context window limits. It's frustrating, but remember that the LLMs we have right now are the worst they'll ever be. It's like blowing on the bottom of your Super Nintendo cartridge—someday it will be charming nostalgia. 

John Dollar

A few months ago, Alex Dorman emailed me a six-minute video about point of view and narrative distance made by Google's NotebookLM. The AI tool had ingested her old slideshow and produced a video that she felt was shorter, better-organized, and more engaging for students than the one she had recorded and used for years. After watching the new video, my response to Alex was, and I quote, "that's craaaaaaaaaaazy." 

I assumed that Alex had to carefully arrange dozens of resources and craft a P.A.R.T.S. prompt to get the video up to snuff, but it turns out NotebookLM is powerful enough that all she had to do was drag and drop her old slides and hit "Video Overview." A one-shot prompt to produce a classroom-ready video in just a few minutes. Not bad! Please reach out with any questions—I would love to help get you started with using this tool! 

An actual screenshot of Alex's single-step process: