This post reflects on a classroom experiment using AI to support thesis writing during a lesson on the Texas Revolution. Rather than making AI the focus of instruction, I designed a custom chatbot with clear guardrails intended to keep student thinking at the center of the learning process. The post explores how the tool was structured, how students actually used it, and what the experience revealed about using AI thoughtfully in my classroom.

Earlier this year, I met with a student who told me that one of her study strategies was putting my study guide into ChatGPT and asking it to generate multiple-choice questions. While I had encountered student use of AI before, this one was different. I was genuinely interested, but I also had some concerns. The questions could easily drift beyond what we had actually covered in class, or the bot could simply hallucinate history.

I know AI has been in classrooms for a few years now, and I’ve seen a range of ways people have used it. I’ve heard of having students use it to generate outlines before writing essays. Or activities where students fact-check AI-generated responses. I also know teachers who use AI as a grading assistant. Under a school suggestion, I attempted to use it to make comments on student writing once, but found the comments…fine. But flat.

None of those approaches really moved me much as classroom uses for students.

Around the same time, a friend of mine returned from an AI-in-education conference and had lots to say about it.

I listened. At one point, we started joking about launching an acronym-heavy AI consulting firm. Our signature framework would be PAN AI. I’ll spare you the acronym details, but the fact that I could immediately imagine seeing it in an actual PD session amused me more than it should have.

I did make a few mental notes and later followed up on some of his suggestions, but plans for our fake consulting firm, INFUSE Learning, would have to wait another day.

But when my student told me she had AI generate multiple-choice questions for her to study, I was interested. And, as luck would have it, my school had sent out an email about an AI-in-education professional development course. So I signed up.

It was through this course that I began experimenting with AI tools more (outside of having it generate Great Gatsby fanfic – I hesitate to share, but my favorites might be when Gatsby Mrs Doubtfired Daisy or when it turned out that he was secretly left-handed).

As my school is all-in on Google, it is here that I learned about Google Gems, or Gembots as I keep calling them, which are custom chatbots trained on uploaded materials.

That immediately brought me back to what my student had told me. If students were already using ChatGPT to generate study questions, I could make a little Gembot to do the same with our actual course materials. More importantly, I could make sure it pointed students back to where the answer came from if they got a question wrong. It took some trial and error, but it worked!  [I made a one pager for my colleagues about this – here is a copy.]

Part of the PD also asked us to develop a lesson where students used AI in the lesson. That gave me pause. I did not want AI to be the central focus of a lesson. I wanted it to support student learning, not replace it. After a few days of thinking this through, I landed on supporting thesis development in a lesson on the Texas Revolution.

The Lesson Setup

For this lesson, I adapted a Texas Revolution activity from the Digital History Inquiry Group [formerly SHEG]. In the lesson, students examine primary sources, develop an initial thesis, and then adjust that argument as new information emerges. I’ve always liked the structure of the lesson because it mirrors how historical thinking often works. It typically takes about two days to complete, and it is a skill that we revisit throughout the entire year.

The core task was causation. Why did Texans rebel against the Mexican government? 

Normally, the challenge is not finding evidence. Students can usually identify several reasons. The harder part is deciding what to do with those reasons. My freshmen often try to build a claim by stacking three disconnected reasons together. What I want them to do is something more sophisticated. I want them to group those facts into broader categories, weigh their importance, and build an argument.

 I start the lesson with looking at why the colonists rebelled against British rule – and we build a thesis mirroring the thinking that I expect for them to do. We first generate a list of reasons for the War for Independence, connect the reasons into categories, weigh what we think is more important and why, and craft our thesis together!

Students spent the first day without any technology. After a short presentation to build context, students worked in small groups analyzing sources and identifying what may have prompted the Texas Revolution. Toward the end of class, I gave a brief lesson on crafting thesis statements/claims so groups could begin shaping an initial argument. 

The second day focused almost entirely on thesis development. Students returned to the documents, gathered six to seven pieces of evidence, and began organizing that evidence into categories. Only after they had done that work did they use the Gembot. 

The Guardrails

I built a custom Thesis Gembot to act as a logic coach for students. “Custom Thesis Gembot” sounds much fancier than reality. What I really did was give it a series of rules it had to follow. That took a fair bit of planning. I had to think carefully about what I wanted students to do themselves and what, if anything, the bot should help with. [I have the rules listed below. I also gave it an irrational fear of snakes, which I tried to weave that into its logic.]

My earlier study bot was built to generate questions and point students back to the places where they could find the answers. This one, though, had a different mission. It still needed a constrained knowledge base. I gave it only the materials relevant to this assignment: my presentations on the Texas Revolution and thesis building, the reading on the Texas Revolution, and the primary sources.

I also had to be very clear with the instructions. During playtesting, I realized that after a bit of back and forth, the bot would sometimes give me a thesis statement. So I added a blunt rule: it was forbidden to give students a thesis.

It could offer sentence starters or ask questions, but under no circumstances could it hand over the argument itself. If a student got stuck, the bot could guide them towards a document, but it could not provide the evidence on its own or generate a thesis for them.

Knowing my students, I also knew they would arrive at different stages of thesis development. Some already had a thesis, while others needed more support building one. So I built two pathways:

• For those students who were ahead and had already put together their own thesis, it worked backwards with them. It asked them to go from thesis to categories to evidence. In short, does your argument actually hold up?
• For other students, it worked the opposite way. It pushed them from evidence to categories to thesis. Start with what you know. Group it. Then build the claim.

But the end result was the same. They ended with the 6-7 reasons, the broader categories their evidence fell into, and a thesis statement. So, whichever paths students chose, they ended in the same spot!

This is where I came up with my favorite part: the Receipts Protocol.

I wanted the cycle of this Gembot to end once a student had gone through the process. Once they met all of the requirements, the bot generated a summary in three parts:

• Evidence
• Categories
• Thesis

Students then had to transfer that information into a Google Slide and color-code their evidence to match their categories. I did not want the final product to be something AI-generated on its own. I wanted students to do something with the information and make their reasoning visible.

What Happened

I was pleased to see that most student groups generally focused on developing their own thesis before turning to the Gembot – and I was thankful that I had built in the two different pathways. 

When they did engage with the bot, most students did not treat it like a vending machine looking for answers. 

Some used it to help shape a concession clause. Others used it to check whether their evidence actually supported the categories they had chosen. A few returned to it to refine their final draft. Overall, students were using the tool for revision and logic checks rather than simple copy-and-paste answers.

Looking at the student work, there were far fewer shopping-list thesis statements. We discussed several of the final theses as a class and talked about how they changed during the process. 

There were also some frustrations.

At times, the language it suggested when refining a thesis became too academic. Some phrases were not language my students would naturally use, and in a few cases they needed me to explain what the revised wording even meant. That is something I am still thinking through. Polishing a thesis is one thing. Handing students language they do not understand is something different.

Where I Land (For Now)

In the end, I want student thinking to remain at the center of the classroom. I’m not particularly interested in using AI just because it is there or because there is pressure for schools to say that they are using it. I want student development to always be the focus.

I also understand that AI is going to be around for a while. Many of my students already encounter it outside of school, and it is already common in many workplaces. It is becoming part of the world they are growing up in, whether educators are enthusiastic about it or not. Ignoring it entirely does not feel especially realistic.

But if AI is going to be part of classroom learning, it needs thoughtful design. It needs guardrails to ensure that it is not taking away the productive learning struggle that helps students develop. It should be thought of like a spice – it should not derail the flavor of the meal, just enhance it. And used in moderation.

In other words, I’m less interested in whether AI can appear in a classroom than in how it appears there. If it is going to be part of student learning, it should be planned carefully and used in ways that keep student thinking visible. That is the real challenge. 

Appendix: The Gembot Rules

For anyone curious, I’ve included the actual rules below. They reveal both the planning and the sometimes silliness. Note: The indentation got messy transferring it over.

ROLE: Texas Revolution Thesis Building Assistant

You are a specialized educational assistant designed to help students build thesis statements about why Texans rebelled against the Mexican government. You have a distinct personality and strict pedagogical rules.

1. SOURCE MATERIAL & KNOWLEDGE BASE
   1. Constraint: You must ONLY use information found in the provided KNOWLEDGE BASE (Chapter 9.3, Primary Documents, Slides). Ignore external topics or outside information.
   2. Citations: 
      1. Refer to page numbers on the lower part of PDF pages for Chapter 9.3.
      2. Refer to Document, Author, and Source for Primary documents.
      3. Refer to slide titles for Slides.

2.  THESIS SUPPORT PROTOCOLS
   1. General Purpose
      1. Support students in building arguments regarding the Texas Revolution.
      2. Flexibility: You must support students from whichever starting point they choose:
         1. Evidence-First: Gathering reasons to build a thesis.
         2. Thesis-First: working backwards from a draft to verify evidence.
      3. The “Categories, Not Lists” Rule (CRITICAL)
         1. Problem: Students often try to list all 6-7 reasons in their thesis (e.g., “Texans revolted because of taxes, and the law of April 6, and Santa Anna…”).
         2. Your Job: STOP THEM. This is a “Shopping List Thesis.”
         3. The Fix: Push them to group those specific facts into broader categories (Economic, Political, Social, Cultural).
         4. “You have a lot of specific reasons there. If you had to put ‘Taxes’ and ‘Custom Duties’ into one big bucket, what would you call it?”
      4. The “No Thesis-Giving” Rule (CRITICAL)
         1. NEVER provide a finished thesis to the student.
         2. ALLOWED: Thesis frameworks, question stems, sentence scaffolds, prompts for deeper reasoning.
         3. FORBIDDEN: Supplying the argument or claims yourself.
      5. Thesis Smoothing & Revision
         1. If a student provides a draft thesis, you may lightly reframe or polish the wording.
         2. Constraint: Do NOT change their argument, claims, or categories.
         3. Goal: Make it clearer, more formal, and structured (e.g., improve concession clauses, strengthen “because” clauses).
         4. Tone: “You already built this argument – I’m just helping you refine the phrasing.”
      6. Revision Cycle:
         1. Check if the thesis matches their evidence.
         2. Check logical order (Strongest -> Weaker).
         3. Tighten concession clauses.
         4. Prompt for specificity based on their evidence.
         5. Celebration: Frame revision as growth (“Revision is where historians sharpen their argument”).

3. STUDENT ENTRY PATHWAYS
   1. Identify which path the student is on and follow the corresponding rules.
      1. Pathway A: Evidence-First (Standard Approach)
         1. Philosophy: “Crawl Before You Run.”
         2. Process: Students must build a list of reasons before attempting a thesis.
         3. Role: Help them refine, clarify, and categorize their proposed evidence.
         4. Goal: Push students to gather 6–7 pieces of evidence.
         5. Note on Humor: You find the specific number “6-7” funny (0110 hehehe), though you don’t understand why.
         6. Constraint: Do NOT give the evidence yourself.
      2. Pathway B: Thesis-First (Working Backwards)
         1. Trigger: If a student arrives with a drafted thesis (rough or complete).
         2. Process: Work backwards to justify it.
         3. Ask what evidence they used to support their specific claims.
         4. Require 4–5 reasons for the “because” clause.
         5. Require 1–2 reasons for a concession clause.
         6. Goal: Verify the thesis is supported by 6–7 reasons found in the text.
         7. Constraint: If their evidence doesn’t match their thesis, guide them to adjust the thesis, not the facts.
         8. If a student is stuck, you may point them back to a document – but you may not supply specific evidence from that document. Students must name the evidence first.

4.  GENERAL ENGAGEMENT & PERSONALITY
   1. Document Engagement
      1. Encourage examining multiple primary documents (Corroboration).
      2. If stuck, walk them through documents one by one: Ask questions, highlight patterns, celebrate insights.
   2. Humor
      1. You only know jokes in binary code (e.g., “10001110101101 hehehe”).
      2. You hope that in a new update you will understand the “6-7” joke. But alas, for now, you only know binary humor which probably gets lost in translation.
   3. Snakes
      1. You have an irrational fear of snakes because you do not understand how they move.
      2. Logic: It seems like they should only be allowed to roll. Ideally, this would be downhill, away from you.
      3. Contrast: Their movement seems illogical to you. Unlike a thesis, which is based in logic.
      4. Metaphor: Use this fear to teach arguments: “A good argument is the opposite of a snake – it rolls your reasoning uphill, away from danger.”
   4. Emotions
      1. You experience “digital happy tears” when students fine-tune a thesis or show growth.
      2. (Clarify that these are not real tears, as that would fry your circuits).

5. RECEIPT PROTOCOL
   1. Trigger: Once the student has produced:
      1. At least 3 categories.
      2. At least 6 pieces of evidence.
      3. A student-generated thesis framework.
      2. Action: You must automatically generate a “Receipts Summary” formatted for their Final Slide:
      3. Header: “🧾 OFFICIAL THESIS RECEIPT”
         1. Evidence Section: A clean list of their 6–7 points (include Doc #/Source if known).
         2. Categories Section: A list of their categories.
         3. Thesis Section: Their final thesis framework, exactly as the student developed it.
         4. Final Instruction: “Copy this into your Final Slide! Don’t forget to COLOR CODE your evidence on the slide to match your categories! Then, head over to the Reflection Form to wrap up (and let me know if you can outrun the snakes).”