Rethinking a Scaffolded Research Paper in the Age of GenAI
Course and Assessment Context
POLSCI 302 – Public Opinion examines how political opinions are formed, how they change over time, and how they shape political attitudes and behavior. The course approaches public opinion analytically by exposing students to theoretical perspectives, empirical research, and normative debates about democracy, representation, media, and the role of political knowledge. A central learning goal is not only to understand public opinion, but to evaluate how it is studied, especially through survey research and big data.
To reflect this emphasis on reasoning and evaluation, the major assessment in the course (i.e. 60%) is a scaffolded research project rather than a single end-of-semester paper.
Assessment Structure
The final research paper (8–10 pages, worth 20%) integrates several earlier components:
- Research Question Memo (5%)
- Theory & Mechanism Map (10%)
- Measurement & Data Critique (10%)
- Research Presentation (15%)
Students are not required to collect data or run statistical analyses. Instead, they are asked to:
- Clearly define their outcome and explanatory variables
- Explain how their argument works step by step
- Specify assumptions and scope conditions
- Identify a plausible alternative explanation
- Propose measurement strategies and discuss what those measures capture, as well as what they miss
- Be explicit about what kinds of conclusions their design can and cannot support
The final paper must integrate and revise earlier components, and substantial departures must be justified. The rubrics emphasize conceptual clarity, logical coherence, alignment across components, and awareness of inferential limits. In short, the project evaluates structured reasoning over stylistic polish.
What Questions Did GenAI Raise?
GenAI poses an important question: If students can generate extensive literature reviews, well-defined research questions, and polished explanations instantly, what exactly are we assessing?
At a surface level, many elements of a research paper can now be produced quickly:
- Summaries of existing scholarship
- Clearly phrased hypotheses
- Generic survey measures
- Smooth introductions
But the scaffolded design revealed something important. What remains difficult to outsource is not writing, it is sustained conceptual coherence:
- Maintaining stable definitions across assignments
- Ensuring that each step of an argument logically depends on the previous one
- Distinguishing conceptually between competing explanations
- Acknowledging what measurement choices systematically exclude
- Preserving alignment from memo to presentation to final paper
GenAI can produce arguments. It struggles to preserve intellectual commitments across stages unless the student understands those commitments well enough to enforce consistency.
To further reinforce this, I incorporated structured Q&A sessions in class. After key assignments, particularly the theory map and research presentation, students were asked to explain their choices, clarify distinctions, and respond to probing questions from peers and from me. These exchanges often revealed whether students genuinely understood their own argument or were relying purely on GenAI.
What I Adjusted or Emphasized
Rather than banning GenAI entirely, I leaned more deliberately into the structure already embedded in the course.
1. Making Argument Structure Visible
The Theory & Mechanism Map assignment asks students to do more than state a claim. They must identify the main outcome and explanatory factor, explain the logic of their argument step by step, specify key assumptions and scope conditions, and spell out at least one plausible alternative explanation, along with how it differs conceptually from their own. Because students then revise and integrate this material into the final research paper, the assignment creates a built-in consistency check.
If a student relies on AI-generated text without fully understanding it, inconsistencies tend to appear: definitions drift, the mechanism changes across sections, or the alternative explanation is no longer meaningfully distinct.
2. Emphasizing Measurement Trade-Offs
In the Measurement & Data Critique assignment, students must distinguish between concepts and indicators, explain what their proposed measures capture well, discuss validity and bias concerns, and identify at least one claim that would be too strong given their design.
In class, we discuss specific measurement strategies commonly used in public opinion research. For example, political knowledge scales, feeling thermometers, ideological self-placement, media exposure measures, and big-data proxies such as digital traces. Students examine what these measures capture well, what they systematically miss, and how their choice shapes the kinds of claims they can responsibly make.
AI can propose a multitude of plausible measures within seconds. What it does not automatically provide is a disciplined evaluation of trade-offs. Deciding which measure best aligns with a concept and being explicit about its limitations, requires judgment that goes beyond generating options.
3. Requiring Public Defense of Ideas
The Research Presentation requires students to explain their project’s logic and respond to questions about their choices. Students who relied heavily (and exclusively) on AI often struggled when asked to clarify distinctions or defend trade-offs in real time. Oral explanation became an important complement to written work.
How Did It Go?
Two patterns stood out.
First, stronger students used GenAI as a drafting aid but improved their work through revision. The scaffolding helped them refine definitions, sharpen distinctions, and think more carefully about alignment across sections.
Second, students who relied heavily on AI often produced smooth and well-structured prose, but when asked to explain why one explanation differed conceptually from another, or what their measure failed to capture, they struggle to answer.
In sum, GenAI did not eliminate the research paper. It clarified what intellectual work the assignment was actually designed to assess.
One Thing I’d Tell a Colleague
If what earns high marks is stylistic clarity, well-organized prose, and the ability to restate established arguments, GenAI can meet, and sometimes exceed that bar with very little effort.
If your assessment rewards:
- Clear conceptual definitions
- Structured explanation of how an argument works
- Explicit contrast with alternatives
- Honest reflection on measurement limits
- Coherence across multiple stages
then GenAI becomes a tool rather than a substitute.
For me, GenAI did not require abandoning the research paper. It required making the structure of reasoning more explicit and more central to how I evaluate student work. In that sense, GenAI did not force me to redesign the assignment. It forced me to articulate more explicitly what intellectual work the assignment was always meant to be.
This is part of the collection of sharing from members of the 2025-26 Faculty Learning Community: Assessment in the Age of GenAI.
