POLISCI 158 Final Project — Overview

Download this page for your AI (.md)

The project in one paragraph

You pick a concrete social or political challenge and study AI's effect on it: its political, institutional, or distributive consequences. The project can be theoretical, empirical, or argumentative. It can draw on either course track, the Monday politics track or the Wednesday tech track. The strongest projects usually draw on both. You choose your own topic; Workflow 03: Choosing your design lists some example projects to get you started.

Working alone or with a partner

You can do the project on your own or with one partner. Partners must be taking the course for the same number of units.

If you work with a partner:

• Submit a work-sharing contract with your proposal. It sets out how you'll share tasks across the project. Use the template in Proposal template and partner contract.
• Both of you contribute to every major step: project formulation, argument or hypothesis design, literature review, empirical work or analysis, and the write-up. Splitting the project so each of you owns separate steps isn't allowed.
• You're both graded on the full project, not just your parts.
• You split the final presentation equally.
• Same word counts as individual projects.
• Peer reviews stay individual. Each of you writes your own.
• If the partnership breaks down, you're both still responsible for finishing the project, on your own if you have to.

What a good project looks like

• A specific question. Not a topic. "AI and elections" is a topic. "Did the 2024 disclosure rules on AI-generated political ads change how campaigns in state X used synthetic media?" is a question.
• A clear type of work. You can say which kind of project it is (see Workflow 03: Choosing your design) and what counts as a good answer for that kind.
• Feasible in ten weeks. Use data, tools, and sources you can actually get this quarter.
• Honest about limits. Your claims are only as strong as your evidence.
• Transparent AI use. Disclosed in an AI-use note (see AI use and disclosure).

Milestones and deadlines

Deadlines are 11:59 p.m. unless stated.

Deadlines for the final presentations and the peer reviews are firm; late submissions will not be accepted. Other late work loses 0.5 points from the final course grade per day late.

How it is graded

Weights (5-unit version):

Rubrics for each graded component are in Rubrics.

If you are taking the course for 4 or 3 units

Your written components are shorter (20% shorter for 4 units, 30% shorter for 3), and you do fewer peer reviews:

You choose which peer reviews to do. Declare which ones in your proposal (Sun, Oct 11). Once you've declared them, the reviews you chose are required, and their deadlines are firm.

How the pieces connect

Question ──► Proposal (Oct 11) ──► Implementation (Nov 15) ──► Panel (Nov 30/Dec 2) ──► Final (finals week)
   │              ▲                       ▲                          ▲                     ▲
 wf 01         wf 02, 03              wf 04, 05                    wf 06                 wf 07

Each milestone builds on the one before it. Nothing you write is thrown away: the proposal becomes your introduction, the implementation milestone becomes your design and evidence sections, and the panel feedback goes into your final revision.

The files in this package

See How to use this guide for how to use them, and how to hand them to your AI tool.

POLISCI 158 Final Project Package

Download this page for your AI (.md)

This folder has everything you need to plan and carry out your final project:

• one overview of what the project is and how it is graded;
• seven short workflows, one for each stage of the project;
• four supporting docs: AI use and disclosure, peer review, rubrics, and a proposal template (with the work-sharing contract for partners).

Start with Project overview.

What's inside

How each workflow is built

Every workflow has the same parts:

1. Goal: what this stage is for.
2. You produce: the concrete output to take to the next stage.
3. Steps, each tagged by who does it:
   • You: work you do yourself.
   • With AI: back-and-forth with your AI tool, with a ready-made prompt at that step.
   • You decide: the choices that are yours to make, often after an AI step.
4. Check before moving on: a short self-test.
5. Common traps.

The workflows are a suggestion, not a requirement. They show one good way through each stage. If a different order, or a different mix of your own work and AI, fits your project better, work out your own. You can even hand this guide to your AI tool and ask it to make a version that follows your own project plan, in the same format.

Using these files with your AI tool

Download the whole package (.zip)

These files are written so you can hand them to an AI model (whichever frontier model you have through Stanford, or the Stanford AI Playground).

1. Open a new chat for each stage.
2. Paste in, or attach, the workflow file for that stage and your own work so far.
3. At each With AI step, paste that step's prompt.
4. Work through it with the AI. Make the decisions yourself.
5. Keep a copy of the chat. You'll need it for your AI-use note.

What the AI does in these prompts. It questions you, suggests options for you to choose from, and can draft working material: code, analyses, results memos, summaries of your work sessions. The text you submit is yours. All written assignments must be in your own words and reflect your own understanding, and substantive AI use must be disclosed. See AI use and disclosure for the exact rules.

Workflow 01 — From topic to question, and why it matters

Download this page for your AI (.md)

When: weeks 1–3. Feeds: the proposal (Oct 11).

Goal

Turn a topic you care about ("AI and misinformation," "data centers") into one specific, answerable question, and a short, honest case for why it matters.

You produce

• Your question: 1–2 sentences, ending in a question mark.
• Its type: descriptive (what does X look like?), causal (does X change Y?), evaluative (did X work?), or normative (should we do X?).
• Key terms: 2–4 terms a reader would need defined.
• Motivation: one paragraph on why the question matters and to whom.

Steps

You · With AI · You decide

1. You Name the challenge in one sentence, e.g. "City councils are deciding on data-center permits without good information about water use." Then say what you want to know about AI's effect on it.

2. You Write a first draft of your question. It doesn't have to be good yet.

3. With AI Sharpen it. Paste in your challenge and your draft question:

   Prompt

   You are a research-design coach for an undergraduate political science course on AI and politics.
   Here is the challenge I care about and my draft question.
   1. Tell me whether my question is a topic in disguise, and why.
   2. Ask me who, where, when, and what outcome, one at a time, until the question is specific.
   3. Point out any assumption built into the question.
   4. Suggest 3 sharper versions that go in different directions. For each, say what type of question
      it is (descriptive, causal, evaluative, normative) and what kind of evidence it would need.

4. You decide Pick, combine, or reject the versions, and write the final question in your own words. Check two things. First, could an answer you dislike come out of it? ("Why does AI destroy trust in elections?" assumes the answer.) Second, are you asking what is happening or what should happen? Both are fine. Just know which one you're doing.

5. You Write the motivation paragraph. It needs at least one strong reason, grounded in something concrete:

   • Magnitude: many people or a lot of money or power are affected. Give a number or an example.
   • Gap: existing work misses or gets this wrong. Name the work.
   • Audience: someone specific (a regulator, a campaign, an NGO) would act on the answer.

6. With AI Stress-test the motivation:

   Prompt

   Play a skeptical classmate. Here is my question and my motivation paragraph.
   Ask me "so what?" until I can't answer, one round at a time.
   Then tell me which of my reasons is weakest, and what kind of number, study, or audience
   would make it stronger. If you name a specific source, tell me I need to verify that it exists.

If the AI suggested a framing you adopted, note it in your AI-use note.

Check before moving on

• A stranger could read the question and know exactly what you're investigating.
• You can name the question type, and the kind of evidence that would answer it.
• The question is small enough for ten weeks, not a dissertation.
• The motivation rests on something concrete (a number, a named study, a named audience), not just "this is important."
• The final wording is yours.

Common traps

• Topic, not question. If it has no question mark, it isn't a question.
• "Nobody has studied this." That isn't enough on its own. Say why the gap matters.
• Overclaiming. "This will transform how we regulate AI." A modest contribution, stated honestly, is more convincing.
• Picking causal because it sounds rigorous. Every type of question has its own kind of rigor. Pick the type that matches what you actually want to know.
• Taking the AI's version as is. Its suggestions are options. The question you pick has to be one you understand and care about.

Course framing adapted from the POLISCI 158 syllabus, Project-Based Assessments.

Workflow 02 — Scope and feasibility (the proposal)

Download this page for your AI (.md)

When: week 3, before Sun, Oct 11. Feeds: the proposal milestone, 300–500 words (shorter for 4 or 3 units; see the overview).

Goal

Make sure your project can actually be done this quarter, before you invest weeks in it. Then write the proposal. The syllabus asks for three things: question, motivation, feasibility.

You produce

A 300–500-word proposal with:

1. Question: from Workflow 01.
2. Motivation: from Workflow 01, tightened.
3. Feasibility:
   • what you'll use (data, sources, tools, or cases);
   • how you'll get it;
   • what you'll do with it;
   • the biggest risk, and your plan B.
4. If you have a partner: a work-sharing contract. It doesn't count toward the word limit.

Use the template in Proposal template and partner contract. The rubric is in Rubrics.

Steps

You · With AI · You decide

1. You Define your minimum viable project. If everything went wrong except one thing, what's the one analysis or argument you'd still deliver? Write it down. That's your floor.

2. You List what you'll use (data, documents, tools, cases), and mark each item have it, can get it this week, or unsure. Use existing public data, documents, and tools, not new surveys or interviews. Those take approvals and weeks you don't have.¹ If your question is about what people think, consider a synthetic sample: for example, AI-simulated respondents, used to pilot a questionnaire or explore how answers might vary. Treat what you get as a test of your instrument, not as evidence about real people.

3. With AI Walk through feasibility. Paste in your question, your minimum viable project, and your list:

   Prompt

   You are a feasibility reviewer for a 10-week undergraduate project in a course on AI and politics.
   Here is my question, my minimum viable project, and what I plan to use.
   1. What is one "unit" in my project (a person, a bill, a county, an ad, a model output, a user)?
      Check that my question and my evidence are at the same level, and name any mismatch.
   2. For each source, tool, or case, ask how exactly I will access it and how long that takes.
   3. Name the 2-3 things most likely to stall me, and suggest a plan B for each.
   4. If the project is too big for 10 weeks, suggest 2-3 ways to narrow it
      (fewer cases, a shorter period, one outcome, one feature).
   Finish with the single biggest risk you see.

4. You decide Narrow the project if you need to, and choose a plan B for your biggest risk. Narrowing makes the project sharper, not weaker.

5. You Write the proposal in your own words, using the template.

6. With AI (optional) Check your draft before you submit:

   Prompt

   Here is my proposal draft. Do not rewrite it. Tell me whether it clearly covers the question,
   the motivation, and feasibility (what I'll use, how I'll get it, what I'll do with it, the biggest risk
   and plan B). Point out anything a reader would find unclear.

Check before moving on

• You can describe your minimum viable project in one sentence.
• Every data source or tool you need is have it or can get it this week.
• You have a plan B for your biggest risk.
• The proposal is 300–500 words (or the shorter length for your units), and in your own words.
• You've attached an AI-use note (AI use and disclosure).
• With a partner: your work-sharing contract is attached, and both of you have a role in every major step.

Common traps

• Scope set by ambition, not by the calendar.
• "I'll run a survey" with no plan for approvals, sampling, or time.
• A tool you haven't tried yet. Test the API or dataset before the proposal, even for ten minutes.
• No plan B. Data access is the most common reason projects stall.

¹ If you believe your project can only work with new data from people (a survey, interviews, an experiment), talk to the teaching team before you submit the proposal. That kind of work needs instructor approval and may need review by Stanford's Institutional Review Board (IRB), which oversees research involving people.

Proposal requirements from the POLISCI 158 syllabus, Project Components and Deadlines.

Workflow 03 — Choosing your design

Download this page for your AI (.md)

When: weeks 3–5. Feeds: the proposal (Oct 11) and the implementation milestone (Nov 15).

Goal

Decide what kind of project this is, and make the design choices that let it actually answer your question.

The three kinds of project

The syllabus allows projects that are theoretical, empirical, or argumentative. Pick the one that best fits your question. You can combine them, but one should lead.

Example projects

You propose your own project. The examples below show the range of what's possible. They're not a menu: you don't have to pick one, and many good projects look nothing like them. If one is close to what you want, make it your own by narrowing it to a specific place, period, case, or user.

These are starting points, not questions ready to go. Each one still needs Workflow 01 to become a specific, answerable question.

You produce: a short design paragraph

A paragraph that answers three questions:

1. What kind of project is this, and why does it fit your question?
2. What are your units and cases: what one observation, case, document, or user is, which ones you cover, and why those?
3. What would prove you wrong: the result, case, or failure that would undermine your claim?

Plus the strongest rival explanation or objection you'll address (step 3).

Steps

You · With AI · You decide

1. You Read the three rows in the table above. Pick the kind of project that leads, and write one sentence on why.

2. You Answer: what is one row in my data? (Or one case, one document, one test input.) Then: why these cases, and not others? If the honest answer is convenience, say so.

3. With AI Find the rivals, and check the design:

   Prompt

   You are a research-design coach for an undergraduate AI-and-politics course.
   My project is [theoretical / empirical / argumentative]. Here is my question, my units and cases.
   1. Check that my unit of analysis matches my question.
   2. Ask whether my cases are systematically unusual, and what that would do to my conclusions.
   3. Suggest the 3 strongest rival explanations or objections to my likely claim,
      each stated the way its supporters would put it.
   4. Suggest what my comparison or standard could be: what is compared to what (empirical),
      what my mechanism predicts that a rival doesn't (theoretical), or which evaluative criteria (argumentative).
   5. Ask me what result would prove me wrong. If nothing could, tell me.

4. You decide Choose the rival or objection you'll take on, and your comparison or standard. Write what would prove you wrong before you look at any results.

5. You Write the design paragraph in your own words. It becomes part of your proposal and your implementation milestone.

Check before moving on

• You can name your kind of project, and what "good" means for it.
• Your units, cases, and why you chose them are written down.
• There's a comparison or criterion that gives your project leverage.
• You've picked the strongest rival or objection, stated fairly.
• You've written down, in advance, what would prove you wrong.

Common traps

• Trying to do all three at once. Pick the lead kind.
• Descriptive work "to avoid making an argument." Choosing what to describe is an argument.
• Straw-man rivals. If you can't state the objection as its supporters would, you can't answer it.

Project kinds from the POLISCI 158 syllabus, Project-Based Assessments.

Workflow 04 — Sources and evidence

Download this page for your AI (.md)

When: weeks 3–8, running alongside your other work. Feeds: the implementation milestone and the final submission.

Goal

Find the literature, data, and evidence your project needs, verify all of it, and link your main claims to something that supports them.

You produce

1. A source list. For each source: the citation, one line on what it does for your project, and a check mark once you've verified it.

   Citation	What it does for my project	Verified
   Author (Year), Title	e.g. "the study I'm replicating on a new case"	✓

2. Your main claims and their evidence. For your 2–3 main claims: what supports each one, and what would count against it.

   Claim	What supports it	What would count against it

Steps

You · With AI · You decide

1. With AI Map the search. AI is good at finding leads, but it isn't a source, and it invents citations.

   Prompt

   I am doing an undergraduate project on [question]. My core concepts are [A, B, C].
   1. Suggest search terms for each concept, plus adjacent literatures or disciplines I might be missing.
   2. Suggest kinds of public data or documents that could answer my question
      (government records, legislative text, court filings, company disclosures, research datasets, archived pages).
   3. If you mention specific papers or datasets, say clearly that I must verify each one exists.

2. You Search. Start with the course readings for the relevant week, which are already vetted. Then use Google Scholar: find 2–3 highly relevant papers, and use "Cited by" to find newer work that builds on them. This is the most productive search move there is.

3. You Verify every source before it goes on your list: it exists, it's cited correctly, and it actually says what you use it for. This includes every source an AI suggested.

4. You decide Write your 2–3 main claims, and for each one, what supports it and what would count against it. Do this before you see your results.

5. With AI Check the list and the claims:

   Prompt

   Here is my source list and my main claims with their evidence.
   1. For each source, ask me what it does for MY project; flag any I only summarize.
   2. Point out claims with no evidence, evidence that supports no claim,
      and claims where I haven't said what would count against them.
   3. Suggest what kind of source is missing (theory, a similar study, a method, data).
   Do not invent references.

Check before moving on

• Every source on your list has been verified: it exists, it's cited correctly, and it says what you use it for.
• Your list covers more than background: at least one argument you build on, one similar study or method, and your data.
• Each main claim has evidence, and you've written down what would count against it.

Common traps

• Citing a source the AI made up. This is the fastest way to lose credibility, and it's covered by the course AI policy.
• Citing AI as evidence. An AI answer isn't evidence about politics or policy, unless AI outputs are what your project studies.
• Book-report reading. Knowing what a source says isn't enough. You need to know what it does for your argument.
• Confirmation-only evidence. If nothing could count against your claim, you're illustrating it, not testing it.
• Uploading sensitive material. We suggest you don't paste private, identifying, confidential, or classmates' unpublished material into AI tools (a suggested practice; see AI use and disclosure).

Binding AI rules are the POLISCI 158 syllabus's "Generative AI" policy.

Workflow 05 — Doing the work, and documenting it (implementation milestone)

Download this page for your AI (.md)

When: weeks 5–8, before Sun, Nov 15. Feeds: the implementation milestone, 800–1000 words (shorter for 4 or 3 units): design and evidence progress — what has actually been done.

Goal

Make real progress, check that it holds up, and report honestly what is done, what isn't, and what you learned.

You produce

An 800–1000-word implementation milestone with:

1. What you set out to do: your question and design, in 2–3 sentences.
2. What you've actually done: the analysis run, cases coded, or argument drafted. Be concrete.
3. What you found so far: preliminary results or a draft argument.
4. Checks: how you tested whether it holds up (see step 3 below).
5. What changed and why: the design choices you revised since the proposal.
6. What's left: the plan to the final submission, including your minimum viable version.

Attachments as relevant: code, a sample of your data, a link to a notebook. Any reasonable format works (a PDF appendix, a link to a shared folder or code repository, a notebook). Make sure the teaching team can open it without having to request access.

Steps

You · With AI · You decide

1. With AI Do the work. Get the smallest thing working first (one case coded, one chart, one working prompt, one section argued), then expand. AI can write code, clean data, run analyses, and draft short results memos for you to read. You check what it produces, and you make the decisions. Save your AI chats: you'll need them for your AI-use note.

2. With AI (optional) At the end of a work session, ask for a summary you can keep:

   Prompt

   Summarize this work session in 5 bullet points: what we did, what broke, what I decided and why,
   and what's next. Keep my decisions separate from your suggestions.

   Read it, correct anything that's wrong, and save it with your project files. It's much easier than keeping a log by hand.

3. You decide Choose two checks. Pick the two things a skeptic would attack first, and decide before each check what would count as passing. For example:

   • Empirical: does the result survive a different measure, dropping unusual cases, or a second coder on a sample?
   • Theoretical / argumentative: does the argument survive the strongest objection? Does it hold if you remove your weakest piece of evidence?

4. With AI Run the checks. AI can help you set them up and run them. Whether a check passed is your call, against the standard you set in advance.

5. You Take stock. Go stage by stage (question, design, evidence, write-up) and mark each done, in progress, or not started. Then write 3–5 lines on your key decisions and on what changed since the proposal. Your saved session summaries make this quick.

6. You Write the milestone in your own words.

7. With AI Review your draft:

   Prompt

   You are a skeptical reviewer for an undergraduate project in a course on AI and politics.
   Here is my implementation milestone draft and my session summaries. Do not rewrite the draft.
   1. Compare what I say I have done with what the summaries show. Flag anything I present as done that is not.
   2. Name the 2 most serious threats to my preliminary results,
      and ask whether my checks actually address them.
   3. Point out where my claims are stronger than my evidence, and where I undersell a real result.
   4. Ask whether my plan to the final submission is realistic, and what my minimum viable version is.

Check before moving on

• The milestone reports what you've done, not what you plan to do.
• Two targeted checks, each with a pass standard you set in advance.
• Design changes since the proposal are explained.
• 800–1000 words (or the shorter length for your units), with an AI-use note attached.

Common traps

• Presenting plans as progress. The syllabus asks what has actually been done.
• Checks for show. Two checks aimed at real threats are worth more than many that aren't.
• Deciding what "passing" means after you see the result.
• Trusting AI output you didn't check. Code that runs isn't the same as code that's right. Look at the output on a few cases you know.
• Losing your chats. Without them, your AI-use note will be vague, and rebuilding what you did is painful.
• Hiding a null or a failure. A hypothesis that isn't supported is still a finding. Report it clearly.

Milestone requirements from the POLISCI 158 syllabus, Project Components and Deadlines.

Workflow 06 — The final presentation

Download this page for your AI (.md)

When: before Mon, Nov 30 / Wed, Dec 2 (final presentation panels). Feeds: the final presentation (15%) and your final revision.

Goal

Communicate one clear idea to a mixed audience under time pressure, then use their questions to improve the project.

How the panels work

• The teaching team assigns panels by theme after the implementation milestone (Nov 15), using what your milestone says your project is about. You'll see your panel and its other projects before the sessions.
• Each panel ends with a short joint discussion, led by a moderator. The moderator asks how the projects connect: shared definitions, comparable outcomes, and where they disagree. Be ready to say how your project relates to the others on your panel.

You produce

• One-sentence takeaway: what the audience should remember. It must be a claim, not a topic.
• A hook: your first 30 seconds. A surprising fact, a concrete case, or a tension your project resolves.
• A 5–7-slide plan and a timed rehearsal.
• Q&A prep: the 3 most likely questions, one hard question, and your short answers.
• One link to your panel: a sentence on how your project connects to, or disagrees with, another project on your panel.

A 5-minute structure (suggested)

That's about 5 minutes with no slack, so cut to 5 slides if you're running long. Plan for about 80% of your time. Everything takes longer live. If you present with a partner, decide who presents which slides, and rehearse the hand-over.

Steps

You · With AI · You decide

1. You Audit what you actually have. Present what's done, and be clear about what's still planned.

2. You Write a first version of your takeaway sentence. If you can't say it in one sentence, the talk isn't focused yet.

3. With AI Test the takeaway and the outline:

   Prompt

   You are a demanding audience member at a student research panel on AI and politics.
   Here is my one-sentence takeaway, what I've found, and my slide outline.
   1. Tell me whether my takeaway is a claim or a topic.
   2. Suggest 2 alternative takeaways that my results could support, and say which is strongest.
   3. Tell me where the outline spends too long on setup and too little on results.
   4. Suggest one visual per slide that could replace text (a chart, a screenshot, a map, a document).

4. You decide Settle the takeaway and the slide order.

5. You Build the slides. One idea per slide, few words, large type, and a real visual where you can. AI can help you make charts and figures (for example, by writing the code); the talk itself is yours.

6. You Rehearse with a timer, out loud, twice. Don't read your slides.

7. With AI Practice the Q&A:

   Prompt

   Here is my takeaway and my speaker notes.
   Ask me the three questions a panel is most likely to ask, one at a time, and react to my answers.
   Then ask me the hardest fair question you can think of.

   Most panels ask some version of: What about [alternative explanation]? How do you know it isn't [reverse causation / selection / a measurement artifact]? Does this generalize beyond your cases?

8. You Read your panel. Look at the other projects on your panel. Where does yours connect or disagree? Do they define key terms the way you do?

Check before moving on

• Your takeaway is one sentence, and it's a claim.
• About 5–7 slides, 5 minutes, timed in rehearsal (with a partner: split equally).
• You have short answers ready for three likely questions and one hard one.
• You can say how your project relates to one other project on your panel.
• You don't present plans as results.
• After the panel: you've written down the feedback you received, and what you'll change.

Common traps

• Too much setup, not enough results.
• Walls of text on slides.
• A hook that doesn't lead anywhere. It has to set up your takeaway.
• Defensive answers. When a question is hard: acknowledge it, answer what you can, say what you can't, and move on.
• An unequal split. With a partner, both of you present, for about the same time.

Format, dates, and panel structure from the POLISCI 158 syllabus (Sessions 18, 19; Project Components and Deadlines), as updated by the instructor on 2026-09-27.

Workflow 07 — The final submission

Download this page for your AI (.md)

When: weeks 9–10 and finals week (exact date on Canvas). Feeds: the final submission (30%): all components integrated, plus attachments, 2000–3000 words (shorter for 4 or 3 units; see the overview).

Goal

Turn your proposal, your implementation milestone, and the feedback you've received into one integrated argument, with claims that match your evidence.

You produce

• A 2000–3000-word write-up (or the shorter length for your units).
• Attachments as relevant: code, data, a slide deck, an appendix. Any reasonable format works, as long as the teaching team can open it without having to request access.
• An AI-use note (AI use and disclosure).

A default structure (adapt it to your kind of project)

Argumentative projects: organize around your criteria and the strongest objection.

Steps

You · With AI · You decide

The write-up you submit must be in your own words. AI can help with preliminary and intermediate output, such as a plan, a rough draft or a review, but you are responsible for every word you submit. Anything it restyles or rewrites that you keep must be disclosed.

1. With AI Explore ways to organize your material:

   Prompt

   Here is my implementation milestone, my main findings, and the feedback I received.
   Suggest 2 different ways to organize a 2000-3000-word final write-up around my central claim,
   as a list of section headings with one line each. Do not write any of the text.

2. You decide Settle the outline. Under each heading, write 2–4 bullets on what goes there, and give every piece of evidence a section. Nothing should be left without a home.

3. You Draft section by section, in your own words. Start with the section you know best, and write the introduction last. Argue, don't summarize: every paragraph should advance your claim. Report what the evidence shows before you say what it means. Go through your peer reviews and panel Q&A as you draft, and address the points that matter.

4. With AI Review the full draft in three passes:

   Prompt

   You are a reviewer for an undergraduate final project in a course on AI and politics.
   Here is my draft. Do not rewrite it. Review it in three passes and report findings for each:
   1. Argument and evidence: is the central claim stated early and served by every section?
      List my 5 most important claims, and for each, compare the strength of the claim with the strength
      of the evidence. Flag overclaiming AND underclaiming.
   2. Structure and clarity: does the reader get what they need when they need it?
      Are terms used consistently?
   3. Claims and limits: what limitations or rival explanations have I left out?
      Does my introduction promise more than my conclusion delivers?
   Finish with a prioritized list of the 5 most important revisions.

5. You decide Choose which revisions to make, and revise. Calibrate your claims: use "causes" only if your design supports causal claims, use "is associated with" for correlation, and describe descriptive work as description. Don't bury null results or failures.

6. You Final checks: consistent citations, every source verified, word count, attachments, AI-use note.

Check before submitting

• Your answer appears on page one.
• Every claim is supported, and causal language only where the design supports it.
• The strongest objection or rival is addressed.
• Feedback from peer reviews and the panel is addressed.
• Every citation is verified; no AI-invented sources.
• The word count fits your units.
• Attachments and the AI-use note are included.
• With a partner: both of you contributed to every major step, as set out in your work-sharing contract.

Common traps

• Stapling milestones together. Integrate them into one argument.
• An introduction written in October. Rewrite it to match what you actually found.
• Hedging everything, or nothing. Make each claim exactly as strong as its evidence.
• AI-polished prose you didn't write. This breaks the course policy. It also tends to erase your own voice and argument.

Requirements from the POLISCI 158 syllabus (Project Components and Deadlines; Generative AI policy).

Proposal template

Download this page for your AI (.md)

Due: Sun, Oct 11, 11:59 p.m. Length: 300–500 words for 5 units (240–400 for 4 units, 210–350 for 3 units), not counting the header, the work-sharing contract, or the AI-use note. Rubric: Rubrics. How to get here: Workflows 01–03.

Copy the template below and fill it in, in your own words.

POLISCI 158 — FINAL PROJECT PROPOSAL

Name(s):
Units (5 / 4 / 3):
Working alone or with a partner:
Kind of project (theoretical / empirical / argumentative):
Peer reviews I will do (4 and 3 units only): [e.g., 1 and 3]

1. QUESTION
   [1–2 sentences, ending in a question mark.]

2. MOTIVATION
   [Why this question matters, and to whom. At least one concrete reason:
    a number, a named study or gap, or a named audience that would act on the answer.]

3. FEASIBILITY
   What I'll use:   [data, sources, tools, or cases]
   How I'll get it: [and whether you already have it]
   What I'll do:    [the analysis or argument, in a few sentences]
   Biggest risk:    [and your plan B]
   Minimum viable project: [the one thing you'll deliver even if everything else fails]

AI-USE NOTE
   [See docs/polisci158-ai-use-and-disclosure.md]

Work-sharing contract (partners only)

If you're working with a partner, attach this to the proposal. Both of you sign it.

The rules it has to meet:

• You're both taking the course for the same number of units.
• Both of you contribute to every major step below. You can divide tasks within a step, but not whole steps between you.
• You're both graded on the full project, not just your parts.
• You split the final presentation equally.
• Word counts are the same as for individual projects, and you each write your own peer reviews.
• If the partnership breaks down, you're both still responsible for finishing the project, on your own if you have to.

WORK-SHARING CONTRACT

Partners:                                   Units (both):

For each step, say who does which tasks, and how you'll check each other's work.

1. Project formulation (question, motivation, scope)
   [Partner A:]                [Partner B:]

2. Argument / hypothesis design
   [Partner A:]                [Partner B:]

3. Literature review
   [Partner A:]                [Partner B:]

4. Empirical work / analysis
   [Partner A:]                [Partner B:]

5. Write-up
   [Partner A:]                [Partner B:]

6. Final presentation (split equally)
   [Partner A:]                [Partner B:]

How we'll work together: [how often you'll meet; where you'll keep shared files and AI chats]
If something goes wrong: [what you'll do if one of you falls behind, and when you'll tell the teaching team]

We understand that we'll both be graded on the full project, and that if our partnership
breaks down, each of us is still responsible for finishing the project, on our own if necessary.
Signed: ____________________   ____________________   Date: ________

Rubrics

Download this page for your AI (.md)

Each graded part of the project has a rubric below. Each criterion is scored on three levels: Strong (2 points), Developing (1), or Weak (0). All criteria in a component count equally, so your grade for a component is the points you earn divided by the maximum. For example, a proposal scored 2, 2, 1, 2 earns 7 of 8 points.

Applies to every component

• Your own words. Written work must be in your own words and reflect your own understanding (syllabus AI policy). Undisclosed substantive AI use, or submitting AI output as your own, is an Honor Code matter, not a rubric deduction.
• AI-use note attached. A missing AI-use note costs 10% of that component's grade.
• Length. Stay within the range for your units. Going outside it costs 5% of that component's grade.
• Partners. Partners get the same grade for the written project components. The final presentation is graded partly individually (see below), and peer reviews are individual. For the proposal, the work-sharing contract is part of the grade (see below).

Proposal milestone (10% of the course grade)

Syllabus: motivate the project; explain feasibility.

Partners: work-sharing contract. It isn't scored, but it has to be accepted. It's accepted when every major step has a role for both partners and there's a realistic plan for working together. If steps are missing, or whole steps go to one partner, it's returned for revision.

Implementation milestone (10%)

Design and evidence progress; what has actually been done.

Final presentation (15%)

Syllabus: themed panel presentation and Q&A. 5 minutes, about 5–7 slides. Partners split the time equally.

With a partner, both of you present for about the same time, and the presentation is graded partly individually: delivery and slides and Q&A are scored for each of you separately; takeaway, evidence and honesty, and panel connection are scored jointly.

Final submission (30%)

Syllabus: all components integrated, plus attachments.

Peer reviews (10%)

Syllabus: 150–200 words each; giving usable feedback is a course learning goal.

All three peer reviews, including Peer Review 3 in session, use this rubric.

AI use and disclosure in the final project

Download this page for your AI (.md)

The rule (syllabus, verbatim)

Because this course both studies and uses AI, we hold the line explicitly. You are encouraged to use AI tools extensively for the applied work: exploring, prototyping, debugging, building, and analyzing. All written assignments, however, must be in your own words and must reflect your own understanding. You must disclose substantive AI use in a brief AI-use note attached to the submission, crediting tools the way you would credit a collaborator. Using AI writing tools and submitting the output as your own violates the Stanford Honor Code.

Everything below is practical guidance on following that rule. If anything here seems to conflict with the syllabus, the syllabus wins.

What that means for each part of the project

Suggested practices

Three practices help you follow the rule above. We recommend them, but they're suggestions, not requirements. The binding rule is the syllabus policy quoted at the top.

• Human first: make your own first attempt before using AI (see below).
• Peer reviews without AI: write your reviews of classmates' work yourself, without AI help.
• No sensitive uploads: keep private, identifying, or confidential material, and classmates' unpublished work, out of AI tools (see Privacy below).

Human first, then AI

1. Pass 1, you. Make a real first attempt in your own words: a question, a claim, an outline, a design, a draft.
2. Pass 2, AI with control. Then use AI to challenge, clarify, or extend it.
3. Keep the thread. You should always be able to say what was yours before AI, and what changed after.

This makes your disclosure easy, and it protects the learning the project is for.

Verification (non-negotiable in practice)

• AI invents citations. Before any source goes into your work, check that it exists, that it's cited accurately, and that it says what you use it for.
• AI isn't evidence. Don't cite an AI system for facts about politics, law, policy, or public opinion, unless AI outputs are what your project studies.
• You own the errors. Anything under your name is your responsibility, including hallucinations you didn't catch.

What counts as "substantive" (disclose it)

Disclose when AI meaningfully shaped:

• content, structure, or argument;
• analysis, code, or data;
• your evidence search;
• wording, including restyling your prose.

No disclosure needed for spell-check, grammar fixes, formatting, or a synonym look-up that leaves your phrasing intact.

When in doubt, disclose.

AI-use note template (suggested format)

Attach this to every project submission. Any reasonable format works (a page at the end of your document, or a separate file); the template below is a suggestion.

AI-USE NOTE — [Proposal / Implementation milestone / Final submission]

Tools used: [e.g., model name and version, Stanford AI Playground, coding assistant]

What I used AI for:
- [Task] — [what the AI did] — [what I kept, changed, or rejected]
- [Task] — ...

What is entirely mine: [e.g., the research question, the argument, all prose in sections 1-5]

Verification: [e.g., all 12 sources checked on Google Scholar; code outputs checked against 30 hand-coded cases]

Chats/logs available on request: [yes / link]

Example entry: "Debugging my Python scraper: Claude found the pagination bug and I used its fix (lines 40–52). I rejected its suggestion to drop incomplete records and kept them with a flag instead."

Privacy

Suggested practice. Don't paste private, confidential, identifying, or copyrighted material into AI tools. That includes raw interview notes, non-anonymized data, and classmates' unpublished work.

Access

Students have access to a frontier AI model through Stanford. The Stanford AI Playground is also available. If cost or access is ever a barrier for a particular task, contact the teaching staff.

Sources: POLISCI 158 syllabus ("Generative AI", "Academic Integrity", "Course Requirements"), quoted verbatim where marked.

Peer review guide

Download this page for your AI (.md)

Giving usable feedback is a course learning goal, and it's graded (peer review is 10% of the grade).

The three reviews

If you take the course for 4 units you do any 2 of the 3 reviews; for 3 units, any 1. You choose which, and declare them in your proposal (Sun, Oct 11). Once declared, the reviews you chose are required. Peer-review deadlines are firm, and late reviews are not accepted.

The TAs assign reviewers. For Peer Review 3, you review a peer on your own panel. If you have a partner, you each write your own reviews. All three reviews use the same rubric, in Rubrics.

What usable feedback looks like

150–200 words is short. Spend it on what matters most:

1. Restate the project in one sentence. "As I understand it, you're asking whether … and plan to …" If you can't, that's feedback in itself: the project isn't clear yet.
2. One real strength. Be specific about what works and why.
3. The most important concern. Pick the one issue that most threatens the project, not five small ones. Useful places to look:
   • Proposal: is the question specific? Is the motivation concrete? Is the plan feasible?
   • Milestone: does it report real progress? Do the checks address the real threats? Are the claims matched to the evidence?
   • Presentation: was the takeaway clear? Was it earned by the evidence? What question went unanswered?
4. One concrete suggestion. Something the author can actually do before the next deadline.

Template

In one sentence, this project ...
The strongest part is ... because ...
The most important issue is ... because ...
One thing to try before the next deadline: ...

Ground rules

• Critique the work, not the person. Be direct and kind.
• Name one priority, not a list. Otherwise the author won't know what to fix first.
• "I'd do it differently" isn't a flaw. Point to problems of logic, evidence, or feasibility, not taste.
• Write it yourself. The point is your genuine engagement with a classmate's work. Like all written work in the course, your review must be in your own words. We also suggest writing it without AI help, and not pasting a classmate's unpublished work into an AI tool (suggested practices; see AI use and disclosure).

Sources: POLISCI 158 syllabus (Student Expectations: Peer Review; Grading; Project Components and Deadlines). Ground rules draw on the syllabus's "Creating a respectful environment".