Back to the blog

Thesis: building your research question and outline with AI

Thesis4 min read · 26 September 2026

The GetPack team

You have to submit a thesis (mémoire) and you don’t know where to start: your starting question feels too broad, the outline won’t take shape, and the bibliography is a project in itself. AI can help you structure this work — but there’s an essential difference between using it to think and using it to write in your place. French universities that have published a charter on the subject are clear about this.

What a university charter actually says

The Université d’Angers charter, which also applies to Polytech Angers, sets out a simple principle: generative AI “can be used as an assistance tool in preparing assignments, reports, and dissertations,” particularly as a “language assistant and help with finding information.” But it immediately adds a limit: “unless stated otherwise, work produced by learners cannot be produced by such tools.” In other words, AI helps you prepare, not produce the final text.

The charter also stresses three very concrete obligations if you use AI on your thesis:

  • Verify and cite: any information or citation obtained via AI must be verified and properly referenced, or it counts as plagiarism.
  • Get your advisor’s agreement before folding AI into your working process, and when in doubt, discuss it with them rather than guessing what’s tolerated.
  • Disclose your use in the final document (a dedicated section, footnote, or appendix), explaining precisely how you used the tool. Université d’Angers even provides a questionnaire for this.

It specifies a point directly relevant to a research thesis: “generative Artificial Intelligence must not be used to fabricate research data for a survey.” If your thesis relies on interviews or fieldwork, the data must come from your actual collection, never be generated.

The charter of the UFR Sciences Humaines et Sociales at Université Paris Cité goes further on one point: by default, absent instructions from the instructor, no use of AI is authorized. It’s up to each instructor or thesis advisor to explicitly state what’s permitted. If AI is authorized, you must then lay out your working methodology in the assignment, cite the relevant passages, and remain fully responsible for the content produced — including its errors. This charter is blunt about what not to do: letting AI generate the bulk of the content, or presenting its answers as established fact when they are, in its own words, “probabilistic outputs.”

These are two different charters, but the underlying message is the same: complementary use only, never a substitute, mandatory transparency, and systematic self-verification. The penalty for non-compliance can range, under the Angers charter, from rejection of the assignment to referral to a disciplinary committee.

Where AI genuinely helps on a thesis

Staying within this framework, there are three moments in thesis work where an AI assistant saves you real time without ever writing in your place:

1. Moving from a starting question to a research question (problématique). You arrive with an intuition that’s too broad (“the impact of remote work on employees”); a good back-and-forth lets you narrow it down through follow-up questions: what fieldwork, what time period, what specific population, what tension or contradiction your thesis is going to try to illuminate. You’re the one who formulates the answer; the AI asks the questions that move things forward.

2. Building a thesis outline and methodology. Once the research question is set, you need to break down the work: theoretical framework, chosen fieldwork or corpus, data collection method (interviews, survey, document analysis), and a schedule up to the submission date. Here again, AI helps you structure the steps and spot the gaps, but the content of each section is still yours to write.

3. Organizing the bibliography. Gathering your references as you write, checking they’re complete (author, year, publisher, DOI if available), and formatting them in the right citation style — this is mechanical work where AI can genuinely save time, as long as it never invents a reference that doesn’t exist or fills a missing DOI with a made-up value.

What to do before you start

Before you even open an AI tool for your thesis, only one question matters: what does your institution allow? The two charters cited here show that the rule isn’t the same everywhere — one starts from the principle that complementary use is allowed unless banned, the other starts from the opposite principle. Ask your thesis advisor explicitly what’s tolerated, and if an AI charter exists at your institution, read it before using one. That’s exactly what both texts opened for this article say, word for word.

Key takeaways

  • AI can help you move from a starting question to a research question, build an outline and a schedule, and organize a bibliography — never to write the thesis in your place.
  • The Angers charter allows assistance use provided you verify, cite, and disclose your use; the Paris Cité (UFR SHS) charter bans it by default, unless the instructor explicitly authorizes it.
  • Neither charter allows fabricating research or fieldwork data with AI.
  • Undisclosed or unauthorized use is treated as a breach of academic integrity, with penalties ranging from rejection of the work to referral to a disciplinary committee.
  • Always ask your thesis advisor what’s permitted, and check whether your institution has published its own charter before using one.

To support you at this exact stage — moving from your starting question to a solid research question, building your outline and schedule — the memoire-problematique skill applies exactly this framework: it asks questions and structures your thinking, it never writes the thesis for you.

Sources

  1. Charte d'utilisation de l'IA générative — Université d'Angers / Polytech Angers · accessed 26 September 2026
  2. Charte d'utilisation de l'intelligence artificielle générative par les étudiants — UFR SHS, Université Paris Cité · accessed 26 September 2026

Read the next article

Method26 September 2026

Building your exam study schedule with AI, the right way

Method26 September 2026

Revising with AI, honestly: active recall, Feynman, quizzes

Career26 September 2026

Choosing your apprenticeship with real employment data

Code27 September 2026

Learning to code in the age of AI: what you still need to know how to do yourself

Code27 September 2026

The one-page spec to write before you prompt an AI to code

Thesis27 September 2026

How to cite ChatGPT or Claude in a thesis: APA, MLA, ISO 690

Code27 September 2026

Claude Code for beginners: install, first launch, CLAUDE.md

AI news27 September 2026

Claude Opus 5.5: what really changes (and how to use it to study)

Analysis27 September 2026

The AI race: why some people are scared and others aren't

Code27 September 2026

Building your first MCP server, step by step

Code27 September 2026

Deploying your first site for free: Vercel, Netlify, Cloudflare Pages, or GitHub Pages

Explainer27 September 2026

AI detectors: are Turnitin, GPTZero and Compilatio reliable?

Code27 September 2026

Writing your own Claude skill: structure, SKILL.md, and a description that triggers

Career27 September 2026

France's national student-entrepreneur status (SNEE) and the PEPITE network, explained

Analysis27 September 2026

France in the AI race: Mistral, energy and talent

AI news27 September 2026

French Tech and AI: the French startups to know in 2026

Code27 September 2026

Git without fear: commit, branch, remote explained, then the commands that save you

Explainer27 September 2026

Chinese AI: DeepSeek, Qwen, Kimi… why Europe is wary

Health27 September 2026

AI in medical school: study for PASS, LAS and the EDN safely

Tools27 September 2026

Free AI for students: every offer and discount (2026)

Career27 September 2026

AI on an internship or apprenticeship: what’s allowed, what isn’t

Code27 September 2026

From IDE to ADE: coding with AI agents in 2026

Code27 September 2026

Reading an error without panicking: the anatomy of a stack trace (Python and JavaScript)

Tools27 September 2026

Best AI for students in 2026: the honest comparison

Tools27 September 2026

New AI models in 2026: which one should you study with?

Tools27 September 2026

French AI tools you’ve never heard of: Noota, Moshi, Vibe…

Analysis27 September 2026

Why AI is so expensive (and American AI even more so)

Code27 September 2026

How to prompt an AI coding tool well: the method that changes everything

Code27 September 2026

Securing a vibe-coded app: 7 mistakes to fix before you publish

Code27 September 2026

Slopsquatting: when AI recommends packages that don't exist

Weekly brief27 September 2026

AI news roundup: the week of September 21–27, 2026

Code27 September 2026

Vibe coding: what it actually means (and how not to mess it up)

AI news27 September 2026

Why Yann LeCun wants AMI: AI beyond LLMs

Tutorial26 September 2026

Connect an MCP connector to Claude without writing a line of code

Degrees26 September 2026

Choosing your program with real MonMaster and InserSup data

Method26 September 2026

APA 7, ISO 690, Vancouver: how to cite your sources properly

Explainer26 September 2026

ChatGPT's 'hidden codes' on TikTok: fact vs. fiction

Health26 September 2026

Medicine: revise for the EDN with France's public drug database

Career26 September 2026

Internship pay and apprentice wages in 2026: the rules

Explainer26 September 2026

AI and academic integrity: what universities actually say

Tutorial26 September 2026

Installing a skill in Claude in 2 minutes

Join GetPack

Already have an account? Sign in