Skip to content

The thesis in the age of AI

If you want to see every problem in this series in one place, look at the master’s thesis at a business school: the months of individual research that close a Master of Science in management, finance or marketing and, at many schools, the MBA as well. One deliverable. Months of work nobody watches. One coach with a fixed number of meetings. A second reader who appears at the end. And now a logbook in which the student records their own use of AI, after the fact. It is the purest form of assign and hope, and it is also where the remedy is easiest to picture.

What the rules say now

Rotterdam School of Management publishes its thesis manual, so take it as the worked example. The thesis is 25 to 70 pages plus a 1,000-word executive summary. A coach and a second assessor form the committee. Supervision is usually about six meetings, the committee assesses the proposal and the thesis within two weeks, and the student will usually not meet the second assessor before the oral. The grade is 70% the written work, judged on a seven-dimension matrix, and 30% the oral.

On AI the manual is clear and, by the standards of 2026, well drafted. AI output may not be submitted as the student’s own without citation. Sources may not be fabricated. Confidential data may not be uploaded to tools the school has not approved. Any permitted use must be documented in a logbook or appendix, and the student must be able to show their original work and justify or reproduce AI outputs if an examiner asks. AMBA now expects every accredited school to have a policy of this kind, with disclosure and citation of AI use, in its 2025 guidance.

Many theses in European programmes are not research theses at all but projects inside the student’s employer. Nyenrode’s modular EMBA ends in a thesis on a practical management issue from the student’s own organisation, and Kellogg-WHU’s in an in-company thesis. The same rules apply, with the added fact that the data belongs to someone else.

What breaks

Four things, and they are visible in the rules themselves.

The months between meetings. Six meetings over a thesis is one every few weeks. Between them the coach has no idea whether the student is reading, drafting, stuck or gone. The manual’s answer is to tell students to take the initiative and ask. Some do.

The logbook is written afterwards, by the person it is about. It is the right idea, a record of AI use, built on the wrong mechanism, self-report. The first piece in this series gives the number: 74% of students at one London business school did not declare AI use even with no penalty for declaring. A logbook that depends on the same honesty has the same gap.

Confidential data. The rule says not to upload it to unapproved tools, and 94% of UK undergraduates use generative AI for assessed work, in HEPI’s 2026 survey. An in-company thesis without an approved tool is an invitation to break the rule quietly.

The defence is where the truth comes out, late. Nova SBE has lengthened its thesis defences for exactly this reason, as GMAC’s 2026 deans survey reports. A longer oral at the end is a better check. It is still a check at the end, after the months in which the problem could have been caught, and it costs the committee’s most expensive hour.

The thesis as a method

Every thesis coordinator already has a method. The question, the literature, the design, the data, the analysis, the conclusions, in that order, with a sense of what each one must contain before the next makes sense. It lives in the manual’s grading matrix and in the coach’s head. Put it where the student works and most of what breaks stops breaking.

Each stage becomes a step with its criteria written down. The student cannot start the analysis until the design is signed. Every draft of a step is read first against the criteria, by something that does not need a meeting to do it, and goes back with what is missing. The coach reads what is ready and signs or returns it with one note. Six meetings become six decisions, taken when the work is ready rather than when the calendar says.

The logbook stops being a separate document. If the student’s reading, questions, drafts and the AI tutor’s pushback are recorded as they happen, the record of AI use is a by-product, not a confession. The confidential data stays inside a tool the school approved, because that is where the work is done. And the defence changes character: the committee arrives having read not a stack of pages but a record of how the pages came to be, and asks about that.

What a thesis method must have

  • Stages as gated steps, each with what it must contain, in the coordinator’s own words, applied to every student the same way.
  • A first reading that is not the coach’s, so a draft is never the coach’s first contact with the work.
  • The coach’s signature at each stage, recorded, with the return notes.
  • AI use recorded as it happens, not reported afterwards.
  • The employer’s data inside the school’s approved tool, with a processing agreement behind it.
  • A record the second assessor and the committee can read before the defence.

Where Ulern stands

We build Ulern, the applied-learning platform for business schools and executive education, so read this section knowing that.

A thesis in Ulern is a project: the coordinator’s stages as steps with their criteria, a plan per participant that pairs each step with the topics they need, the AI tutor reading each draft against the criteria and teaching from the school’s material, and the supervisor signing at each stage. The record of it exports with the thesis. The method is the coordinator’s, written once; we supply the steps machinery, not the method. No school has yet run its theses this way on Ulern, and a thesis is where a school’s own rules matter most, so the first one will be built with its examination board in the room.

What 56 of Europe’s leading business schools publish about AI in the thesis, school by school and quoted from their own pages, is in AI rules for the thesis.


This is the sixth of seven pieces in Assign and hope, a series on what goes wrong while a programme’s project runs, and what a remedy must have. The others: Did they write it, or did the AI? · Activities are not outcomes · Supervision does not scale · The 90 days after the programme · What corporate clients now ask of schools · What to ask an AI tool before December 2027

← All posts