Long projects and dissertations

A different kind of work

A dissertation or final-year project fails in ways a three-week essay does not. Not usually through inability, but through drift: months in which the question changes shape, the reading expands without converging, and nothing is written until the last six weeks.

What follows applies to any long piece: a research project, a capstone, a portfolio, a design project.

The question, and its instability

Your question will change. That is normal and it should be deliberate rather than accidental. Write it down, dated, in one sentence, and rewrite it whenever it moves, keeping the old versions.

That file of successive questions is worth a great deal at the end: it is the intellectual history of the project, it makes writing the introduction straightforward, and in a viva it demonstrates exactly the judgement you are being examined on.

A good use of a model early: here is my question. What would I need to establish to answer it? What is ambiguous in how I have phrased it? What would somebody object to about the scope? Diagnosis of your question, not a supply of one.

Write from week one

The single most useful habit. Not the dissertation: a working document, updated weekly, containing what you have read, what you now think, and what you do not know.

This does three things. It converts reading into thinking as you go, rather than in a panic at the end. It means you always have text, so the blank page never arrives. And it produces a continuous, timestamped record of the work developing, which is the process record from Module 7 in its strongest form.

Students who write only at the end lose most of what they read in months one and two, because they never encoded it.

The literature, without drowning

A method that converges:

1. Start with two or three recent reviews. They give you the map and a curated reference list.
2. Follow citations backwards to the foundational work everybody cites, and forwards through Scholar’s cited by to the recent work.
3. Stop when you stop meeting new names. Convergence is the signal, and it arrives sooner than you fear.
4. Four lines per paper, in your own words, in Zotero, from Module 5.

A model is useful for step zero (the vocabulary a field uses, so your searches match) and for explaining methods sections you cannot follow. It is not useful for producing the reference list, and a bibliography you did not build is a bibliography you cannot defend in a viva.

Supervision

Your supervisor is the most valuable resource in the project and the most under-used. Practical points:

• Send something before every meeting, even if it is bad. A meeting about a document is worth five about intentions.
• Ask specific questions. Is my second research question answerable with this data gets a useful answer; how am I doing does not.
• Write up every meeting (what was agreed, what you will do) and email it to them. This prevents the common disaster of two people remembering different agreements, and it is a record.
• Say when you are stuck, early. Supervisors have seen every version of stuck and mostly want to help. The failure mode is disappearing for six weeks out of embarrassment.

Ethics and data

If your project involves people, ethics approval comes before data collection, not after, and there is no retrospective route. Your approval says where data will be stored and who may access it: a commercial AI service is almost certainly not on that list, and uploading participant data is a research-integrity breach independent of any AI policy. Anonymise before anything leaves your approved storage, remembering from Module 2 that removing names is not anonymisation when the combination of details identifies somebody.

Where AI helps across the project

Honestly and specifically: understanding difficult papers, explaining statistical methods, generating viva questions, critiquing your structure, checking your draft against the criteria, debugging analysis code, and helping you say clearly something you already decided to say.

And where it does not: choosing the question, deciding what the data means, judging which literature matters, or writing anything you would be asked to defend. Those are the parts being assessed and the parts you will be examined on.

The schedule that works

Count backwards from submission, and put the last month aside for writing and revision, not for new work. Then:

• Reading and question refinement: first third.
• Data collection or core work: second third, and it will take longer than planned.
• Analysis and writing: final third, with the last month for revision only.

Everybody overruns the middle. The way to protect the end is to start writing in the first third, so the final month is revision rather than composition.