You have the data. You have three years of reading. You have folders of notes, half-written sections, a supervisor waiting on chapter four and a funded period that ends before the thesis does. What you do not have is a chapter — and every week that passes makes starting harder, because the material keeps growing while the blank document does not.
This is the most common failure point in the UK doctorate, and it is almost never a knowledge problem. It is a problem of converting accumulated material into linear argument. If you want to start that conversion now, you can do it in Tesify.
What writing up actually costs you
Be clear about the stakes, because they are not only academic. An overrun typically means an unfunded writing-up year, during which the stipend has stopped and the work has not. Advance HE’s Postgraduate Research Experience Survey 2025 — 35,475 responses across 93 institutions — found that among doctoral researchers who had considered leaving, 20% cited mental or emotional health and 14% cited financial difficulties. The writing-up period is where those two pressures meet.
And the delay is self-reinforcing. Every month you do not draft, the volume of material to synthesise grows, the memory of why you made a decision fades, and the perceived size of the task increases. The single most valuable thing you can do is convert something — anything — into prose this week.
Why the notes will not become a chapter on their own
Because notes and chapters are different objects. Notes are associative, indexed by when you encountered something. A chapter is linear and argumentative, indexed by what the reader needs next. There is no amount of reorganising your notes that turns one into the other; the conversion is an act of authorship.
The specific trap is waiting to feel ready. Doctoral researchers routinely believe that one more paper, one more analysis, one more pass through the data will make the chapter writable. It will not, because the missing element is not information — it is a decision about what you are claiming.
The staged workflow
Stage 1: Write the claim before the chapter
One sentence: what does this chapter establish? Not what it is about — what it establishes. “This chapter examines participants’ experiences of supervision” is a topic. “This chapter shows that researchers interpret supervisory availability through disciplinary norms about independence, which explains why identical contact hours produce opposite satisfaction judgements” is a claim.
Everything downstream is easier once this sentence exists, because it tells you what to include and, more importantly, what to leave out. If you cannot write it yet, that is the work — not the drafting.
Stage 2: Build a skeleton of claims, not topics
List the five to eight things the chapter must establish, in the order a reader needs them. Each becomes a section, and each section heading should be a claim you could defend, even if the final version is a conventional noun phrase.
This is where AI assistance is genuinely useful and entirely defensible: give it your own claim sentence and your own list of sections and ask whether the sequence has gaps, whether any step assumes something not yet established, and where a reader would object. That is structural critique of your material, and it is the fastest route out of a stall.
Stage 3: Attach your evidence to each claim
Under each section, list the specific evidence that supports it: which participants, which results, which sources. This is where three years of notes finally earn their keep, and it is a sorting task rather than a writing task — which is why it is achievable on a bad day.
If a section has no evidence beneath it, you have found something important. Either the claim is unsupported and must go, or the analysis you need has not been done yet. Better to discover that now than in your viva.
Stage 4: Draft badly, deliberately
Write the worst acceptable version of each section, quickly, without stopping to check citations or polish sentences. The purpose is to get the argument onto the page where you can see whether it holds.
Most doctoral researchers cannot do this, because years of training have made them unable to write a sentence they know is imperfect. This is precisely why writing up stalls: the standard that produces good final chapters produces no first drafts at all. Separate the two activities and both get easier.
Stage 5: Revise against the claim
Now bring your standards back. For each paragraph, ask whether it advances the chapter’s claim. If it does not, it goes — however good it is and however long it took. Material cut here can live in an appendix or a future paper.
Stage 6: Verify everything
Every citation opened and checked. Every number traced back to your analysis. Every quotation matched against the source. Do this as a separate pass, because verification and composition use different attention, and doing them together is how errors survive.
The line you cannot cross
This audience will not forgive a fudge on this, so here it is plainly.
What AI can defensibly do: critique the structure of an argument you have built; suggest an order for sections you have specified; tighten sentences you have written; identify where a draft assumes something it has not established; help you get an imperfect first version onto the page so that revision has something to work on.
What it must never do: decide what your findings mean; generate citations or reference lists; summarise papers you have not read; write the contribution claim; or produce text you could not explain and defend.
The test is your viva. UK examiners are checking, among other things, that the thesis is yours and that you engaged with the research process — the QAA notes this is part of why the outcome is not revealed in advance of the oral examination. A chapter you cannot discuss is a chapter that will be found, and being unable to explain your own argument is a far worse position than a rough draft.
Two practical rules follow. Never accept a reference from any tool you have not personally opened and read — fabricated citations are discovered in seconds and are catastrophic at doctoral level. And check your institution’s policy on generative AI and any declaration requirement, since these differ between universities and sometimes between departments.
Where Tesify fits
Tesify is built for stages 2 to 4 specifically — the conversion of material you already own into structured drafted chapters. You bring your claim, your evidence and your reading; it helps you get from a skeleton to prose without the week-long stall in front of a blank document.
It does not read your sources for you, decide your contribution or invent your references. The scholarship, the interpretation and the argument remain 100% written by you, which is the only version that survives examination. More than 9,000 students have used it to get stalled writing moving, across more than 15,000 chapters.
Start drafting your chapter in Tesify.
What to write first
Not the introduction. It has to promise what the thesis delivers, and you do not yet know precisely what that is.
Start with a results or findings chapter, because the material is most concrete and the claims are closest to your data. Then the discussion, which is where the thesis makes its case — our guide to writing the discussion chapter sets out the seven moves it needs. Drafting the discussion usually clarifies your contribution, which is the claim examined in our guide to what counts as an original contribution to knowledge. Write the introduction last.
Two practical points while you are at it. Set your timeline against your actual funded period rather than an assumed one, since as our data on UK PhD stipends and funding shows, UKRI publishes no sector-wide duration. And if your document is going to reach 80,000 words with several hundred references, make sure your writing environment can carry it — our comparison of LaTeX and Word at thesis scale covers where each one fails.
Frequently asked questions
Is using AI to help write my thesis research misconduct?
It depends on your institution’s policy and on what you use it for. Structural critique and sentence-level editing of your own writing are treated very differently from generating content or sources. Read your regulations, follow any declaration requirement, and never submit text you cannot defend.
Will my examiners be able to tell?
The reliable signal is not stylistic — it is whether you can discuss your own reasoning in the viva. Examiners probe why you made particular analytical decisions and what alternatives you rejected. A workflow where you make every interpretive decision yourself leaves you able to answer; one where you did not, does not.
How should I declare AI use in my thesis?
Specifically, where your institution requires it. State what the tool did and what it did not — for example, that it was used for structural feedback and clarity editing on text you wrote, and not to identify sources, summarise literature or generate content. A precise declaration reads as command of the boundary.
What about the confidentiality of my unpublished research?
A fair question, and one you should ask of any tool before uploading a chapter. Unpublished doctoral research may be commercially sensitive, may involve participant data with ethical restrictions, and may affect later publication. Read the provider’s current documentation on storage, access and training use, and check whether your data protection obligations to participants permit uploading their data at all.
Is it worth paying for a writing tool on a stipend?
That is your judgement to make against your own budget, and there is a free tier to test the fit before committing. The comparison worth making is not tool cost against zero, but against the cost of an unfunded overrun term.
I have been stuck for months. Where do I actually start?
With stage 1, on the chapter whose material is most complete. Write the single sentence stating what that chapter establishes. If you cannot, the blockage is analytical rather than a writing problem, and that is the conversation to have with your supervisor this week rather than next term.
Should I tell my supervisor I am using an AI tool?
Yes. Supervisors would far rather discuss your workflow openly than discover it at examination, and many will have a view on what your department expects. Raising it yourself also puts you on the right side of any declaration requirement.