You have forty papers, a folder of notes, and a literature review chapter that will not start. The temptation is obvious: paste the topic into a chatbot and ask for a review. That specific shortcut is also the fastest route to an academic misconduct meeting — not because AI is forbidden at most UK universities, but because of what happens to the citations.
There is a version of this that is defensible, produces a better chapter, and takes less time than staring at the page. It requires being precise about which parts of the job you can delegate and which parts are the job. If you want to start drafting with that structure already in place, you can do it in Tesify.
First: read your own university’s policy
UK institutions have taken meaningfully different positions on generative AI, and some departments differ from their own university’s default. Policies typically distinguish between permitted assistive use, use that must be declared, and use that constitutes misconduct — and the boundaries are not the same everywhere.
So before anything else: find your institution’s academic integrity policy and your module handbook, and find out whether your department requires a declaration. If it does, plan to write one. Nothing on this page overrides what your own regulations say, and “a website said it was fine” is not a defence anyone has ever successfully run.
The line that actually matters
Forget the question “is AI allowed”. The useful question is: can you explain and defend every claim in your chapter as your own understanding?
If your supervisor stops at any paragraph and asks “why did you group these three studies together, and what does this one actually show?”, you need an answer. That test does not care which tool you used. It cares whether the reasoning happened in your head.
This maps onto a clean division of labour.
Reasonably delegated
- Reorganising a chapter you have drafted into a clearer order
- Tightening sentences that are grammatically tangled
- Suggesting thematic groupings for studies you have read and summarised
- Explaining an unfamiliar methodological term so you can go and read properly about it
- Producing a first outline that you then argue with and rewrite
- Checking whether your topic sentences actually match the paragraphs beneath them
Never delegated
- Deciding what the literature says
- Judging whether a study is any good
- Choosing which studies matter to your argument
- Generating citations or reference lists
- Summarising a paper you have not read
- Writing the gap your project fills
The second list is the intellectual content of a literature review. Delegate any of it and you have not written a review; you have commissioned one.
Why AI-generated citations end badly
General-purpose language models produce references that look correct — plausible authors, a real-sounding journal, a well-formed volume and page range — for papers that do not exist. They also attach real authors to work they never wrote, and real papers to findings they never reported.
This is not a rare glitch, and it is uniquely catastrophic in a literature review because a marker’s first instinct with an unfamiliar reference is to look it up. A fabricated citation is discovered in about thirty seconds, and it reads as either fabrication of sources or citing work you never read. Both are misconduct.
The rule that removes the risk entirely: never accept a citation from any tool you have not personally opened and read. Not the abstract — the paper. If you cannot produce the PDF, it does not go in your chapter.
This applies to summaries too. A tool’s account of a study’s findings can be subtly wrong in ways you will not detect unless you have read the source, and a chapter that misreports what three studies found is worse than one that discusses two studies accurately.
A workflow that survives scrutiny
- Search and screen yourself. Use your library databases. Record your search terms and databases as you go — you will need them if your project turns out to require a structured review, and it takes two minutes now versus an afternoon of reconstruction later.
- Read and summarise each paper in your own words. Four or five lines each: what they did, what they found, what the limitation is, why it matters to you. This is the step people try to skip, and it is the step the entire chapter is built from.
- Group your own summaries into themes. Lay them out and look for disagreements, not topics. The strongest reviews are organised around unresolved questions.
- Now use AI on your own material. With your summaries in front of you, ask for a structural critique: is this ordering logical, does each theme lead into the next, where does the argument jump. This is where assistance genuinely helps, because it operates on content you produced.
- Write the prose yourself, then edit with help. Draft each paragraph, then use a tool to tighten what you wrote. Editing your sentences is a very different act from generating them.
- Verify everything. Every reference opened, every claim traceable to a paper you have read, every quotation checked against the original.
This produces a better chapter than the shortcut would, for a reason worth noticing: the thematic structure comes from your reading, so it reflects what the literature actually contains rather than what a model predicts a review on your topic usually looks like.
Where the time actually goes
Students imagine AI saves them the writing. In practice it saves you the staring — the hour lost to an empty page, the paragraph rewritten six times before you work out what it was supposed to say. That is real time, and reclaiming it is legitimate.
What it does not save is the reading. There is no version of a literature review where you do not read the literature. Any workflow promising otherwise is describing a chapter that will not survive a supervision meeting, let alone a viva-style discussion of your work.
Declaring your use
Where your institution requires a declaration, write it specifically. “AI was used” tells an examiner nothing and invites suspicion. Something like this tells them exactly what happened:
Generative AI was used to review the structure of drafted chapters and to suggest improvements to sentence-level clarity in text written by the author. It was not used to identify sources, to summarise literature, or to generate content. All sources cited were located through library database searches and read in full by the author.
A precise declaration is a strength. It demonstrates you understood the boundary rather than hoping nobody would ask.
How this fits the rest of the chapter
None of this changes what a literature review has to be. It still has to end by earning your research question, and it still has to be organised by argument rather than by paper. If you are not yet sure which kind of review your department expects — a narrative review inside an empirical dissertation, or a structured review as the dissertation itself — settle that first using our guide to literature reviews versus systematic reviews, because the answer changes how much of the process you need to document.
If you are earlier than that and still fixing your question, our dissertation topic ideas by subject includes the four feasibility tests worth running before you commit. And if the review is heading towards secondary analysis, the sources an undergraduate can actually access are listed in our guide to UK data sources by subject.
What Tesify does differently
Tesify is built around the division of labour described above. It works from your material — your sources, your notes, your findings — and helps you produce structure and momentum rather than substituting for your reading and your judgement. The argument, the evidence and the conclusions stay 100% written by you, which is the only version that survives a conversation with your supervisor.
More than 9,000 students have used it to get a stalled chapter moving. It is not a way to avoid writing a dissertation; it is a way to stop losing days to the blank page.
Start your literature review in Tesify.
Frequently asked questions
Is using AI to write my dissertation cheating?
It depends on your institution’s policy and on what you use it for. Using a tool to edit prose you wrote is treated very differently from submitting generated text as your own work. Read your academic integrity policy, follow any declaration requirement, and never submit writing you could not explain and defend.
Can Turnitin detect AI-written text?
Detection tools exist and are used by some UK institutions, but their reliability is contested and false positives are a documented concern. This is the wrong question to organise your behaviour around. The durable protection is being able to explain your own argument, produce your reading notes, and show the sources you cited — which no detection outcome can undermine.
What if I am wrongly accused of using AI?
Keep your working. Drafts with version history, dated notes, annotated PDFs and your search records collectively demonstrate a genuine process. Students who keep their materials are in a strong position; students who cannot show any working are not, regardless of what they actually did.
Can I use AI to summarise papers I do not have time to read?
No. This is the specific practice that produces misreported findings and citations of work you have not read. If time is short, narrow your scope and read fewer papers properly — a review of fifteen well-understood studies beats a review of forty misunderstood ones.
Is it acceptable to use AI to improve my English if I am an international student?
Language support is among the most widely accepted uses, and many institutions treat it similarly to proofreading. Check whether yours requires a declaration, and keep the edits at the level of expression rather than letting a tool restate your argument for you.
Will my supervisor be able to tell?
Supervisors who have read your previous work notice shifts in voice, and they notice when a chapter cannot be discussed in a meeting. The reliable signal is not stylistic — it is whether you can answer questions about your own reasoning. That is also exactly what the process described here protects.
What about the data privacy of my unpublished work?
Ask this of any tool before uploading a chapter. Your dissertation is unpublished research, and you should know where it is stored, who can access it, and whether it may be used for training, by reading the provider’s current documentation rather than a third-party summary.