Does using AI for your literature review count as cheating?
The line is clearer than most people think. It is not about how much you used, and it is not about whether anyone can tell.
25 August 2026
Almost everyone doing research now has some version of this worry, and most of the advice available is useless in both directions. One half says avoid these tools entirely, which nobody is doing. The other half sells you something and assures you it is fine. Neither tells you where the line actually is.
Your own institution's policy is the binding answer and you should go and read it, because they differ and they have been rewritten recently. But underneath the policies there is a principle they nearly all share, and once you see it the individual rules stop feeling arbitrary.
The principle
A degree certifies that you can do research. So the work that is being assessed is the thinking. Everything else is logistics.
Nobody believes you personally typed every reference into a database, or that using a spell checker is fraud, or that a colleague pointing you towards a paper invalidates your review. Those are all forms of help with the labour, and research has always been full of them.
What cannot be delegated is the part being certified. Deciding what the literature means. Judging which study to trust. Working out why two papers disagree. Choosing what your argument is. If a tool did that, the certificate is describing work that was not yours, and it does not matter how the words got onto the page.
Three zones
Ordinarily fine
Help with the labour
Finding papers you would not have found. Summarising a paper so you can decide whether to read it properly. Formatting references. Fixing your grammar, which matters enormously if you are writing in your second or third language and has never been the thing under assessment.
The test these share: you could have done it yourself given enough time, and doing it faster does not change what you understand at the end.
Declare it
Help with the drafting
Generating text you then rework. Having a tool suggest a structure for a chapter. Asking it to identify themes across papers you have read.
None of this is automatically misconduct, and increasingly it is explicitly permitted. What gets people into trouble is doing it silently. Most institutions and a growing number of journals now require you to say what you used and what for. A short honest declaration costs you nothing. Being found not to have made one is a different conversation entirely.
Not fine
Help with the thinking
Submitting generated prose as your own writing. Citing papers you have not opened. Presenting an analysis you could not reproduce or explain. Letting a tool decide what your gap is and writing that up as your own conclusion.
This is the zone where the question stops being about policy. Even where nobody catches it, you have a thesis you cannot defend.
The test that actually works
Forget word counts and percentages. Take any paragraph in your review and ask whether you could be stopped mid-sentence and asked three questions.
- Which paper is this claim from, and what exactly did it say?
- Why did you believe it enough to build on it?
- What did you leave out, and why?
If you can answer those for every paragraph, how the draft was produced is a matter of disclosure rather than integrity. If you cannot answer them for a paragraph, that paragraph is a problem whether or not any software was involved. People write indefensible paragraphs unaided all the time.
This test is useful because it is the actual test. It is what a viva is, and what a reviewer does, and neither of them is checking your process. They are checking whether the person in front of them understands the work.
The risk nobody warns you about
The fear is being caught. Detection tools are unreliable in both directions, and if you are working honestly and declaring your use, that fear is mostly misplaced.
The real cost is quieter. Reading forty papers is not an obstacle between you and your review. It is how you acquire the judgement that lets you tell a strong study from a weak one, and how you learn what your field argues about. Skip it and you produce a review that reads adequately and leaves you unable to design the study that follows, or defend the one you did.
Which is why the useful question is not "am I allowed to use this?" It is "does this leave me knowing more or less about my field than doing it by hand?" A tool that finds you papers and shows you what they say leaves you knowing more. A tool that hands you finished paragraphs about papers you never opened leaves you knowing nothing, and it will show.
How to declare it
Plainly, in your methods or in an acknowledgement, naming the tool and what you used it for. Something like: used to search the literature and produce first-pass summaries of candidate papers, all of which were read and verified by the author, with all analysis and writing the author's own. If that sentence is not true of your work, write the sentence that is.
Why Gapnostic does not write your paper
This is the reason for the one rule Gapnostic is built around. It does the reading, the searching and the cross-checking, and it stops there. It will not draft your chapter, and that is deliberate rather than a feature we have not got to yet.
Everything it produces carries the paper and the sentence behind it, so you can answer all three questions above about any claim in your review. The thinking and the writing stay yours, which is the only arrangement that survives a viva.