How to Disclose AI Use in Dissertation: Honest Opinion by a Dissertation Expert
I have mentored dissertation students for over 12 years now. Till two years back, nobody asked me how to disclose AI use in dissertation writing. Now it comes up in almost every second consultation call. As an expert dissertation writer, I would rather give you the direct answer than the hedge-everything version most guides offer.
Fair question, and everyone is asking it. Everyone is using ChatGPT or Claude for something, even if it is just to fix a clumsy sentence. The problem is nobody told them how disclosure actually works, so they are Googling at 1 AM before submission, which is the worst time to make this decision.
I read through a stack of research articles, official university policies, and regulator documents before writing this. Some of it is useful. A good chunk of it is written by AI detection vendors who have a business reason to make you feel scared, or by editing-tool companies selling “humaniser” software. I will tell you plainly where I think they get it wrong, and I have linked the primary sources so you can check my reading of them yourself.
Why This Question Is Blowing Up Right Now
Two things changed since late 2023. First, the UK’s Russell Group published its Principles on the Use of Generative AI Tools in Education in July 2023. A large share of UK research-intensive universities now expect explicit AI-use disclosure in submitted work as a result. Second, Turnitin’s own reporting shows the scale of the problem: submissions scoring 80% or higher on AI writing detection climbed from about 3.3% of papers at launch in 2023 to roughly 14.8% by early 2026.
That jump is why supervisors and committees got twitchy. It is not because AI use itself is new. It is because the volume crossed a threshold universities could no longer ignore.
As a thesis writing expert working across UK, US, and Middle East student cohorts, I notice dissertation expectations have shifted faster than most departments have updated their written policy. Students are being held to an unwritten, stricter standard even where the official rulebook has not caught up.
My Honest Take After 12 Years of Doing This
Here is where I will disagree with a lot of what I read while researching this piece. Several articles claim there is a “safe” AI percentage, something like “10 to 20% is fine, above 50% is dangerous.” That is false precision dressed up as advice.
Turnitin’s own guidance states that any AI score under 20% is marked with an asterisk precisely because it is not reliable enough to act on. So an article telling you “12% is safe” is quoting a number Turnitin’s own documentation says you should not trust. I would rather tell a student the honest, less satisfying truth: there is no universal safe number, only what your specific university’s policy and your own actual usage pattern support.
Do You Even Need to Disclose? A Quick Decision Framework
Not every use of AI needs a formal statement. Here is how I break it down for my own students, task by task.
- Spelling and grammar check only (Word’s built-in checker, basic Grammarly), generally no disclosure needed. This is treated the same as a dictionary.
- Language polish and tone improvement on your own drafted text, disclose in your acknowledgements, keep it brief.
- Brainstorming, outline suggestions, or clarifying a methodology term, disclose, and be specific about what you asked it to do.
- Drafting actual paragraphs, analysis, or arguments, this needs disclosure and, in most institutions, prior committee approval. Several universities, including the University of Georgia, prohibit this outright unless the advisory committee has specifically authorised it.
- Statistical analysis or code help (SPSS syntax, R scripts, Stata do-files), this sits in its own category, and I will explain why below.
If your use falls anywhere in categories 2 to 5, disclose it. When in doubt, disclose. I have never seen a disclosed, honestly described use of AI cause more damage than an undisclosed one that got noticed later.
AI Use Disclosure Statement: What It Actually Means
An AI use declaration is a short, written statement in your dissertation that names the AI tool you used, describes exactly what task it helped with, and confirms that you reviewed and take responsibility for the final content. It is not the same as a plagiarism declaration, and it does not admit wrongdoing. It is documentation, similar to citing a source.
This distinction matters more than people realise. A plagiarism declaration says “this work is original.” An AI use declaration says “here is exactly how a tool assisted, and here is what remains entirely mine.” The two statements sit in the same dissertation without contradicting each other, as long as your actual usage matches what the AI use declaration describes.
Turnitin AI Detection: What the Score Really Tells You
This is where most of the panic comes from, so let me slow down here. The short version of how AI detection works in Turnitin: the system compares statistical patterns in your writing, word predictability, sentence rhythm, structure, against patterns typical of large language model output. It produces a percentage estimate, not a certainty.
Turnitin is not the only tool in this space. GPTZero and Originality.ai run on similar principles and are commonly used alongside it. Cross-checking a flagged paragraph against a second detector tells you more than trusting any single score in isolation.
AI Writing Score vs Similarity Score
These are two completely separate numbers on your Turnitin report, and confusing them is the single biggest source of misunderstanding I see among students. The similarity score checks for text matching other sources. The AI writing score estimates how much of your text resembles patterns generated by large language models.
You can have a low similarity score and a high AI score at the same time. That combination alone does not prove plagiarism. It suggests your writing pattern statistically resembles AI output, which is a different claim entirely.
What AI Score Is Too High for a Dissertation? (Turnitin AI Detection Percentage, Explained Honestly)
Turnitin does not attribute a reliable score below 20%, marking anything from 1 to 19% with an asterisk instead of a number. Above 20%, scores are considered more reliable and tend to invite closer review, but Turnitin itself is clear that no score, on its own, proves misconduct. A 2026 peer-reviewed evaluation by Hadra, Cambridge, and Mesbah, published in the International Journal for Educational Integrity, found Turnitin’s overall accuracy around 0.61 on a mixed set of human, AI, and hybrid texts, well short of the “near-perfect” impression some marketing material creates.
I want to flag something directly here. Turnitin’s own transparency material claims 98%+ accuracy with under 1% false positives, but that figure applies only to documents with more than 20% AI-generated text, under Turnitin’s own test conditions. Independent academic testing across a broader mix of writing reports meaningfully lower accuracy. I would rather you see both numbers before deciding how worried to be.
If you want to check dissertation for AI before submitting, use your institution’s own Turnitin student-view access where available, or ask your supervisor to run a preliminary check rather than relying on a random free online detector. Cross-checking the AI similarity report against your own honest usage log tells you far more than chasing a specific percentage.
Here is a quick reference so you are not guessing at what a number on your own report actually means.
| Turnitin AI Score | What Turnitin Says | What It Means for You |
|---|---|---|
| 0% | No AI writing detected | No action needed |
| 1-19% (shown as *%) | Not reliable enough to report as a number | Don’t treat this as evidence either way |
| 20-49% | Reported as a number, moderate confidence | Worth discussing with your supervisor if undisclosed |
| 50-79% | Higher-confidence flag | Check your disclosure statement matches this level of use |
| 80%+ | High-confidence flag | This band grew from 3.3% to 14.8% of submissions between 2023 and early 2026, so expect closer review |
False Positives and Non-Native English Writers
A 2023 Stanford study found that seven widely used AI detectors flagged 61% of TOEFL essays by non-native English speakers as AI-written, even though every essay was written entirely by a human. Turnitin has publicly acknowledged this concern and adjusted its thresholds since. If you are writing your dissertation in your second or third language, this bias is a real risk, not a hypothetical one. That describes a large share of the international and multilingual students I mentor across Gulf universities.
This is exactly why disclosure protects you rather than exposing you. A clearly written, honest AI use statement gives your committee context before they see any detector score. Silence gives them nothing to weigh a false flag against.
University AI Policy: UK, US, and Middle East
If you searched something like “university AI policy dissertation,” here is the honest answer: there is no single one. It varies a lot by country and even by department within the same university, so treat what follows as a map, not your specific rulebook.
In the UK, is AI allowed in dissertations at all? Mostly yes, within limits, and mostly with disclosure attached, a stance the Russell Group’s 2023 principles pushed most research-intensive universities toward. Cambridge’s own guidance treats undisclosed AI use in summative assessments as academic misconduct comparable to plagiarism, and it publishes a downloadable declaration template for students. Oxford follows a similar “cite it like any other source” logic.
If you are specifically searching for a dissertation expert UK universities will recognise as reliable, know this: Russell Group versus non-Russell Group policy nuances matter more than most generic guides let on. Always cross-check your own department’s written policy rather than relying on a country-wide assumption.
In the US, every generative AI academic integrity policy reads a little differently depending on the school. The University of Georgia prohibits generative AI in theses and dissertations unless the advisory committee specifically authorises it. Any approved use must then be documented in a statement within the thesis itself. Illinois State goes further, requiring all generative AI use in the dissertation process to be disclosed to the committee, full stop, regardless of how minor.
Stanford’s stance is simpler: disclose AI use and follow your instructor’s specific guidelines, since the university-wide rule leaves interpretation to individual departments. Most graduate school AI guidelines I have reviewed follow this same pattern, a broad principle at the top, with the real decision pushed down to your supervisor or committee.
In the Middle East, the pattern is disclosure-permitted, submission-restricted. United Arab Emirates University’s official Generative AI Policy allows students to use GenAI tools for academic and research purposes, but requires acknowledgement of that use and mandates that faculty run AI-detection checks on submitted work. Violations, including unreflective AI-generated text presented as original, carry academic penalties up to zero marks on the affected section.
This pattern holds broadly across UAE institutions. Most UAE universities require a Turnitin similarity score under 15 to 20%, with the AI-detection layer checked separately, and generally permit AI tools for brainstorming, outlining, and reference discovery while prohibiting AI-generated text in the submitted dissertation itself. Khalifa University applies particularly strict STEM research standards, and UAEU additionally requires bilingual Arabic and English abstracts regardless of your AI use.
My honest read on this region: the policies are newer and less tested by actual misconduct cases than the UK’s or the US’s, so committees tend to default to caution. Disclose more conservatively here, not less. If a policy document does not explicitly permit a specific use, treat it as requiring committee approval first, rather than assuming silence means permission.
Here is the same three-country picture side by side.
| Country | General Stance | Key Authority | Disclosure Expected? |
|---|---|---|---|
| UK | Disclosure over blanket ban, set out sector-wide | Russell Group Principles (2023); Cambridge, Oxford guidance | Yes, at most research-intensive universities |
| US | Varies by university and department, no single federal rule | UGA (prohibits unless approved); Illinois State (mandatory disclosure); Stanford (defers to instructor) | Yes, but scope is set locally |
| Middle East (UAE) | Disclosure permitted, submission restricted; AI checked separately from similarity | UAEU official GenAI Policy; Khalifa University STEM standards | Yes, disclosure required wherever GenAI is used |
How to Disclose AI Use in Dissertation: A Step-by-Step Statement Guide
Here is the exact process I walk my own mentees through when they ask me how to write an AI disclosure statement for their thesis or dissertation.
- Name the tool precisely. Write “ChatGPT (OpenAI, GPT-5), accessed March 2026” rather than just “AI.” Vague naming is the most common mistake reviewers flag.
- Describe the specific task, not the general vibe. Say “used to suggest alternative phrasing for three paragraphs in the literature review” rather than “used for help with writing.”
- State what you personally reviewed and changed. This is the sentence that actually protects you, showing you retained intellectual control over the final text.
- Place it in the correct section. Language-only assistance goes in your Acknowledgements, sometimes written as a short generative AI acknowledgement section rather than folded into general thanks. Substantive, committee-approved use goes in a dedicated Declaration section, sometimes alongside your Methods chapter if it affected your analysis approach.
What counts as acceptable AI use in thesis writing ultimately comes down to two things. First, whether your committee approved it in advance for anything beyond editing. Second, whether your final disclosure honestly matches what you actually did.
A Note on Statistical Analysis and Coding Help
This is a category most AI-disclosure articles skip entirely, and it is one I deal with constantly through my own statistical consulting work. If you used AI to help write an R script or debug SPSS syntax, that is methodological assistance, not writing assistance. It usually belongs in your Methods chapter rather than your Acknowledgements. Disclose the tool, the specific script or procedure it helped with, and confirm you verified the output against your actual data.
If you are unsure how to document this correctly, a statistical consultant can walk you through it properly before submission, rather than you guessing.
Citing ChatGPT or Claude in APA 7 and MLA 9 (If Your Style Guide Requires It)
A disclosure statement and a citation are two different things, and mixing them up is a common mistake. A disclosure statement tells your committee you used a tool. A citation is what you add in-text if you actually quote or closely paraphrase AI-generated text somewhere in your argument.
In APA 7, the AI tool is treated as an algorithm rather than an author, so you credit the company instead. The reference looks like OpenAI. (2026). ChatGPT (GPT-5 version) [Large language model]. https://chat.openai.com/chat, and in text you cite it as (OpenAI, 2026), the same way you would cite any software.
In MLA 9, the approach is the opposite. The AI tool is never treated as an author, and the entry is built around your actual prompt as the “title of source”: "Summarise the key debates on X" prompt. ChatGPT, model GPT-5, OpenAI, 14 Mar. 2026, chatgpt.com.
Neither format replaces your disclosure statement. Use both together: the disclosure statement in your Acknowledgements or Declaration section, and the in-text citation only where you have actually quoted or closely paraphrased AI output.
Ready-to-Use AI Use Declaration Template (Copy and Adapt)
Pick the sample AI usage statement that matches your actual usage. Do not upgrade or downgrade the wording to sound more impressive or more modest than what actually happened.
Template 1: Editing or language polish only: “I used [tool name, version, access dates] to check grammar and improve sentence clarity in my drafted text. All arguments, analysis, methodology, and conclusions in this dissertation are my own original work.”
Template 2: Research assistance or brainstorming: “I used [tool name, version, access dates] to brainstorm chapter structure and clarify terminology related to [specific topic]. The output was used only as a starting point; all content, analysis, and conclusions were independently developed and written by me.”
Template 3: Committee-approved substantive use: “With prior approval from my supervisory committee dated [date], I used [tool name, version] to assist with [specific approved task, e.g., drafting an initial literature summary in Chapter 2]. I reviewed, verified, and substantially revised all AI-assisted content, and I take full responsibility for its accuracy and originality.”
Will Disclosing AI Use Hurt Your Grade? My Honest Answer
No, not by itself, and I say this after sitting through several committee conversations where disclosure came up directly. What actually damages a grade is weak analysis, thin argumentation, or a mismatch between what you disclosed and what a detector or a sharp-eyed examiner later notices. Disclosure is not the risky part. The gap between what you say you did and what you actually did is the risky part.
If you are wondering how to make disclosed AI use acceptable to committee members specifically, the honest answer is precision and consistency, not minimising what you did. Here are two composite examples built from patterns I have seen repeatedly across mentoring conversations, names and details changed throughout.
Case study: disclosed. A student used ChatGPT to restructure a messy literature review section. She disclosed it clearly in her acknowledgements, naming the tool and the exact task. Her committee raised zero concerns during her viva, and one member specifically said the disclosure made the chapter easier to evaluate fairly.
Case study: undisclosed. A different student used AI at a similar level but did not disclose it. A committee member noticed unusually uniform sentence rhythm in one chapter and raised it directly. The conversation that followed was far more uncomfortable, and far longer, than any disclosure statement would have been.
The pattern I see repeatedly: disclosure closes the conversation before it starts. Silence opens it up later, at a worse time, with less control over the framing.
What If You Already Submitted Without Disclosing?
Talk to your supervisor before anyone else does. This is not pleasant advice, but it is the honest one. Explain exactly what you used the tool for, provide the same detail a proper disclosure statement would have included, and ask how your institution handles retroactive declarations.
Some departments allow a supplementary declaration to be added even after submission, especially if a viva or defence has not happened yet. If you are staring at this exact situation right now, unsure what your options are, this guide on what to do if you have been accused of using AI in your dissertation walks through the practical next steps.
How to Reduce AI Detection False Positives, Honestly
A few honest, non-gimmicky things actually help here, separate from any “AI detector bypass” tool, which I will not recommend because most of them just add noise without addressing the actual concern. None of this is about hunting for a writing style that avoids AI flags artificially. It is about writing naturally, in your own voice. The whole point is to reduce risk of AI misconduct flag situations as a side effect of honest process, not as the goal itself.
- Write your first full draft yourself, then use AI only for targeted revision passes, not generation from scratch.
- Vary your sentence length and structure naturally. Uniform sentence rhythm across long stretches is one of the genuine statistical signals detectors pick up on, disclosed or not.
- Keep a simple log of your AI interactions (prompts and dates) as you go. It costs you two minutes and gives you a paper trail if anyone ever questions your process.
- Get a second pair of expert eyes on your draft before submission. A dissertation mentor who has actually sat through committee reviews will catch inconsistencies you cannot see in your own writing.
When to Get Expert Eyes on This Before You Submit
I will be direct about this, since I run a dissertation coaching practice and it would be dishonest not to mention it. Plenty of students search “dissertation editing services near me” at 2 AM, panicking about a single flagged paragraph. You do not need someone physically nearby for this. You need someone who understands both the writing and the policy side of AI disclosure.
If you are unsure whether your university would count your AI use as something needing disclosure, or you want a second opinion before your viva, this is worth a proper consultation rather than a forum post. Between structuring your chapter organisation, tightening formatting that examiners actually notice, and getting the AI disclosure wording right, small errors compound. If you are still deciding how to know if you need a dissertation expert at this stage, an AI disclosure review is a low-cost, high-value place to start.
Frequently Asked Questions
How do I write an AI acknowledgement section? Name the specific tool and version, state exactly what task it assisted with, and confirm you reviewed and take responsibility for the final content. Keep it to two or three sentences placed in your Acknowledgements or a dedicated Declaration section, depending on how substantial the AI’s role was.
Can Turnitin detect ChatGPT accurately? Turnitin can flag text patterns statistically associated with AI generation, but its own independent accuracy testing shows results well below its marketed best-case figures, especially on hybrid human-AI writing. Treat any AI score as one signal to discuss with your supervisor, not a final verdict.
Is using Grammarly considered AI use I must disclose? Basic grammar and spelling correction generally does not require disclosure, since it is treated similarly to a dictionary or spell-checker. If you use Grammarly’s more advanced rewriting or tone-adjustment features on substantial passages, disclosing that use is the safer, more transparent choice. This is really the same question as “do I need to disclose ChatGPT use” for something minor, and the answer follows the same logic: small mechanical edits, no; substantial rewriting, yes.
What happens if I do not disclose AI use and get flagged later? Consequences depend entirely on your specific university’s academic integrity process, ranging from a conversation with your supervisor to a formal misconduct review. Undisclosed use that surfaces later is treated far more seriously than the same use disclosed upfront, because the concern shifts from “how much AI” to “why did you hide it.”
Can non-native English speakers get falsely flagged as AI? Yes. The Stanford study cited above found detectors flagging 61% of non-native English writers’ genuine, human-written work as AI-generated in testing. This is a documented limitation of current detection tools, not a reflection of your actual writing.
Do I need to declare AI use if I only used it for editing? For basic grammar and spelling correction, generally no. For anything beyond that, tone changes, restructuring, or rewriting whole passages, disclose it in your Acknowledgements, since the safer default is always disclosure over silence.
Siddharth Gupta has spent over 12 years guiding dissertation and thesis students through research design, statistical analysis, and academic writing. Through statssy.com’s dissertation coaching and statistical consulting practice, he has personally reviewed AI disclosure statements and dissertation drafts for clients across the UK, the US, and the Middle East. Connect on LinkedIn.