AI Disclosure Statement for a Thesis: Three Templates You Can Copy (2026)

What an AI disclosure statement is

An AI disclosure statement is now a standard part of a thesis submission at a large share of universities. It is a short signed section stating which AI tools you used, for what, where in the document, and how you checked the output.

It is not the same thing as your declaration of authorship. The declaration of authorship says the work is yours. The disclosure statement says what tools were involved in producing it, and it exists because the first statement stopped being self-explanatory around 2023.

Names vary — AI usage statement, declaration of AI use, generative AI acknowledgement, statement on the use of AI tools. The substance is the same: a named tool, a stated purpose, a located section, and evidence that a human checked the result.

Two things are worth knowing before you write one. First, there is no common standard: the required format differs by institution and sometimes by faculty. Second, in the German-speaking system the pattern is now well documented — a survey of 64 German universities found 58 requiring a declaration of AI use and none banning AI outright. Assume your institution wants something, and find out what.

This article covers what has to be in the statement, why institutions moved from catching AI to requiring declarations, and three templates — no AI use, language and structure only, and substantive use. Copy the one that describes what you actually did.

Why the rules moved from detection to declaration

The short version: detection did not hold up, and universities that acted on it got burned.

More than fifty universities have now disabled or demoted Turnitin's AI writing indicator. Vanderbilt turned it off in 2023, within months of launch, and published its reasoning. Waterloo followed in September 2025, Curtin in January 2026, Washington State in February 2026. The stated grounds repeat across the announcements: accuracy claims that could not be independently verified, false positives concentrated among students writing in a second language and students with distinctive writing patterns, and the structural problem that a student cannot disprove a score.

What replaced it is a disclosure regime. Instead of asking whether a machine wrote this, institutions ask you to say what you used, and treat the omission as the offence. The German legal guidance published on 16 January 2026 by the Baden-Württemberg higher-education network states the principle plainly: the deception is not using AI, it is failing to mark it — and the burden of proof for academic misconduct lies with the examination authority.

That cuts both ways. You are much less likely to be accused on the basis of a score alone. You are much more exposed if your statement turns out to be incomplete.

What has to be in it

Whatever the local format, five elements recur. If your institution has no template, include all five and you will satisfy most policies:

• The tool, by product name and version. Not AI, not a large language model. ChatGPT (GPT-5), DeepL Write, Elicit, Whisper (large-v3).
• The purpose. Brainstorming, literature search, translation, transcription, language editing, code, data analysis.
• The location. Chapter and section numbers. Policies that ask for the affected passages mean numbers, not the word partly.
• The verification. What you did to check the output — which claims you traced to a source, what you rejected and why. This is the part examiners actually read.
• The acceptance of responsibility. An explicit sentence taking responsibility for errors, bias and copyright in the final text. Some institutions require this wording verbatim.

Some policies go further. Where a tabular appendix is expected, a table of tool, purpose, affected sections and critical evaluation is the common shape. Some faculties additionally ask you to mark the passages AI created or altered inside the text itself, and some reserve the right to request a prompt and output log.

Template 1: no AI used

Write this even if you used nothing. A missing statement reads as an oversight; a present one is an answer.

Statement on the use of artificial intelligence

No generative artificial intelligence tools were used in the research, preparation or writing of this thesis. Standard spelling and grammar checking built into my word processor, and reference management software (Zotero 7), were used without any generative or text-producing features. All text, analysis and argumentation are my own.

If your reference manager or word processor has an AI assistant that you did use, this template does not apply to you — use template 2.

Template 2: language and structure only

This is the most common honest case: you wrote the thesis, and a tool smoothed the English or gave you feedback on the outline. Fill in the bracketed parts.

Statement on the use of artificial intelligence

I used generative AI tools in the preparation of this thesis for language and structure only. No AI tool generated content, arguments, results or references that appear in this work.

DeepL Write (web, academic mode) was used between [dates] to rephrase sentences I had already written, in Chapter [3] and Section [5.2]. Each suggestion was accepted or rejected individually; no passage was adopted unchecked. In [four] places the tool replaced technical terms with general vocabulary and I reverted those changes.

ChatGPT (GPT-5, web) was used on [date] for feedback on the chapter structure of Chapter [2]. No text from that exchange appears in the thesis; I adopted only the suggestion to reorder Sections [2.2] and [2.3].

I verified all output independently and I take full responsibility for the content of this thesis, including any errors, bias or copyright issues in the final text.

The specificity is not decoration. A statement that says AI was used for language polishing throughout tells an examiner nothing and invites the question in the viva. A statement naming chapters, counts and reversals closes it.

Template 3: substantive use

Use this where AI touched the research itself — literature search, data handling, transcription, code, or drafted text you then revised. Check first that your institution permits it; some allow substantive use only with prior agreement.

Statement on the use of artificial intelligence

The following generative AI tools were used in the preparation of this thesis. All output was independently verified and all final wording and argumentation are my own.

1. ChatGPT (GPT-5, web), [12 March 2026]: generation of candidate research questions during topic development, informing Section [1.2]. No wording was adopted; the final research question was written by me and agreed with my supervisor on [18 March 2026].

2. Elicit (web), [4 April 2026]: literature search on [topic], supporting Sections [2.1] and [2.2]. Eight results were returned; each was checked against DOI and library catalogue. Two could not be located and were not used. All sources cited were read in full.

3. Whisper (large-v3, run locally), [2–5 May 2026]: automatic transcription of six expert interviews, Section [4.1] and Appendix [C]. All transcripts were checked against the audio; technical terms and company names were consistently misrecognised and corrected manually. Analysis was carried out on the corrected transcripts only.

4. ChatGPT (GPT-5, web), [June 2026]: drafting of an initial summary of Section [5.1], subsequently rewritten by me. Approximately [15 per cent] of the sentence structures in that section originate from that draft; no factual claim in it is unverified.

I take full responsibility for the content of this thesis, including any errors, bias or copyright issues in the final text. A prompt and output log is available on request [or: is included as Appendix D].

Only offer a prompt log if you actually kept one. Offering a log you would have to reconstruct from memory is a worse position than not offering it, because a reconstructed log presented as a record is itself a false statement.

If the thesis becomes a paper

The publishing side has converged faster than the university side. Elsevier and Springer Nature both require authors to disclose the use of generative AI in the writing process in a dedicated statement, and both are explicit that an AI system cannot be listed as an author, because authorship carries accountability a tool cannot hold. arXiv now requires disclosure of generative AI use as well, having tightened its policies after a wave of submissions with fabricated references.

Two practical consequences. Keep your thesis disclosure detailed enough that you can convert it into a journal statement without guessing. And note that the journal rules apply to the writing, not the research tools: using AI for statistical analysis is described in your methods section, not in the AI statement.

Before you write the statement, read your own document

One practical step before you sign anything. If you ran whole chapters through a rewriting tool, those chapters read differently from the rest of your thesis — flatter, more uniform, with your habits sanded off. Knowing that before submission helps you describe the use accurately and means a question in the viva does not catch you off guard.

Our free AI check, up to 1,500 words, gives you a read on that, with no account and no signup. It publishes its own error rate next to the verdict: roughly 3 in 100 human-written academic documents receive the high probability verdict on English long-form text, measured on 40 human and 25 AI documents. Treat it as a mirror, not a verdict — a score is not evidence in either direction, and your disclosure statement is the thing that carries weight.

The hard part of all of this is the record, not the statement. Nobody writes down their prompts while they work, and the log cannot be made honestly afterwards. We are considering building something that keeps that log for you in the background and outputs it in whatever shape your institution asks for. If you would use that, tell us.