In my research, AI disclosure is not a perfunctory afterthought but a matter of scholarly integrity and methodological transparency. I treat it with the same seriousness as citing a critical source or declaring a potential conflict of interest.

I do not hide AI involvement behind vague phrases such as “AI assistance was used.” Instead, I explicitly identify:

1. The precise research stage at which the AI intervened — whether in hypothesis shaping, literature triage, data transformation, or structural drafting.

2. The granularity of delegation — whether the AI served as a conceptual provocateur, a linguistic editor, or a computational aide.

3. The exact identity of the system — including model name, provider, and version, e.g., “OpenAI GPT-5.”

My disclosures follow a GAIDeT-aligned structure, for example:

Certain preliminary textual formulations and editorial suggestions were generated using OpenAI’s GPT-5. All analytical reasoning, theoretical framing, and final conclusions remain entirely the author’s intellectual work. AI outputs were subjected to rigorous human verification prior to inclusion.

This approach achieves three objectives:

(1) Clarity — No reader is left guessing about the role of AI.

(2) Accountability — The human author retains unequivocal ownership of the work’s intellectual substance.

(3) Replicability — Future researchers can understand and reproduce the methodological environment.

Generative AI is a powerful instrument, but without disciplined disclosure, it risks becoming an invisible hand that erodes academic trust. By codifying AI involvement with precision, we preserve both the credibility of individual research and the legitimacy of the scientific record.

Hello,

Very interested in this topic as I could not find a suitable taxonomy or system for making such a declaration within a recent PhD thesis. One of issues I would raise with your github tool is that you focus exclusively on LLMs. In the study I use AI infused software such as MaxQDA - and we also need to accept that AI is now incorporated into email packages, and even standard software like Excel. So a focus on LLMs is a good start, but far far short of what will be needed to make a comprehensive declaration.

Hope this is of value,

K

Kieran Hanrahan Hello Kieran,

Thank you very much for your thoughtful comment. You are absolutely right — GAIDeT is intentionally limited to generative AI (mostly LLMs) and does not cover other types of AI-infused tools such as MaxQDA, email clients, or Excel. We are well aware that this leaves a large area outside its scope. Our decision was motivated by two factors:

1. Urgency of the problem — the strongest pressure from journals, funders, and policymakers at the moment concerns generative AI (particularly text generation and editing).

2. Feasibility of standardisation — since the functions of LLMs are relatively new and more easily distinguishable as discrete “delegated tasks,” we felt this was the right place to start.

That said, we absolutely agree that in the future the taxonomy should evolve towards a more comprehensive framework that also includes AI-embedded and “ambient” systems in research workflows. Your example with MaxQDA is a great illustration.

We would be glad to hear more about your experience with this, especially in the context of your PhD thesis — it could be very useful for the next iteration of GAIDeT.