Journals start to change their AI policies

Journals start to accept the use of AI for writing

Last week, I was preparing an article for submission to a SAGE journal and checked their current AI policy. I was positively surprised by their new AI guidelines.

SAGE now distinguishes between assistive AI and generative AI. Assistive uses include things such as: • improving language, grammar, spelling, punctuation and style • improving the structure and flow of existing text • formatting references

These uses DO NO LONGER NEED TO BE DISCLOSED.

More substantive generative use, for example, AI contributing to conceptual development, analysis, literature review or the generation of text, does need to be disclosed. They emphasize that the authors remain responsible for the work, which I think should go without saying.

sagepub.com/journals/publication-ethics-polic…

This is a very interesting development that seems to occur in parallel to the current race around AI detection. On one side, we see claims that AI-written text can be identified, followed almost immediately by tools promising to “humanize” that text so the detectors will not find it.

On the other side, publishers are beginning to acknowledge reality: People write with AI.

The meaningful distinction is increasingly not whether AI touched a text. It is how it was used, what intellectual work was delegated to it, whether that use needs to be disclosed, and who remains accountable for what is published. That seems to me a considerably more productive direction.

I am aware of the counterarguments that since the advent of AI, editors receive a flood of submissions and a lot of it is AI slop. One editor told me very confidently that he recognizes immediately whether it has been written with AI and that the article will be desk rejected.

Others say that it is difficult to detect and therefore difficult to sanction. I find this a very narrow perspective. Not everything that is written with AI is slop and needs to be sanctioned. Writing with AI is different from telling a chatbot: “Write me an article about....” and then submitting it without much further thought.

This is similar to uploading some data to ChatGPT and asking it to analyse it, e.g. by applying thematic analysis, grounded theory, or whatever. We do not need any more studies or authors telling us that this leads to sub-optimal results and therefore we need to reject the use of AI for qualitative analysis.

The same applies when it comes to writing. When used properly, it will greatly enhance the piece that comes out at the end, because you have not only used it for writing, but for thinking along while writing.

Let me share with you two AI disclosure texts for a publication, one written in the summer of 2025 and one written in the summer of 2026. Probably two things happened: I got better at using AI and at the same time the technology improved. The way I am now working with AI was not possible a year ago, largely due to an increase in context window limits, model abilities and also audio use.

2025 ChatGPT (OpenAI) was used to support the creation of the manuscript. The author was responsible for the content drafting, argumentation, and expert evaluation. ChatGPT was used, in particular, for section-by-section linguistic revision, streamlining, and improving readability, as well as occasionally as a sparring partner for clarifying and structuring arguments. All suggested changes were reviewed, revised, and integrated into the final version by the author.

2026 ChatGPT (OpenAI) was used during manuscript development as a dialogic writing and discussion aid, including to discuss and refine conceptual arguments, improve wording and structure, and assist with reference checking. All substantive arguments, interpretations and final wording were critically reviewed and determined by the authors, and cited sources were verified against the original publications. The authors take full responsibility for the content of the article.

I use the microphone a lot and just speak my thoughts. My partner recently caught one of those interactions and his first reaction was one of surprise. He said: But you are not asking a question. I said: Well, sometimes I do. But more often it is like having a back-and-forth conversation.

I also do not end every speaker utterance in a conversation with a human with a question. A conversation partner will infer and then react to what the other person says. And this is also what an AI interlocutor can do.

We may eventually look back at the obsession with identifying whether individual sentences were “written by AI” as a rather short transitional phase.