Since 2023 there has been a marked increase in the use of software based on artificial intelligence and large language models to both generate and edit scientific manuscripts. One popular such tool is ChatGPT, which is able to produce text that can easily be mistaken for having been generated by a knowledgeable human.

We decided to ask ChatGPT the following question: "Should I use ChatGPT to edit my scientific manuscript?" Its reply was as follows, which is broadly consistent with the considerable amount of information now available about this topic on the Internet:

Using ChatGPT as the sole editing tool for your scientific manuscript is not recommended. While ChatGPT can provide suggestions and help with certain aspects of editing, it is not a substitute for human expertise. Here are some important considerations:

1. Accuracy and scientific correctness: ChatGPT may not have the most up-to-date scientific knowledge or deep understanding of specific scientific fields. It can generate errors, inaccuracies, or inappropriate suggestions that could compromise the scientific integrity of your manuscript.

2. Limited context awareness: ChatGPT may not fully grasp the context, specific goals, or requirements of your research. Human editors with subject matter expertise can better understand the nuances of your work and provide targeted feedback.

3. Scientific conventions and standards: Scientific writing follows specific conventions, formatting guidelines, and citation styles. ChatGPT may not consistently adhere to these conventions or provide appropriate suggestions related to them.

4. Peer review and feedback: Professional editors or colleagues with expertise in your field can provide critical feedback, help address specific scientific requirements, and offer insights to improve the manuscript's overall quality.

While ChatGPT can be used as an additional tool to assist with editing, it's crucial to have human experts review your manuscript for scientific accuracy, coherence, clarity, and adherence to the relevant standards. Their expertise and critical evaluation are essential for ensuring the highest quality of your scientific manuscript.

One crucial aspect not mentioned above is that generative artificial intelligence tools such as ChatGPT work by harvesting text and data from the Internet and other sources. This means that there are major privacy and data-security implications of using such tools in your scientific work. These tools continually learn by processing information that they are provided with, and so you are essentially broadcasting your intellectual property to other users of these tools if you use them at any stage while you are writing a manuscript.

Moreover, and in contrast with ChatGPT, our editors at English Science Editing provide a considerable amount of feedback about the editing that they perform. In particular, when they are not certain about the scientific meaning being conveyed, they will flag this to the authors so that the changes that have been made can be properly checked.

In short, our editors understand the scientific content of documents, which ChatGPT most certainly does not, and this has serious implications when submitting your work for publication in the scientific literature. Finally, it must be remembered that many journals currently will not accept manuscripts that have been generated or edited using software such as ChatGPT. As an example, the following is an excerpt from the editorial policies of the American Association for the Advancement of Science (as at November 2023), which publishes several of the top scientific journals:

Text generated from AI, machine learning, or similar algorithmic tools cannot be used in papers published in Science journals...A violation of this policy constitutes scientific misconduct.