MHRA AI Policy for Authors and Peer Reviewers
Artificial intelligence (AI) technologies include large language models, prompt-based text/image generation, etc. AI offers new opportunities to many industries and raises complex legal and ethical questions for society to resolve. This Policy is not the MHRA’s response to such debates. It is rooted in practical issues which we must address as a scholarly publisher: the need to avoid misrepresentation of facts, misattribution of opinions, and plagiarism; and the need to give clarity to readers and research libraries on how the text in our books and journals has come into being.
This page gives both our Policy for Authors and our Policy for Peer Reviewers.
1. Policy for Authors
A distinction is made between AI generation and AI assistance. AI generation refers to material (text or images) created by or originating from an AI tool, following a prompt from a user. AI assistance refers to situations in which material created by the author(s) is then refined or edited using AI tools.
AI-generated material is not permitted in MHRA journals or books.
While the MHRA does not prohibit the use of AI assistance in refining work for submission, it is not recommended.
Authors are responsible for verifying the accuracy and appropriateness of all material and citations in their work.
Use of AI tools for translation is not recommended, but is permitted for short passages of quotation from languages other than English. In such cases the use of an AI translation should be clearly marked. Pre-existing translations should always be sought in the first instance, and wherever possible translations should be checked by a language specialist.
For those authors whose first language is not English, an AI translation tool may be of use in drafting an English-language text of a book or article, but such drafts will always need checking and improving, and the material should not be submitted to the MHRA for publication until at least one native English speaker has gone through it. For larger-scale use of AI translation, such as to translate an entire article or chapter already existing in another language, please contact the relevant Editor in advance to check whether this will be allowable.
Examples
These are included for illustrative purposes and are not intended to be exhaustive. Being listed as an acceptable use does not mean that a particular practice is recommended or encouraged, and special care should be taken to verify the results of any AI assistance. Because of the way in which AI tools work, suggestions made by such tools may be of low quality and may introduce errors into your work.
Scenario: drafting the text of an article or book chapter.
Acceptable use (assistive): using an AI-based tool to check for spelling and grammar errors (e.g. built-in grammar checkers, predictive text in Microsoft Word). Users should be aware that such tools can be wrong, or use non-UK spellings, or misunderstand when non-English text is being quoted.
Unacceptable use (generative): using an AI tool to rewrite passages and suggest improvements; using an AI tool to suggest links between topics, suggest counter-arguments, or provide a draft structure based on pre-written material provided to the tool; using an AI tool to generate text.
Scenario: conducting a review of prior literature.
Acceptable use (assistive): using an AI search tool to discover relevant publications in the area, which are subsequently read; using an AI referencing tool to format references provided to it in the correct referencing style (this should be checked carefully). Traditional search engines are preferable, for increased accuracy and reduced bias in their results.
Unacceptable use (generative): using an AI tool to summarise publications or paraphrase quotations; relying on an AI-generated answer to a query without checking the sources; listing an AI-generated answer as a source; citing references provided by an AI search tool without having read the referenced source.
Scenario: writing an abstract for a journal article.
Acceptable use (assistive): using an AI-based tool to check the grammar of an abstract you have already written.
Unacceptable use (generative): using an AI tool to generate an abstract on your behalf by submitting your manuscript to the tool, or by using built-in summary tools. Such summary tools work by uploading submitted material to the company’s servers, often without the user’s prior consent.
Scenario: a large collection of data requires analysis.
Acceptable use (assistive): using an AI tool to perform basic tasks such as formatting and sorting. All such uses should be fully declared.
Unacceptable use (generative): using an AI tool to generate graphs or other visualizations based on the data, or to provide written summaries of its findings; using an AI tool for any generation, correction, cleaning, or editing of data. Spreadsheet and graphing software will produce graphs which can be relied on to be correct, and will also serve authors better in generating print-quality images.
Scenario: writing on a literary text that is mainly in English but that includes short Latin tags.
Acceptable use (assistive): using an AI tool to translate the Latin passages to obtain a rough idea of their content; such translations would not normally be included in the finished article, but when they are, the use of AI tools should be fully declared.
Unacceptable use (generative): using an AI tool to translate entire texts or large portions of texts that have major significance for your research; publishing any AI-based translation as your own work.
Scenario: a historical photograph available only in a low-quality version in which it is difficult to make out the subject.
Acceptable use (assistive): cleaning up the image in Photoshop using an AI-based de-noise plugin, or using the healing tool to remove dust specks or damage, where such edits are genuinely helpful to a clear interpretation of the image. Archival photographs have a physical reality which is important in itself, and to remove signs of water damage or other blemishes from them should not be done casually. On the other hand, images now available only through poor-quality preservation (such as scanned newsprint or microfiche archives) may well need tidying.
Unacceptable use (generative): generating a new version of the image; asking an AI tool to produce a similar image to an existing one (since the result will be an invention).
If in any doubt at any point, please contact the relevant Editor.
2. Policy for Peer Reviewers
It is not appropriate to use AI tools for the purposes of peer review, other than for minor assistive tasks.
Peer reviewers are requested not to upload a submitted manuscript or any part thereof to an AI tool of any kind. This includes inbuilt AI summary features (e.g. in Adobe Acrobat, Microsoft Copilot, etc.). Doing so is an infringement of the rightsholder’s intellectual property and may also involve unlawful disclosure of the author’s personal information. If asked to summarise a long document, many AI tools do so by uploading it to the cloud. Such text is frequently then kept and used for model training without consent being asked. This does not keep faith with our authors, may compromise their personal data, and may mean that their still-unchecked, uncorrected drafts are ascribed to them by future search engines.
Peer reviewers should also be aware of our author guidelines on AI use: material generated by AI is not permitted in MHRA publications, and reviewers are requested to be alert for such material wherever possible. Please raise any concerns in this regard with the relevant Editor.