Our Position

Readers of academic books need to be able to trust what they are reading. The arguments should belong to the author, the sources should exist, and the evidence should hold up when checked. Generative software has made it much easier to produce text that has the appearance of scholarship without the substance, and this page sets out how we deal with that.

We do not ban the use of AI in preparing a manuscript. Our expectation is a different one. Whatever tools an author chooses to use, the finished book has to be their own work, and they are answerable for all of it. This does not change because software was involved, or a translator, or an editor. Responsibility stays with the author.

These standards apply to every proposal and manuscript we consider, to every book we have published, and to all authors, volume editors, contributors and third parties involved in preparing that work.

Legitimate Assistance

Using technology to write more clearly is not a problem. Software that corrects grammar and punctuation, tidies up phrasing, helps with translation or keeps formatting consistent is a normal part of writing today. The condition is that the author stays in control of what the text actually says and argues. Authors can also use these tools in their own working process, for example to organise or condense research notes, as long as the manuscript they submit is written and checked by them.

On translation specifically: machine translation is fine where the author has confirmed that the translated text says what they meant it to say.

Unacceptable Use

Problems begin when software output is presented as the author's scholarship. We will not accept manuscripts containing analysis, argument or literature reviews that were generated by AI and submitted as original research. References, quotations and bibliographic entries produced by generative tools are not acceptable under any circumstances, because this material cannot be trusted. Nor should these tools be used to write authoritatively on subjects the author does not actually know, or to invent evidence, data, archival material or research findings.

If we ask an author what part AI played in the preparation of their manuscript, we expect an honest answer. An author who conceals significant AI use when asked about it has breached these standards regardless of whether the text itself turns out to contain errors.

The Risk of Convincing but False Material

The main danger with generative software is that its mistakes are convincing. In academic writing this shows up in a few familiar ways. Factual claims that sound authoritative but have no basis. Bibliographic entries for publications that do not exist. Real works cited for arguments they never made. Details taken from genuine sources and combined into something that appears nowhere. Once material of this kind is in print it gets cited, repeated and built upon, and it becomes very difficult to remove from the record.

For this reason, checking matters more than declaring. A reference that has been disclosed as AI-assisted but never verified is just as much of a problem as one that was never disclosed at all.

What We Expect from Authors

Before submitting, authors should have checked every reference and citation against the actual source, confirmed that every quotation is accurate and traceable, and satisfied themselves that the factual claims in the book rest on evidence they have personally examined. When our editorial team raises a question, we expect a full answer. When an error comes to light, whoever finds it, we expect it to be corrected quickly and openly. It is also sensible for authors to keep their research notes and drafts, since these are usually the fastest way to settle a question about how a particular passage was written.

Our Checks

Manuscripts go through originality screening, spot-checking of citations, verification of author credentials and publication history, and assessment for signs of machine-generated text. In most cases authors will not notice any of this. We look more closely when there is a reason to, for instance sources that cannot be traced, citation patterns that fall apart on inspection, claims that contradict established facts, credentials we cannot confirm, indications that large parts of a text were machine-written, or a concern raised by a reader or reviewer. If problems recur across an author's work, we will review their other titles with us.

Consequences

Most issues found during review are dealt with in the normal course of editing. Something gets corrected, a chapter gets revised, occasionally a manuscript has to be resubmitted. Serious cases have more serious outcomes, which can include rejection of the proposal, review or termination of the publishing contract, and suspension, correction or withdrawal of a published title.

Publication does not close the file. If a genuine concern about a book's integrity comes to our attention, we will look into it, whether the book came out last month or several years ago.