Artificial intelligence is becoming part of the publishing conversation. Authors and publishers are asking where it can save time, what it can be trusted with and how to use it without losing control of the work.

The useful question is not whether AI belongs in publishing. It is which task we want it to help with, and who remains responsible for the result.

Where AI can help

Used carefully, language models can be useful for bounded tasks. They can suggest alternative ways to organise material, identify repeated explanations, help build a glossary from approved text or offer a plainer version of a difficult passage for an author to review.

These can be worthwhile starting points. They can also be wrong, incomplete or too confident. A suggested structure may look tidy while weakening the argument. A summary may leave out the qualification that makes a technical claim accurate. References, calculations and factual statements still need to be checked against their sources.

Efficiency figures can sound precise, but their value depends on the task, the material and the checking that follows. A percentage on its own tells an author very little about whether a book has become clearer or more reliable.

What still needs editorial judgement

In specialist finance, risk and technical non-fiction, the challenge is rarely just getting words onto a page. It is working out what the book is for, which reader it serves and how its ideas should build.

That calls for questions a model cannot settle on its own. Is the central claim supported by the evidence? Does the explanation preserve the technical meaning? Is uncertainty being presented honestly? Does the chapter order help readers understand the subject, or merely reflect the order in which the author discovered it?

An editor brings judgement to those decisions, as well as an understanding of publishing and readers. Subject experts bring the knowledge and accountability behind the work. AI may help explore options, but it cannot take responsibility for the argument, the analysis or the published claims. Springer Nature’s guidance similarly distinguishes assistive uses from decisions that require human judgement and accountability.

Protecting the work

A manuscript may contain unpublished ideas, commercial information, confidential data or material under review. Before putting any of it into an AI tool, authors and publishers should understand the tool’s terms and settings, the organisation’s policy and any obligations to co-authors, clients, publishers or journals. Where permission is needed, get it first. Do not assume that a tool’s privacy or training settings resolve every question about confidentiality or rights.

The position on copyright and AI is still developing. In its March 2026 report, the UK government said there was no consensus on the way forward and that it needed to build the evidence before deciding on copyright reform. That makes it sensible to read publishing agreements closely and discuss AI rights explicitly, rather than assume they are covered by a general grant of rights.

Authors and publishers may also want to distinguish between different uses: training a model on a book, making the book searchable by an AI system, or creating summaries and other outputs from it. Those uses raise different questions about control, permission and value. The terms should be clear before rights are granted.

Keep the author in the work

A book’s value is not simply the information it contains. It lies in the expert’s choice of what matters, the evidence they bring together, the shape of the argument and the care taken to make it useful to a reader.

AI can help with parts of that work. The editorial task is to make sure those parts serve the book, preserve the author’s expertise and leave a human being clearly accountable for the final result.

Before using AI on a manuscript, ask:

The answers will differ from project to project. The principle is steady: use the tool where it helps, protect the material entrusted to you and keep judgement with the people who understand the work.