Practical AI in publishing

What does AI actually look like in an editor’s day-to-day work? At the fourth Media Transformation Summit, Erin Servais moved the discussion away from broad promises about the technology and into the editorial workflow itself.

In AI for the Modern Editor: Practical Demonstrations for Publishing, Servais, president of Servais Training Company and founder of AI for Editors, demonstrated how AI can already help with tasks ranging from formatting references to extracting information from manuscripts and applying house-style rules — while drawing a clear line around the editorial judgment that should remain human.

The session was part of the Summit held online on September 29, organized by Publishing Signal and powered by knk. The event brought together publishing professionals from Canada, Brazil, South Africa, France, Germany, the United Kingdom, the United States, Australia and other markets, with sessions spanning AI, readers, discovery, accessibility, audio and publishing technology.

Introducing Servais, knk CEO Sebastian Mayeres suggested it was time to “make AI boring”, as a way to move the conversation from abstract promises to real tasks, real editors and real deadlines.

Servais did exactly that. But the demonstrations also raised a second, perhaps more important question: once AI enters the editorial workflow, where should the machine stop and the editor take over?

Servais began with what she described as a relatively low-risk use case: formatting references and citations. AI tools can convert references between style systems, research missing information and standardize entries — but only if editors understand the limits of the tool. AI, she cautioned, does not have direct access to every style guide and may not know recent changes. “It doesn’t know 100% of the rules with 100% accuracy,” Servais noted, which means editors may still need to fill in the gaps and supply the rules the system is missing. 

The second demonstration focused on data extraction, with AI used to pull and organize information from a manuscript — from character details to timelines and other elements editors often track manually.

Revision, however, came with a stronger warning. Servais demonstrated AI working directly inside Microsoft Word with a custom style guide, applying rules on spelling, punctuation, capitalization and other house-style decisions. She described this as the highest-risk of the three categories because it allows AI to change the text itself.

“We, the human editors, do not want to outsource our judgment,” she stressed. “We can let the AI make suggestions, but we need to make sure that it's the humans who are making the final decisions”.

That distinction became the central message of the demonstration. “These are machine tasks,” Servais explained. “We did not touch the heart of the book. We didn't touch the author's voice, we didn't do anything creative.”

Her rule of thumb was simple: reserve machines for rules-based tasks, and reserve for humans “decisions about creativity, beauty, connection — the things that humans excel with.”

The questions that followed quickly moved from capability to trust, accuracy and responsibility. One audience concern was familiar: if editors still have to check everything AI produces, does the technology actually save time?

For Servais, the answer is yes – provided the task is appropriate. Building character lists, timelines and other editorial aids once took her several hours per project. Having AI generate a first version in minutes changes that equation, even when the output still requires verification.

“I’m not ever going to say, ‘I think this is 100% accurate’”, she acknowledged. But even if a tool completed 90% of the work correctly, she argued, that could still represent a substantial saving compared with starting from scratch.

A more sensitive question concerned copyright, confidentiality and uploading unpublished manuscripts to AI systems. On copyright and confidentiality, Servais stressed that the first step is to follow the publisher’s AI policy and make sure there is explicit permission before uploading any manuscript or sensitive material. “Never upload information without express permission”, she warned, adding that freelancers should check the AI policies of every organization they work with.

Mayeres reinforced the point: the useful question is increasingly not simply whether AI was used, but how it was used, to what extent and who remained in control. In the examples Servais presented, AI was handling structural and rules-based work rather than altering author voice or making creative decisions.

Another concern from the audience was that cleaning up AI output can sometimes take longer than doing the work manually. 

Servais believes part of the problem lies in the instructions. “You have to get really specific and really detailed”, she explained, arguing that editors need to learn how to communicate with AI in the way the system can interpret effectively – not necessarily the way they would communicate with another person.

For editors curious to experiment, Servais recommended starting with a contained task. “If you work with references, try formatting references,” she suggested. For those who do not, she offered another everyday example: take one of those “paragraph-long sentences” that feels impossible to untangle and ask AI to simplify it. “Maybe it won't be 100% the way you want to use it”, she added, “but it will at least unwind it enough that you can see for yourself how to edit it.”

The broader message of the session was similarly restrained. AI may already be capable of taking repetitive, rules-based work off an editor’s desk. But using it well still depends on the things editors have always brought to the process: knowledge of the text, awareness of the rules, judgment about what matters — and responsibility for the final decision.