Macmillan has retired NOA, an AI assistant that staff remember being welcomed in January 2024 with balloons, banners and a plate of doughnuts spelling out its name. Chaptr, the German software company that built it, ended it in October 2025, and the staff who spoke about it describe a tool that was unreliable and little used. Macmillan disputes how that account describes the rollout, and says it uses Google's Gemini for its business but has not told anyone to use it in the publishing process itself.
The failed chatbot is the smaller story. The larger one is what authors at the big publishers say is happening around it.
A chatbot for everything has no job
NOA could write emails, draft copy, do research and come up with ideas, and staff could set its tone to sarcastic or regretful. A tool that is for everything gives nobody a reason to open it on a busy morning. People adopt a tool when it takes over one task they already dislike, such as writing the weekly report or sorting the inbox, and they ignore it when it arrives with a party and a list of possibilities. Replacing it with another general assistant repeats the same mistake.
What the author is paying for
A publisher keeps most of what a book earns. In return, the author gets an editor who reads the book closely, a team that markets it, and a sales force that gets it into shops. Authors now describe the middle of that bargain going hollow: marketing copy with quotes that are not in their book, editors who ask authors to write their own revision plans, and big publishers that use ChatGPT and Claude for publicity and back-cover copy. Authors talk to each other and to their agents, and AI-written text gets spotted fast once people compare notes.
This is where the economics turn. The same chatbot that writes a publicist's back-cover copy writes it for the author too. Sanderson's 2022 campaign showed that the hard part of going alone was never the words. It was printing, packing and shipping, and even he said planning a full year of extras was more than his small company could handle. The words are now cheap for everyone. So the work a publisher automates is the work an author can least see a reason to pay a share for, and the work it cannot automate, an editor's attention, is the only part left to sell.
The same test for any business paid for attention
Take a bookkeeping firm that sends each client a monthly summary. If the summary is AI-written and generic, the client can paste their own spreadsheet into ChatGPT and get the same page for free. What the client cannot get is the accountant noticing that one supplier's invoice looks wrong and phoning them about it.
The better use is the reverse. Let AI do the reconciliation and draft the summary, which might save half an hour per client. For a firm with 40 clients, that is about 20 hours a month, half a working week. Spend those hours on the phone calls and the reading that clients can feel.
A customer will forgive AI behind the scenes much sooner than AI standing where they believed a person was. The authors in this story are not asking publishers to stop. The least they want is to be told when and how it is used, and any business that sells attention can offer that before its customers have to ask.