Macmillan’s AI assistant rollout exposes the publishing industry’s trust problem

Macmillan staff say the publisher introduced an internal AI assistant called NOA, urged employees to use it for work tasks, and later shut it down, before management began moving some workflows toward Google Gemini. The episode matters because it captures the larger strain inside book publishing: companies are trying to use generative AI to save time and money while authors, editors, and staff worry about accuracy, copyright, and trust.

What happened inside Macmillan

According to WIRED, Macmillan employees first noticed signs of NOA in January 2024, when they found launch-party decorations for the tool in the company cafeteria at headquarters. NOA stood for “No Ordinary Assistant,” and it was built by Chaptr with backing from Holtzbrinck, Macmillan’s parent company.

Workers who spoke to WIRED said the branding itself did not sit well with everyone. Some employees found the name offensive, while internal communications framed the tool as a safe place to handle proprietary data.

That promise came with limits. Staff were told that NOA would not share data with third parties or use it to train outside models. At the same time, they were warned that the chatbot could hallucinate and that human fact-checking was still required.

How was NOA supposed to be used?

According to staff accounts reported by WIRED, Macmillan encouraged employees to use NOA for emails, copy, marketing research, idea generation, and even drafting goals tied to performance reviews. Training advice reportedly also told workers to paste only a percentage of a manuscript at a time, rather than the full text.

Some employees, however, say the assistant was not reliable and ended up being used little. That gap between the company’s pitch and day-to-day usefulness is at the center of the dispute.

Why it matters

This is not just a story about one publisher’s software experiment. It shows how quickly internal AI tools can become a labor issue, a trust issue, and an authorship issue at the same time.

For publishers, generative AI promises faster drafting, cheaper research, and less repetitive work. But if staff are asked to use a system that can produce plausible errors, or if they do not trust the tool enough to rely on it, the hoped-for efficiency gains may never arrive. For authors, the concern is different but related: if publishers normalize AI in editorial and marketing workflows, writers may wonder where their own work ends and machine-generated language begins.

The issue is especially sensitive because several of the Big Five publishers are suing AI companies over copyright claims even as they experiment with AI internally. That contradiction is now visible rather than theoretical. Publishers are challenging outside companies in court while trying to decide how much of the same technology belongs inside their own operations.

There is also a commercial consequence. Some authors told WIRED that AI use has strained relationships with publishers enough that they are considering self-publishing. That matters in a market where major names already have strong alternatives. WIRED reported that bestselling authors it spoke with had combined sales of more than 30 million copies, and the leverage that comes with that level of audience can shift negotiations fast.

Background: how the book industry got here

The publishing business has been under pressure to cut costs and speed up work for years, and generative AI arrived as a tool that appeared to offer both. At the same time, the technology has raised familiar but newly urgent questions about copyright, source material, and the quality of machine-generated text.

Many large language models can produce convincing prose that is inaccurate or misleading. In publishing, that is not a small flaw. Editing, rights management, jacket copy, metadata, publicity materials, and acquisitions all depend on precision. A system that sounds confident while getting the facts wrong may save time in some settings and create extra work in others.

The Big Five publishers sit at the center of those tensions. They are large enough to experiment with internal tools and to push AI vendors on contract terms, but exposed enough to public scrutiny that any sign of machine involvement can alarm authors. That is why an internal assistant like NOA becomes more than a workplace utility. It becomes a signal about how management views human labor, editorial judgment, and the future role of the publisher.

How the Macmillan rollout changed

WIRED reported that in October 2025, Macmillan staff were told that Chaptr would sunset NOA. After that, employees said leadership began moving toward Google Gemini in parts of the company’s publishing workflows.

Some staff tried Gemini on agent submission letters and found the output useless. Macmillan, for its part, says Gemini is used for business purposes but denies that leadership directed staff to use it in publishing processes, including submission-letter review.

That denial is important because it suggests the dispute is not just about whether AI is present, but how deep into editorial or publishing decisions it reaches.

What Macmillan says

Macmillan disputes WIRED’s account of how NOA was rolled out and communicated. Kelly Collins, the company’s vice president of communications and community, said:

“there has not been any directive from Macmillan leadership to use Gemini for any part of the publishing process”

The company says its broader AI approach is centered on employee awareness, education, and human judgment. In other words, Macmillan is presenting its strategy as support for staff rather than a mandate to replace them.

That response leaves open a central question: how much of the pressure felt by employees came from formal policy, and how much came from the practical reality of how workplaces respond to new tools?

What employees and authors are saying

WIRED said it heard similar complaints about AI pressure from staff across all Big Five publishers, suggesting Macmillan is not an isolated case. The publication also reported that some authors believe AI use has damaged publisher relationships.

For authors, the stakes are both practical and emotional. A publisher’s marketing copy, submission handling, or review materials are part of the public face of a book. If those tasks are automated, even partially, writers may feel that the company is substituting machine output for the human expertise they expect in a literary relationship.

John Scalzi, whose 2024 Macmillan Tor Publishing Group deal was a 10-book contract, captured that unease bluntly. He told WIRED:

“I would be EXTREMELY annoyed to discover my PR is ‘AI’-generated”

That reaction helps explain why the conversation around AI in publishing has moved beyond efficiency. It is now about control, disclosure, and whether a publisher can adopt new tools without undermining the confidence of the people whose books it sells.

By the numbers

The dispute can be read more clearly through a few concrete figures and dates that came up in WIRED’s reporting.

What happens next

The biggest open question is how far internal AI use will spread inside publishing houses after NOA. WIRED reported that after the assistant was sunset, Macmillan moved toward Google Gemini in some areas, but it is still unclear which workflows are actually involved and how broadly the tool will be used.

Another unresolved issue is transparency. Macmillan says WIRED’s account does not accurately reflect NOA’s rollout and communication, while staff say the company pushed the tool more aggressively than they were comfortable with. More detail from Macmillan or Chaptr could clarify whether employees were being encouraged, required, or merely advised.

There is also a wider industry question: if one Big Five publisher is experimenting with one assistant, and WIRED says similar pressure complaints exist across the sector, then internal AI adoption may be developing faster than publishers are willing to describe publicly. That would leave authors and staff reacting after decisions are already underway.

Watch for three things next:

• whether Macmillan or Chaptr offers more specifics on NOA’s launch and shutdown;
• whether Macmillan clarifies which publishing tasks, if any, now involve Gemini;
• whether authors push for contract language that limits or discloses AI use.

Penguin Random House has already said through Claire von Schilling that the company is exploring how AI may appropriately support its work. That line suggests the larger publishing industry is not retreating from AI. It is trying to define limits in public after testing the tools in private.

What this means for readers

For readers, the immediate effect may not be visible on a bookstore shelf. But the internal decisions publishers make today can shape what gets acquired, how books are described, how quickly campaigns are built, and how much trust survives between writers and the companies that publish them.

That is why the Macmillan NOA episode matters beyond one office or one assistant. It is a case study in how a traditional media business is trying to absorb generative AI without losing the human judgment that gives publishing its value.