Book publishers are deploying AI across operations—but their own employees are increasingly refusing to use it. The paradox: organizational AI adoption at major houses jumped to 63% in 2026, while individual employee use dropped to 38%. This isn’t a publishing problem. It’s a change-management crisis that hits every mid-market sector.
Let me cut to it: the publishing industry just handed every mid-market decision-maker a masterclass in how not to deploy AI. And almost nobody noticed.
Here’s what’s happening. HarperCollins bought licenses for Claude, ChatGPT, and Jasper, appointed “AI Champions” to push adoption, and got so enthusiastic they spent more on software than they invested in employee raises. Simon & Schuster ran an employee AI contest with prizes—and then quietly explored workplace-monitoring software (Skan AI’s Blueprint) that could identify which tasks to automate away. Hachette supports operational AI but restricts it from author communications. Across the industry, AI is quietly generating book-cover copy, Amazon listings, Goodreads pitches, and back-cover descriptions, often without authors or even employees knowing it’s happening.
And the result? The Book Industry Study Group surveyed 771 publishing professionals in 2026. Organizational AI adoption climbed to 63%, up from 48% a year earlier. Individual employee AI use? It dropped from 46% to 38%. The organizations are buying the tools. The people doing the work are actively refusing to touch them.
The pattern nobody’s talking about
This isn’t a publishing quirk. I see it constantly across the mid-market companies I work with. The pattern is always the same: executive leadership declares AI a strategic priority, signs a big vendor contract, rolls out mandatory training, and expects adoption to follow. Instead, they get quiet sabotage—employees who drag their feet, who use the tool minimally and grudgingly, or who find ways around it altogether. The promised 20% productivity lift never materializes. Management blames the workers. Workers blame the lack of consultation and the vague threat to their job security.
In publishing specifically, the resistance isn’t subtle. Simon & Schuster employees gathered signatures opposing the monitoring software—not because they hate progress, but because the message they heard was: “We’re watching to see which of you can be replaced.” HarperCollins publicists complained that AI-generated pitches destroy authentic engagement with readers and media contacts. Workers cited doubled and tripled workloads, manuscript confidentiality concerns, and the sinking feeling that the company was willing to spend heavily on automation but not on them.
Among those surveyed, significant shares express ethical opposition to AI or lack interest in using it. Only 72% of those actually using AI report better efficiency—which means 28% are using it and seeing no improvement at all.
The top concerns? Inadequate controls on copyrighted material (86% cite this), accuracy of AI-generated information (89%), and the flood of low-quality AI-generated books destroying the market. These aren’t anti-technology sentiment. They’re legitimate operational and business-ethics concerns that leadership should have addressed BEFORE buying the tool, not after.
Why this matters for your 60-person manufacturing firm or your 80-person agency
The publishing breakdown matters because it exposes a universal adoption failure mode. You can have the best inference engine on the market, the most seamless integration with your workflows, rock-solid data governance—and still watch adoption stall if your team didn’t help design the rollout.
Imagine a 75-person logistics distributor. The ops director sees a vendor demo of an AI agent that could automate order-to-delivery workflows and save 15% of manual data entry. Smart move, right? Executive team agrees. Contract signed. System deployed. Six weeks later, dispatch staff are still manually entering half the orders because the AI occasionally misses a special-handling flag, and “going back to check what the robot did” takes longer than doing it right the first time. The warehouse team, never consulted, never trained on when to trust the system, simply doesn’t. The CFO sees the license cost and zero ROI. The vendor gets blamed. The tool gets shelved.
The real problem wasn’t the tool. It was that the implementation had zero buy-in from the people who actually move the packages. The dispatch team could have told you in week one that special-handling flags needed a different detection strategy. But they weren’t asked. They were told.
This is what I mean by the enterprise AI adoption trap: capability without readiness. You build capability—the technical ability to deploy AI—but you skip readiness: the organizational preparation, the cultural alignment, the workforce buy-in. Capability without readiness doesn’t accelerate. It stalls. And it burns credibility with your team.
The change-management piece nobody unlocks
The question for your business isn’t “Should we deploy AI?” It’s “Are we ready to deploy AI in a way that your team will actually use?”
That sounds soft. It’s not. It’s the hard operational difference between a 5% adoption curve and a 90% adoption curve. Here’s what structured readiness looks like in practice:
- Involve affected teams in the design phase, not the roll-out phase. If your dispatch team had been in the room when you scoped that order-automation project, they’d have said “flag detection is the blocker” before you signed a contract. You’d have chosen a different vendor or a different workflow. You’d have built on solid ground.
- Be transparent about what changes for people’s jobs. Not “we’re keeping everyone and AI just helps you work faster,” which is a lie everyone knows. Actually: “This tool will handle X, Y, and Z data entry. Your job shifts to quality review and exception handling. Here’s the training timeline. Here’s how your compensation changes if that shifts your hours.” Honesty costs less than mystery.
- Address the legitimate concerns first. In publishing, the top concern is copyright safeguards. You can’t just deploy an LLM trained on scraped books and hope nobody notices. Either you solve that problem architecturally, or you don’t deploy. Same principle across industries. If your team’s genuine concern is data security, answer it before launch, not after resistance hardens.
The publishers that are struggling missed this. They bought the tools, announced the mandate, and hoped momentum would overcome friction. It didn’t.
What to do Monday morning
If you’re planning AI deployment in the next 12 months—and most mid-market companies are—do this:
First, audit your organizational readiness. Not “do we have the right servers” readiness. Organizational readiness means: Do your teams understand what’s changing? Have you involved the people who’ll actually use this tool in the vendor selection or workflow design? Do you have a training plan that’s actually fit for your culture? Are you transparent about job-impact risks?
Second, pick a small pilot with a team that’s already curious. Not every team needs to adopt simultaneously. Let a subset of your publicists, dispatch staff, or customer-service reps try the tool, gather feedback, refine the workflow, and then scale. Adoption spreads through success stories, not mandates.
Third, build in a feedback loop and act on it. If 30% of your staff says “this adds work, not reduces it,” that’s data. It means either the tool isn’t right for your workflow, or your training approach isn’t working. Publishing leadership didn’t adjust. They pushed harder. Watch what happens: adoption rates drop further.
The irony is brutal. Publishers invested in AI because they thought it would improve efficiency and reduce headcount pressure. Instead, they’ve created a scenario where employees distrust the technology, adoption stalls, and the ROI vanishes. They spent the money wrong. They addressed a capability gap without addressing a readiness gap.
Your competitor might make the same mistake. If you don’t, you’ll see the adoption curve that HarperCollins should have seen: teams that are aligned, tools that actually get used, and productivity that actually materializes.
This is the actual AI transformation. Not the tool. The people.
0 Comments