Five Signs your Publishing Workflows are not Ready for AI

Artificial intelligence is no longer a distant horizon for the publishing industry. It is already inside production pipelines, editorial teams, and content platforms, reshaping how organisations create, manage, and distribute content at scale. Yet for many publishers, early AI investments are producing results that fall short of expectations. The tools work. The workflows do not.

According to a survey of North American publishing professionals by the Book Industry Study Group (BISG), less than 30% of organisations have official AI policies in place, and only 27% use closed or enterprise models for data protection. The overwhelming majority report multiple pain points, with serious ethical concerns cited by nearly every respondent.

If AI adoption is accelerating and outcomes remain inconsistent, the diagnosis is almost always the same. The workflows were not ready. Below are five signs that yours might not be either.

Sign 1: Metadata Lives in Spreadsheets or Across Multiple Systems

Ask any publisher where their metadata lives and you will often get a pause before an answer. The honest reply is usually some version of "partly in our CMS, partly in a spreadsheet someone owns, and partly in the distributor's system."

This fragmentation is not a minor inconvenience. It is a structural barrier to any form of intelligent automation. Metadata is the connective tissue of modern content infrastructure. It determines how content is indexed, discovered, recommended, and licensed across platforms. When it is inconsistent, incomplete, or siloed, automation breaks down at the first step.

In scholarly and educational publishing, the stakes are particularly visible. The State of Scholarly Metadata Report found that metadata breakages "interfere with business transformation initiatives, contributing to high operational and opportunity costs" for publishers. And as AI-powered discovery tools become the primary interface through which readers, librarians, and educators locate content, poor metadata does not just slow workflows. It renders content invisible.

AI systems require clean, consistent, structured data to function reliably. If your metadata cannot be trusted by a human, it cannot be trusted by a machine.

The question to ask: If someone asked your team to produce a single, complete, accurate metadata record for every title in your catalogue today, how long would that take, and how confident would you be in the result?

Sign 2: Standards Alignment Is Handled Manually

Publishing operates within a web of external standards: curriculum frameworks, accessibility requirements, bibliographic standards, distributor specifications, and an evolving regulatory landscape. For educational publishers in particular, curriculum alignment is not optional. It is a core product requirement.

In most organisations, keeping content aligned with these standards is a manual job. Someone monitors changes, cross-references content, updates tagging, and manages the downstream effects of any shift in the framework. This approach is slow, inconsistent, and expensive. It is also increasingly untenable.

Curriculum frameworks change. Regulatory requirements evolve. New accessibility mandates emerge. Each update triggers a cascade of review and remediation work that falls on already-stretched editorial and production teams. When that work is manual, it cannot scale.

Organisations that have not built systematic, structured approaches to standards alignment will find AI amplifies their inconsistencies rather than resolving them. Automated content generation trained on poorly aligned source material will produce output that fails compliance checks at scale. The manual remediation burden grows, rather than shrinks.

The question to ask: When a curriculum body or regulatory authority updates its framework, what is your current process for identifying which content is affected and how quickly can you act on it?

Sign 3: Accessibility Is Addressed Late in Production

Accessibility in publishing has moved from a best practice to a legal obligation, and the compliance landscape is tightening. In the United States, the Department of Justice published its final rule under ADA Title II in April 2024, establishing enforceable standards for digital content including PDF documents. In 2024, over 4,000 digital accessibility lawsuits were filed in federal and state courts, with the first half of 2025 tracking nearly 20% higher. Separately, the WebAIM Million report found that 95.9% of homepages still fail basic accessibility standards.

For most publishers, accessibility is still treated as a late-stage production task: something checked at the end of a workflow rather than embedded from the start. The consequences are predictable. A 2024 peer-reviewed study analysing 20,000 scholarly PDFs found that fewer than 1 in 30 met all accessibility criteria, and that publishers routinely outsource remediation to specialist providers at significant cost and time, with no guarantee the result will even be accessible. Retroactive fixes are not just more expensive than building accessibility in from the outset. They frequently do not work.

This matters for AI readiness for a specific reason. AI-assisted content production moves fast. If accessibility is a manual checkpoint at the end of the pipeline, it immediately becomes the bottleneck that negates any speed gains AI creates earlier in the workflow. Publishers who want to genuinely benefit from AI acceleration need accessibility baked into their production architecture, not bolted on at the end.

There is also a quality issue. The Federal Register noted that even current AI technology "cannot reliably automate the remediation of STEM materials at scale, and human oversight is required to ensure accessibility." AI is not yet a substitute for structural accessibility design. It is a reason to get that design right earlier.

The question to ask: At what stage of your production process is accessibility currently checked, and what is your average cost and time per title to remediate accessibility issues after the fact?

Sign 4: Content Is Stored in Large, Monolithic Documents

Many publishing organisations still structure their content around the document as the fundamental unit of production. A chapter is a Word file. A course module is a PDF. A title is managed as a single, self-contained object from commissioning through to distribution.

This model made sense when publishing was primarily a print-to-digital conversion exercise. It makes very little sense in a world where content needs to be reused, recombined, personalised, and delivered across multiple channels and formats.

Here is what that looks like in practice. A rights team needs to license a subset of content to a new platform. An editorial team is asked to repurpose a title for a different curriculum. A production team needs to translate and reformat material for a new market. In a document-first environment, each of these tasks requires manual extraction, reformatting, and quality checking at every step. The content cannot be accessed in a structured, granular form, so the human scaffolding never goes away.

AI tools that could theoretically automate content assembly, personalisation, or translation are blocked by exactly this problem. Publishers building structured content environments, whether through XML-first workflows, component content management systems, or DITA-based architectures, can unlock automation at every stage of the content lifecycle. Those still operating in document-first environments will find AI delivers far less return than expected, because the cost of preparing content for it remains stubbornly manual.

The question to ask: If you were asked to create a new product that reused content from five existing titles in a different format for a different audience, how much of that work could be automated today, and how much would require manual copy-and-paste?

Sign 5: AI Tools Are Being Used Without Governance

Perhaps the most common and most consequential sign of AI unreadiness is the simplest: tools are being used without a framework to govern their use.

This is not a criticism of the teams adopting them. AI tools are genuinely useful, and the pressure to demonstrate productivity gains is real. But ungoverned AI adoption inside a publishing organisation carries operational, legal, and reputational risks that are difficult to quantify until something goes wrong.

The data tells a consistent story. Research from the National Association of Corporate Directors found that while 95% of senior leaders say their organisations are investing in AI, only 34% are incorporating AI governance, and just 32% are addressing bias in AI outputs. The IAPP's 2024 Governance Survey found that only 28% of organisations have formally defined oversight roles for AI governance. For publishers specifically, the BISG survey found that only 27% are using closed or enterprise models for their AI tools, meaning the majority are exposing potentially sensitive data to public-facing systems without clear data protection policies.

For publishers, the risks are specific. AI-generated content that has not been validated against editorial standards can introduce factual errors into educational or professional materials. AI tools trained on third-party content raise unresolved intellectual property questions. Automated outputs that have not been reviewed for accessibility or curriculum alignment can fail compliance requirements at scale. Without a governance framework that defines what AI is used for, how outputs are validated, and who is accountable, each of these risks compounds quietly until it becomes a serious problem.

Speed of adoption without governance is not efficiency. It is deferred risk.

The question to ask: If a regulator or institutional client asked you to demonstrate how AI is being used in your content production process today, what would you be able to show them, and what could you not?

Where Does Your Organisation Sit?

If several of these signs are familiar, that is not a cause for alarm. It is a starting point. Most publishing organisations are somewhere on a maturity curve between legacy workflows and AI-ready infrastructure, and the gap between those two positions is closable with the right diagnostic and the right prioritisation.

The organisations that will gain the most from AI are not those that move fastest. They are those that move with the clearest understanding of where they are starting from. Understanding where your organisation stands with AI today is the first step toward building workflows that AI can genuinely improve, rather than expose.

Find out where you stand. Take the free AI Readiness Assessment for Publishers.

About the Author

Gillian Malone-Johnstone is Head of Customer Success at Syllabyte, where she helps publishers leverage AI-driven solutions for standards alignment, metadata optimisation, and accessibility. With over a decade bridging AI-powered adaptive learning systems and real-world publisher workflows, Gillian has spent her career proving that the best educational technology amplifies human expertise rather than replacing it. Connect with her on LinkedIn.