Every page could have been page one a long time ago: are we ready now?

Every page could have been page one long ago. When I read Mark Baker’s Every Page is Page One: Topic-Based Writing for Technical Communication and the Web in 2014, long before artificial intelligence became part of my daily vocabulary, the idea felt both obvious and almost impossible: a user manual could be complete without forcing every reader through the same sequence.

I did not read it as a trend. I read it as an explanation of something I had already been experiencing in technical documentation: user manuals were written as linear books, while readers entered them through a problem, a task, a machine state, a fault code, or a question.

The book gave that experience a name, a structure, and practical principles for doing things differently – and better. It also made me imagine a user manual that did not have to behave like a book, even when delivered as a PDF.

I started dreaming of hyperlinked reference books

Back then, I wanted to build a manual that behaved like a reference book – not a conventional book that readers had to read from front to back, but a structured reference that could be entered through different routes.

I imagined an orientation page that acted as a map of the machine. From there, readers could enter through tabs for:

• the task they needed to perform;
• the machine state they were facing;
• the component they were working on;
• the hazard they needed to control;
• the operating mode they had selected;
• the fault code displayed on the HMI; or
• the configuration of the machine in front of them.

The familiar categories – Safety, Installation, Operation, Maintenance, Troubleshooting, and Technical Data – would remain. They would function as tabs and access points, opening different search and navigation routes rather than imposing one mandatory reading order.

A topic such as checking a protective door could be reached through commissioning, maintenance, safety, or troubleshooting. The content would remain one controlled topic instead of being copied into four places.

That was the form I could see before there was a practical environment in which to build it: not a sequence of pages, but a navigable reference system aligned with how people actually approach a machine.

“Every Page is Page One”

Baker’s book was published in 2013, long before the current wave of generative AI. Its central observation was simple: readers do not always begin with the first page.

They search, follow links, scan QR codes, open topics from support cases, or arrive through a fault code or component number. The page they reach must work as a first page because, for that reader and that situation, it is the first page.

Wikipedia offers a familiar comparison. A reader can arrive at an article without knowing the wider structure of the encyclopedia. The article provides enough context to make sense on its own, while links lead to definitions, related concepts, detail, and neighbouring subjects. Technical documentation has stricter applicability and safety boundaries, but the entry-point logic is comparable.

For Baker, a useful topic has a specific purpose, establishes its context, follows a recognizable type, stays at an appropriate level, addresses a defined reader, and links richly to related information.

For me, the significance went beyond topic-based writing. The book suggested a different shape for the documentation system itself: a complete manual that does not force every reader through the same sequence.

Why the PDF could only approximate the idea

At the time, industrial manuals were still shaped by pagination, page references, translation packages, print requirements, approval cycles, and the expectation that a manual should look like a complete book.

PDF was an efficient delivery format, but it carried the habits of print with it. Bookmarks, hyperlinks, clickable diagrams, overview maps, and carefully designed page types could have made a PDF more navigable. Interactive PDFs could now approximate parts of the vision; but again, hardly any are more than digital versions of the printed originals.

PDFs could not provide the full flexibility of an online information space. The content would still be tied to a file, a release, and a revision. More importantly, the production environment lacked many of the conditions required for a genuinely non-linear manual:

• reusable modular content;
• stable identifiers for topics, components, configurations, and documents;
• defined page types;
• applicability and audience metadata;
• meaningful relationships between topics;
• a controlled source of truth;
• change impact analysis; and
• a publishing environment capable of generating several views from one content base.

The problem was not that the idea was unrealistic. The problem was that documentation workflows were optimised for producing one approved document, not for maintaining a navigable information space.

What EPPO means in practice

The EPPO Projects on AI TechDoc Knowledge use Baker’s concept as the basis for fictional online documentation projects. EPPO stands for Every Page is Page One.

An EPPO topic is not simply a short fragment cut out of a larger manual. It is a deliberately designed information object with:

• a specific purpose;
• a recognizable page type;
• a defined audience and context;
• explicit applicability;
• meaningful relationships to neighbouring topics; and
• a clear next step for the reader.

A task topic may state which machine and configuration it applies to, which prerequisites are required, which operating mode is assumed, what action is permitted, what result is expected, and what to do if the result does not occur.

A troubleshooting topic follows another pattern. A component topic follows another. A conceptual landing page provides orientation across the information space. The consistency of the page types gives the reader stability; the links between them provide depth.

EPPO therefore does not mean a collection of disconnected short pages. Each page must be usable on arrival, but it must also explain its relationships to other information:

• Read this before starting the task.
• This topic applies to the current machine state.
• Use this procedure for the previous configuration.
• This page explains the protective measure behind the instruction.
• Follow this recovery sequence after the fault.

My first example: an integrated machinery line

In my view, the EPPO model becomes especially important when several machines are integrated into one production line. The line is not adequately documented by placing supplier manuals next to one another. It has its own operating logic, interfaces, transfer points, control states, recovery procedures, and safety boundaries.

The AI TechDoc EPPO project provides a snapshot of how documentation could be different. It uses a fictional packaging line with six machines from four suppliers: infeed and accumulation conveyor, filler and capper, labeller, inspection and reject station, case packer, and palletiser with stretch wrapper.

A reader standing at the inspection station after a line interruption may need to know:

• what condition is visible;
• which machines are affected;
• what the line controller indicates;
• whether isolation is required;
• which reset conditions must be verified;
• which supplier procedure applies; and
• what action is permitted after the safe state has been confirmed.

A linear collection of manuals does not naturally answer that line-level question. An EPPO information space can provide a line-level page and connect it to the relevant machine-level procedures while preserving the source and responsibility boundaries.

The same principle applies to conceptual landing pages. A documentation landing page does not sell a product. It helps readers understand where they are, which configurations and roles are relevant, and which route to follow next.

What changes in the age of AI

AI makes the EPPO idea more visible and much easier to be implemented, but it does not remove the need for structure. It increases the need for well-designed structure.

An AI assistant can help a reader find a page, summarise a relationship, or guide a route that begins with a fault code or a natural-language question. To do that responsibly, it needs to know:

• which topic answers the question;
• which configuration the topic applies to;
• which prerequisites are assumed;
• which source is authoritative;
• which related topics define the boundaries; and
• when to ask for clarification or escalate.

Without page types, applicability metadata, ownership, source relationships, and change history, an AI system may produce a fluent answer that combines information from incompatible contexts.

AI does not replace Every Page is Page One. It makes the quality of the pages, relationships, and boundaries more consequential.

The idea was early and not unrealistic

Mark Baker’s book spoke to me because it described documentation from the reader’s and thus the user’s point of view. People do not experience a machine as a chapter sequence. They experience a task, a condition, a question, an interruption, a component, a risk, or a decision.

I could see that long before the tools existed to support it properly. I could see it even in the constrained form of a PDF: an orientation page as a map, tabs as alternative entry routes, and controlled topics connected through meaningful links.

The idea did not become a working documentation system at the time because the production models were not ready for it. Today, online documentation, integrated machinery, configuration-dependent information, and AI-assisted access have made the same design problem impossible to ignore.

EPPO offers a practical direction: build the documentation as an information space, give every page a purpose and a context, preserve one controlled source, and allow several valid routes through it.

The idea was not too idealistic. It was early.

But now it's high time that his ideas get implemented because the AI era makes it workable and manageable. Here's a snapshot of what I call an EPPO project for an integrated machinery line: Packaging machine.