One of the more surprising consequences of this work was that knowledge engineering did not lead us toward publishing more.
It led us toward publishing less.
That was not the objective when the work began. Like many publishers, we had spent years responding to information demand by creating content. New questions produced new articles. Search opportunities produced new pages. Slightly different formulations of familiar subjects justified additional explanations.
The result could look impressive in a content inventory.
It could also become extraordinarily expensive to maintain.
The twenty-seven Medicare dental articles discussed in the previous chapter were not unusual because someone had made twenty-seven bad editorial decisions. Most of the individual articles had perfectly defensible reasons to exist when they were created.
The problem emerged collectively.
We had repeatedly expressed substantially overlapping knowledge as independently maintained content.
Once we began treating the Knowledge Space beneath those publications as the durable factual asset, some of the pressure to keep producing additional artifacts disappeared.
The same thing happened to the pages themselves. Facts that had been explained repeatedly in prose could sometimes be represented more clearly in a table. Collections could become lists. Definitions could be maintained coherently. Relationships could connect one Knowledge Object to another instead of requiring every page to explain everything it touched.
The pages became simpler.
There was less to write.
And because the underlying knowledge was no longer being independently recreated everywhere it appeared, there was considerably less to maintain.
I know because we had already built the other version.
21.1 Write What Humans Need Written
Knowledge-Space publishing does not begin with a campaign against content.
There are things people need writers to do.
A photographer deciding between two camera systems may want more than specifications. Weight, sensor format, autofocus capability, lens selection, and price can establish important facts. They do not necessarily explain what it is like to carry one system through a day of wildlife photography or why one tradeoff might matter more than another.
That requires experience and judgment.
Write it.
A Medicare beneficiary trying to understand the practical difference between Original Medicare and Medicare Advantage needs more than a collection of plan attributes. The decision involves tradeoffs, terminology, costs, provider access, and personal circumstances that deserve careful explanation.
Explain them.
A person learning a difficult concept may need an example. Someone performing a task may need instructions. A consequential decision may require qualification. Research may deserve interpretation. An argument may require evidence and reasoning.
Those are information products worth writing.
The problem begins when prose is also asked to become the primary storage mechanism for facts the publisher already knows elsewhere.
Consider Medicare dental coverage.
One article might explain whether Original Medicare generally covers routine dental care. Another might discuss Medicare Advantage dental benefits. Another might focus on dentures. Another might discuss implants. Another might explain how to find dental coverage.
Those articles can serve genuinely different human needs.
But they also share knowledge.
If each writer independently explains the same underlying Medicare rules, definitions, exceptions, and distinctions, the publisher has turned shared knowledge into repeated prose.
That creates two kinds of unnecessary work.
The first occurs during publication. Writers repeatedly research and restate facts the organization already possesses.
The second appears later.
When the underlying knowledge changes, someone has to remember where all of those restatements went.
The answer is not to combine every dental question into one enormous page. That merely replaces duplication with an unpleasant human experience.
The better distinction is between what needs to be known once and what deserves to be communicated differently for a particular person or purpose.
The shared factual knowledge can remain coherent underneath the publications.
The explanation can change according to the human need.
That often produces less writing, not because the publisher has decided to communicate less, but because writers are no longer required to keep reconstructing the factual foundation before they can communicate something useful.
21.2 Engineer What the Organization Knows
This changes where a publisher can begin.
Suppose a new information need appears tomorrow.
The familiar response is to ask what should be published.
Should we write an article? Add an FAQ? Create a comparison? Build another location page? Expand an existing guide?
Those may eventually be the right questions.
Knowledge-Space publishing inserts another question first:
What do we already know?
The answer may be scattered across the organization.
Some of it may already appear on webpages. Some may live in databases or product systems. Some may be encoded in application logic. Methodology documents may explain how derived values are produced. Subject-matter experts may understand relationships and exceptions that have never been represented explicitly in the public information architecture.
The immediate task is not to put all of that online.
It is to understand the coherent knowledge relevant to the publishing problem.
What does it describe? Which assertions belong together? What context governs them? Which relationships matter? Where did the information come from? What did the publisher derive? What limits the valid application of the facts?
Only after those questions are reasonably understood does representation become useful.
Some knowledge will belong naturally in a table. Some will become definitions. Some will connect through relationships. Some will require prose. Some semantic distinctions may need to become more explicit. Some internal information will have no reason to become public at all.
This is the practical meaning of modeling before rendering.
It also means the publisher does not need to transform an entire organization before benefiting from the approach.
Start where the maintenance problem already hurts.
A manufacturer might begin with one product family whose specifications and compatibility information are repeatedly duplicated across the site.
A university might begin with one program whose requirements are represented differently across admissions, departmental, and catalog resources.
A healthcare publisher might begin with one body of knowledge that appears across dozens of explanatory articles.
The question is not how quickly the entire website can be converted into KNOLs.
The question is whether one important body of knowledge can stop being recreated every time it is published.
Once that happens, the economic difference becomes visible.
Writers can spend less time recovering facts the organization already knows.
Editors can spend less time reconciling conflicting expressions of those facts.
Developers can stop hard-coding some distinctions independently into every interface that needs them.
Updates can occur closer to the knowledge that actually changed.
The organization still publishes.
It simply stops treating every publication as though it were the first time the organization had encountered the subject.
21.3 Let Synthesis Do Its Job
The reduction in publishing becomes even more significant when we return to the Question Space.
The camera retailer does not need to publish an article titled:
Best Full-Frame Cameras Under $2,500 That Weigh Less Than 700 Grams With IBIS and 4K/120 Video
simply because someone might ask that question.
If the retailer already maintains the necessary camera knowledge consistently, the question can provide the conditions when it arrives.
Change the price ceiling to $2,000 and the publisher does not need another article.
Change the weight limit.
Require a particular lens mount.
Exclude one manufacturer.
Add weather sealing.
These are new questions.
They are not necessarily new knowledge.
Generative synthesis makes that distinction operationally useful. A system capable of retrieving, comparing, filtering, and combining published knowledge can construct information products that did not exist beforehand.
That is precisely what synthesis is good at.
The publisher does not need to compete with that capability by attempting to prewrite every permutation.
Instead, the publisher can test whether its Knowledge Space actually supports the questions it should support.
Ask questions that were never used to design the publication.
Some should retrieve individual assertions. Others should combine several dimensions. Some should follow relationships. Some should deliberately omit context required for resolution.
Then observe the failures.
A failed comparison may reveal inconsistent knowledge.
A wrong subject may expose an identity problem.
An unsupported conclusion may reveal missing provenance.
An overconfident answer may expose a resolution boundary that the publisher failed to preserve.
A legitimate question that cannot be answered at all may reveal genuinely missing knowledge.
The response to those failures is not automatically another page.
Sometimes the Knowledge Space needs repair.
That changes the publishing loop.
Instead of repeatedly moving from question to artifact, the publisher can move through a cycle of modeling, publishing, querying, observing, correcting, and republishing.
The questions become tests of the maintained knowledge.
This is one of the places where publishing less can produce a better information system.
A thousand narrowly targeted answer artifacts can create the appearance of comprehensive coverage while reproducing the same factual weaknesses a thousand times.
A smaller, better-maintained Knowledge Space can support questions the publisher never anticipated while giving future publications a more reliable factual foundation.
The objective is not fewer pages for the sake of having fewer pages.
It is fewer unnecessary copies of the same knowledge.
21.4 Let the Question Arrive Later
For most of the Web’s history, publishers had a strong reason to anticipate information demand.
If someone might ask a question, the information product generally needed to exist before the person searched for it. The publisher researched the demand, created the page, and hoped the search system connected the two.
That model encouraged an enormous amount of useful publishing.
It also encouraged publishers to treat every newly discovered question as evidence that another artifact might need to exist.
Generative systems weaken that constraint.
An answer can now be assembled after the question arrives.
That does not eliminate the publisher from the process. It changes what the publisher can contribute beforehand.
The publisher can maintain the facts.
It can preserve identity.
It can establish relationships.
It can maintain definitions.
It can preserve provenance.
It can represent context and temporal state.
It can make clear where a fact applies and where the available knowledge stops supporting a definitive conclusion.
Then a question can arrive that nobody anticipated.
If the knowledge supports the question, synthesis can operate on it.
If the question exposes genuinely missing knowledge, the publisher has learned something useful about the Knowledge Space.
If it requires explanation, judgment, or human communication, the publisher can create that information product because the need justifies it.
And if it is merely another permutation of facts the publisher already maintains, perhaps nothing new needs to be written at all.
This creates a different relationship between publishing volume and publishing value.
More pages are not automatically more knowledge.
More articles are not automatically more authority.
A publisher can produce less new prose while increasing the number of information needs its maintained knowledge can support.
Maintenance can decline for the same reason. A fact does not need to be rediscovered and independently corrected everywhere it was once rewritten. The publisher can increasingly maintain the knowledge closer to its authoritative state and allow different publications to express it according to their purposes.
That is not a theoretical efficiency for us.
We arrived at it after maintaining the other model.
The expanding content inventory looked like an asset while it was growing. Its true cost became apparent when the same underlying knowledge had to be revised across dozens, hundreds, or thousands of expressions.
Knowledge-Space publishing does not eliminate maintenance.
It moves more of that maintenance upstream, where one correction can have broader utility.
And it does not eliminate writing.
It gives writers permission to spend more of their time on the things worth writing.
That may be one of the most practical consequences of the entire architecture.
The publisher does not have to know the future Question Space. People will combine facts in ways nobody predicted. They will introduce new conditions, traverse unexpected relationships, and ask questions using language no keyword researcher ever recorded.
The publisher does not need to chase every permutation.
It needs to know what it knows, preserve that knowledge well, and recognize when a human information need genuinely deserves something more.
Write what needs to be written.
Engineer what needs to be known.
Let the question arrive later.