I began this investigation because visibility appeared to be disappearing.
The pages were still there. Many still ranked. The information remained accurate and the domain remained authoritative. Yet traffic no longer followed those signals as predictably as it once had. The measurements I had spent years using to understand visibility were telling me that much of the system still worked, while the business outcomes were telling me that something important had changed.
What I eventually realized was that visibility had not simply declined. Its relationship to the page had changed.
For most of the commercial history of the Web, visibility had a destination. Search helped someone discover a document, and the document was where the information task was completed. Publishers built their businesses around that journey, and the measurements we used reflected it.
Answer engines introduce another possibility. The machine can encounter the publisher’s information and use it before the person ever reaches the source. It can retrieve facts, compare them, interpret relationships, apply context, and synthesize a response.
The page remains important. Traffic remains important. A publisher still benefits from rankings, citations, referrals, brand recognition, and visits.
But those outcomes no longer describe every way in which published information can become visible.
19.1 Visibility Used to Have a Destination
The traditional search relationship was relatively easy to understand.
A publisher created an information product. Search engines discovered and ranked it. A person encountered the result, decided whether it appeared useful, and visited the page.
The page was the destination.
That arrangement gave publishers a practical way to think about visibility. Rankings told us where a resource appeared. Impressions told us whether people had an opportunity to see it. Click-through rates told us how often that visibility became a visit. Analytics told us what happened after the visitor arrived.
The measurements were imperfect, but they described a recognizable journey from discovery to destination.
That journey also explained why traffic became such a useful proxy for visibility. The machine’s principal role was to connect a person with the publisher’s information product. If the publisher became more visible in search and everything else remained reasonably stable, more people generally had an opportunity to reach the publication.
The relationship was never perfectly linear. Rankings changed. Search-result features absorbed clicks. User behavior varied. Competition mattered. But the basic division of labor remained intact.
The publisher produced the information product.
The machine helped the person find it.
The person consumed it.
Answer engines alter the middle of that relationship.
Consider someone trying to choose a camera. The person may ask for full-frame cameras under a particular weight that include in-body stabilization and support a particular video mode. No publisher may have created a page answering that exact question.
The machine does not necessarily need one.
It can retrieve specifications from several resources, identify cameras satisfying the conditions, compare them, and construct a response.
The publisher’s page may still participate. Its information may be authoritative. Its camera specifications may provide several of the facts necessary to produce the answer.
But the page is no longer necessarily where the information task ends.
That distinction helps explain the experience that opened this book. The pages could still rank. The content could remain accurate. The domain could remain trusted. Yet traffic patterns could diverge from those traditional signals because some of the information work that once occurred after the visit could now occur before it.
Nothing about that makes the destination irrelevant.
It means the destination is no longer the only place where the publisher’s information can participate.
19.2 Knowledge Can Be Visible Without the Page Being Visited
Once machines begin consuming information rather than merely locating documents, the word visibility becomes more complicated.
A page can be visible when it appears in search results.
A publisher can be visible when its brand appears in a generated response.
A source can be visible when the system provides a citation.
But the publisher’s knowledge can also participate in the response itself.
Suppose a retailer publishes authoritative information about the Apex X5 Mark II. The camera weighs 658 grams. It uses the Apex Z lens mount. It includes in-body stabilization. It supports particular video modes.
A person asks an answer engine for cameras meeting a combination of those requirements.
The machine may use the camera’s weight to determine whether it qualifies. It may use the stabilization specification as another condition and the video capability as another. It may compare the resulting camera with products described by other publishers.
The retailer did not write the answer.
Its knowledge nevertheless contributed to it.
We observed the same possibility during the Medicare Options work. County resources containing maintained Medicare knowledge appeared in generated answers to factual questions for which we had not independently written corresponding answers. We could observe the answers and public citations, although the retrieval and synthesis process between publication and response remained outside our view.
This does not make machine use economically equivalent to a visit.
That distinction matters.
An advertising-supported publisher still needs traffic. A retailer benefits when someone visits and purchases. A service business needs inquiries and customers. A subscription publisher needs subscribers. Publishers have legitimate reasons to care whether their contribution produces attribution, recognition, referral traffic, or revenue.
Knowledge use and commercial value are different questions.
The change is that a visit can no longer serve as a complete proxy for whether the publisher’s information participated in the information environment.
Nor can a citation.
A citation tells us that a source was surfaced. It does not necessarily tell us how much of the answer depended on that source. Conversely, information from a publisher may influence a synthesis without producing the kind of visible attribution the publisher would prefer.
Participation and attribution therefore need to be understood separately.
From the publisher’s perspective, attribution is highly desirable. It can establish authority, create recognition, and give the person a path to the source.
But before knowledge can be attributed, cited, compared, summarized, or otherwise used, it first has to survive as usable knowledge.
That takes us beneath conventional measures of visibility.
19.3 Visibility Has Become a Chain
The publishing process now contains several points at which visibility can succeed or fail.
A resource first has to be published in a form available to the public Web. An external system then has to discover it. The relevant information has to be retrievable. Once retrieved, it has to be interpreted correctly. In conditional domains, it may also have to be resolved against the context of the question before it can be used safely. Finally, the publisher may or may not receive attribution for the contribution.
The sequence can be represented simply:
Published
↓
Discoverable
↓
Retrievable
↓
Interpretable
↓
Resolvable
↓
Usable
↓
AttributableThese stages describe different problems.
A publisher may possess excellent knowledge that an external system never discovers. Nothing about KNOL solves that problem by itself. Discovery still depends on many familiar considerations: accessibility, indexing, distribution, authority, linking, relevance, and the other mechanisms through which public resources become known.
Discovery can also succeed while retrieval fails.
The machine may reach a long article containing the information it needs, yet the relevant facts may be difficult to isolate or compare. The same property may be expressed inconsistently across several resources. A value may depend on context established far away from the passage in which it appears.
Retrieval can succeed while interpretation fails.
A machine may recover an accurate performance rating but associate it with the Medicare plan rather than the contract to which it actually belongs. It may find the correct camera specification but fail to distinguish the Mark II from the product family.
And interpretation can succeed while resolution remains incomplete.
The Plan-ID work gave us a particularly clear example. A system could retrieve several authentic segment premiums, understand their relationship to the correct parent plan, and still lack the geography required to determine which premium applied to the person asking the question.
In that case, discovery succeeded.
Retrieval succeeded.
Much of the interpretation succeeded.
The answer still could not safely be resolved.
Successful use introduces another question: attribution. The system may use the information accurately and still fail to identify the publisher in a useful way.
A publisher experiencing any of these failures might describe the outcome with the same phrase:
We aren’t visible.
But the underlying problems are different.
That distinction matters because publishers control different portions of the chain to different degrees.
We cannot inspect or dictate the proprietary retrieval systems, indexes, ranking models, or reasoning processes operated by external companies. We cannot require an answer engine to retrieve a particular assertion, select our resource, or provide a citation.
What we can control is what exists on our side of that boundary: what we publish and how faithfully it represents what we know. That has been the scope of this investigation from the beginning.
A publisher can establish clear identity. It can preserve context that determines meaning. It can make important relationships explicit. It can distinguish source assertions from publisher derivations. It can preserve provenance. It can represent the conditions under which a fact can safely be applied.
Those decisions do not guarantee that an external machine will use the knowledge.
They reduce the amount of meaning the machine must reconstruct if it does.
That is the portion of the visibility chain the publisher can engineer directly.
19.4 Visibility Starts With Being Worth Resolving
When distribution changes, publishers naturally look for techniques that will restore or improve visibility.
That happened with search engines, social platforms, featured snippets, recommendation systems, and now generative answer engines. Each new intermediary produces new questions about formats, rankings, citations, retrieval, and optimization.
Those questions are legitimate.
The architecture developed in this book does not replace them.
A Knowledge Object that nobody can discover has little public utility. Publishers still need distribution. Search remains important. Authority remains important. Links and citations matter. Brand recognition matters. A useful website still needs to be technically accessible and worth visiting.
KNOL addresses a different part of the problem.
Suppose a machine does reach the publisher.
What does it find?
Imagine a company with thousands of pages containing substantial domain expertise. Its information may be accurate and its writers may be excellent. Yet the same factual assertion may appear differently across dozens of resources. Product variants may be difficult to distinguish. Important context may be carried only by surrounding prose. Derived statistics may be attributed loosely to the source data from which the publisher calculated them. Relationships that everyone inside the organization understands may never be represented publicly.
The company has content.
It has knowledge too.
But the two are not necessarily organized in the same way.
That distinction becomes more consequential when a machine attempts to use individual pieces of the publication outside the reading experience for which the pages were originally designed.
The opposite problem is possible as well.
An organization may maintain an excellent internal information model. Its databases may distinguish entities precisely, preserve authoritative sources, maintain relationships, and enforce business rules. If none of that knowledge is represented usefully on the public Web, an external system has little opportunity to benefit from it.
Visibility therefore requires more than one kind of work.
The publisher still has to become discoverable.
But once discovered, it also has to be useful as a source.
This is where the question How do I improve the visibility of my business? becomes more interesting than it first appeared.
For years, much of the answer concerned getting the publisher’s pages in front of more people. That remains part of the answer.
The other part concerns what the publisher makes available once a machine arrives.
Can the system distinguish what the organization is actually talking about? Can it identify comparable facts consistently? Can it preserve the context that determines where and when those facts apply? Can it distinguish a source assertion from a publisher’s calculation or interpretation? Can it follow meaningful relationships? When several facts are all valid, is enough knowledge available to determine which one applies?
These are not ranking questions.
They are questions about whether the publisher has created something another system can use responsibly.
That is what I mean by being worth resolving.
The phrase does not imply that every piece of published knowledge requires a formal resolution process. Most facts are considerably simpler than the segmented Medicare premiums that led us to the concept.
It describes a broader publishing quality.
The knowledge has enough identity, context, structure, provenance, and semantic precision that another consumer has a reasonable opportunity to determine what it means and where it applies.
That is also why knowledge engineering and human publishing do not need to compete.
A good camera page should still be a good camera page. A Medicare resource should still help a beneficiary understand Medicare. A university should still explain its programs. A manufacturer should still help customers understand its products.
The work developed in this book occurs underneath those information experiences.
The publisher represents what it knows more deliberately so that the same authoritative knowledge can remain useful when encountered through more than one path.
I began this investigation with visibility, traffic, and revenue because those were the outcomes I could observe most easily.
The investigation eventually moved beneath them.
Before published knowledge can earn a citation, generate a visit, influence an answer, support a decision, or contribute to commercial value, another consumer first has to be able to make useful sense of it.
That is the additional dimension answer engines introduce.
Visibility now includes the visibility of the knowledge itself.