What Shepard’s data describes is a shift from pageview logic to knowledge-asset logic.
Google is not rewarding volume the way many publishers and knowledge brands still assume. The pages-per-month model that defined the SEO era is contracting, and the publishers who keep funding it at previous levels are funding a contracting asset.
Cyrus Shepard reviewed more than 400 sites that gained or lost organic visibility through the last year of search updates. The pattern in his dataset is specific. The top differentiator was whether a site offered a real product or service. Behind that came task completion, proprietary assets, tight topical focus, and brand demand. Most of the losers were news, informational, and affiliate sites — sites whose entire model assumed that publishing more pages on more topics would keep the traffic coming.
What Shepard’s data describes is a shift from pageview logic to knowledge-asset logic. Pageview logic measures success in impressions, sessions, and yield per article. Knowledge-asset logic measures success in what the publisher controls, what the publisher can expose on its own terms, and what compounds over time. The first model still pays the bills at most publishers. The second one is where the rewards now sit.
This is not a forecast. It is already in the data Shepard collected.
Why Google’s behavior is changing
Google answers more generic informational queries before the user reaches the publisher at all. Search features and AI summaries handle more of the job each quarter, and the old traffic model gets squeezed each time. If the basic answer sits on the results page, generic page production weakens as a strategy.
When a model summarizes the top three results into a competent answer, the publisher who wrote one of those results gets credited as source material rather than the destination. That is a different commercial position. Source material is paid in attribution. Attribution is real, but it is harder to monetize than a session, and most publisher P&Ls are not built to absorb the swap.
What holds up against this is content the user has to come to the site to use — a tracker, a calculator, a structured comparison, a directory the model cannot reproduce from a snippet. Owned utility outperforms summarized information.
The four signals, briefly
Task completion. A reader who finishes the article and still has to open another tab has been handed off to whoever owns the next step. The sites Google is rewarding close that handoff inside the session. Compare prices, check a schedule, run the calculation, file the form. The article can be the entry point. Something else has to be the destination.
Proprietary assets. The winners own something others cannot easily recreate. A dataset, a service layer, specialized inventory, a reference system that compounds. For most publishers, the strongest candidate is sitting in the CMS already — the archive, if it were structured. More on that in the next post.
Topical focus. Tight scope outperforms broad coverage. Google rewards sites that go deep on a defined territory and are recognizable as the place to land for that territory. This is most of why niche publishers held up better than general-interest publishers in Shepard’s dataset.
Brand demand. When Google sends less traffic, brand is what brings readers back. Direct loads, newsletter opens, branded search volume, social returns. In a zero-click environment, those numbers become the last-mile infrastructure of the business — the part that does not depend on platform decisions made in someone else’s product meeting.
These four signals are not independent. They share a substrate, which is the topic of the next post. For now, the point is that they describe a different operating model from the one most publishers are still funding.
What this looks like in three publisher categories
Local media owns history, proximity, and trust that national platforms cannot buy. The advantage sits trapped inside archived articles, PDFs, and unstandardized metadata. School board coverage becomes an election explainer layer. Local business reporting becomes a reference system. Community calendars become event products.
A municipal archive is not a weekend editorial experiment, and it is not a prompt. If a city hall reporter has tagged stories inconsistently for a decade — the fourteen different ways a beat editor has entered “City Council” since 2014 is a real example — a model pointed at that archive will reproduce the inconsistency and hallucinate connective tissue that was never there. A working searchable meeting utility requires a schema audit, entity resolution, interface decisions, and retrieval tuned to verified source text. That is a capital project, and it should be staffed and funded as one.
Sports media arrives with recurring demand, known entities, scheduled moments, and audience habits built around leagues and teams that do not change names every quarter. Few categories get that gift. The tradeoff is that the content layer commoditized faster than almost anywhere else. Final scores and basic player facts are everywhere the user looks before they reach a publisher site.
The defensibility is in what surrounds the coverage. A team beat that produces 800 articles a year is a content asset. The same beat structured into player histories, transaction timelines, and stat-rich season hubs becomes a reusable inventory layer that sponsors will pay against and operators will license. The reporting did not change. The packaging did.
Niche publishers already possess one of the strongest signals in Shepard’s analysis — tight topical focus — and most are under-using it. An entertainment publisher with eight years of artist interviews has the inputs for artist hubs and timeline pages that platforms cannot replicate. A home, finance, or health publisher has the inputs for planners, calculators, and maintenance schedules readers use on a recurring schedule. The expression varies by vertical. The conversion is the same.
Key questions
What is Google rewarding now?
Google is rewarding publishers who own something readers cannot get from a search result page or an AI summary. Cyrus Shepard’s analysis of more than 400 sites that gained or lost organic visibility through the last year of search updates identifies four signals that separate the winners from the losers: task completion, proprietary assets, topical focus, and brand demand. What the data describes is a shift from pageview logic to knowledge-asset logic — measuring success in what the publisher controls and what compounds over time rather than in impressions and sessions.
What are the four signals?
Task completion (close the handoff inside the session — compare prices, check a schedule, run the calculation, file the form). Proprietary assets (own something others cannot easily recreate — a dataset, a service layer, a reference system; for most publishers, the strongest candidate is the archive, if it were structured). Topical focus (tight scope outperforms broad coverage; this is most of why niche publishers held up better than general-interest publishers in Shepard’s dataset). Brand demand (direct loads, newsletter opens, branded search volume, social returns — the last-mile infrastructure of the business in a zero-click environment). These four signals share a substrate — a structured corpus that supports each one.
How is this different from the SEO-era model?
The pages-per-month model that defined the SEO era is contracting. Google now answers more generic informational queries before the reader reaches the publisher at all — search features and AI summaries handle more of the job each quarter, and when a model summarizes the top three results into a competent answer, the publisher who wrote one of those results gets credited as source material rather than the destination. Source material is paid in attribution. Attribution is real, but it is harder to monetize than a session, and most publisher P&Ls are not built to absorb the swap. Owned utility — a tracker, a calculator, a structured comparison, a directory the model cannot reproduce from a snippet — outperforms summarized information.
What does this mean for the publisher P&L?
Funding the new surfaces means reallocating from the old ones. Engineering and data hours come out of headcount currently pointed at generic page production, or out of syndication and affiliate spend that is already losing yield. Funding both models at full strength is not on the table for most publishers and knowledge brands. The question is which model is contracting and which one is growing, and the answer in Shepard’s data is not ambiguous. Two years from now, the publishers still funding volume will have spent the same money and own less of what matters — and the ones who funded the work to structure the corpus will own the asset.
The P&L conversation that has to happen
The uncomfortable part for most publisher P&Ls is that funding the new surfaces means reallocating from the old ones. Engineering and data hours come out of headcount currently pointed at generic page production, or out of syndication and affiliate spend that is already losing yield. Treating the shift as an incremental budget ask is how these initiatives stall in committee and never ship.
That conversation is hard, and it is the actual decision in front of leadership. Funding both models at full strength is not on the table for most publishers and knowledge brands. The question is which model is contracting and which one is growing, and the answer in Shepard’s data is not ambiguous.
Two years from now, the publishers still funding volume will have spent the same money and own less of what matters. The ones who funded the work to structure the corpus will own the asset. Most of the competition is still trying to out-publish the problem.
What the structured corpus actually is, why it sits underneath the four signals — task completion, proprietary assets, topical focus, and brand demand — and how publishers and knowledge brands convert it into revenue is the subject of the next post.
