What Actually Happened
On July 9, 2026, Google updated its core-updates documentation to admit what practitioners had been measuring for months: smaller core updates now roll continuously, without any public announcement, in the gaps between the few large named updates it still posts to the Search Status Dashboard.
Google's own guidance now states that it continually makes updates to its search algorithms, including smaller core updates that it does not announce because they are not widely noticeable, per Google Search Central's core updates documentation. The named updates, meanwhile, are getting rarer and faster. The May 2026 core update finished rolling out in roughly twelve days, one of the fastest on record, according to Search Engine Land. Everything in between those named events is now unlabeled, continuous change.
Then layer the AI answer surface on top. AI Overviews now appear on close to half of tracked queries, they cut click-through on the top organic result by roughly a third, and only a minority of the pages they cite come from the top ten blue links at all, per an AI Overviews analysis from seo-kreativ. So a growing share of visibility is being decided by a machine that reranks every day and cites sources by a logic that ignores your ranking position.
Read together, the message is blunt for anyone who publishes content to be found. There is no stable leaderboard to climb and no scheduled exam to study for. The ground moves every day, quietly, which means any single thing you point at it, a landing page, a blog post, a hero video, is standing still on top of a system that will not stop moving.
Why A One-Off Video Cannot Hold Anymore
A one-off video cannot hold its position anymore because there is no longer a fixed position to hold. Most marketing teams still buy video and content on a punctuated model: commission the brand film, publish the pillar page, run the campaign, then wait. When traffic drops, name an update as the culprit and fund a rescue project. That model assumed updates were discrete events with a clean before and after. Google just told you they are not.
When the algorithm never stops moving, a recovery sprint is aimed at a target that has already left the range by the time the work ships. You cannot do a one-time fix for a continuous condition. This is exactly where the one-off content purchase falls apart, and it is a big part of why a single expensive brand video so often underperforms its invoice. The video is not bad. It is alone.
A single great video, a single pillar page, a single campaign: each one is a bet that a fixed asset will hold its rank in a fixed system. Under always-on updates there is no fixed system to hold rank in. The asset does not decay because it got worse. It decays because everything around it kept moving and it did not, and a video you shot once and never built on has nothing feeding it.
The teams that stay visible are not the ones with the best single asset. They are the ones publishing original, experience-heavy content on a cadence, real customers, real demos, a founder on camera, feeding the machine fresh first-party signals faster than it reranks. That is not a project with an end date. That is a content system, and content systems get built as retainers, not as one-time invoices. It is the whole reason we build video as a growing library rather than a one-time launch.
The Data We Actually See
Start with our own book, and read it as operator observation, not a study. We run structured video retainers for B2B SaaS teams, which means we watch the same clients build a video library continuously for years, and we watch project clients go quiet after a single shipment. The contrast is hard to miss.
First number. In our book, the median project-only client publishes fewer than two owned video assets in a year. The median retainer client ships three to four a month. Under an algorithm that reranks continuously and rewards fresh first-party signals, that cadence gap is not a nuance. It is close to the entire difference between feeding the system and starving it.
Second number. Across our retainer client audits through the 2026 update cycle, the accounts building a library on a continuous cadence held or grew their branded and organic visibility straight through the May update and the unannounced churn that followed. In the project-only comparison set we see the opposite pattern: visibility built by a one-time video or content push tends to fade within about two quarters of the last thing shipped, right on schedule with the continuous reranking Google just confirmed. We cannot run a clean controlled experiment on this, and we are not claiming one. It is simply what the book shows.
Now set that against the public record. Google's core-update documentation now confirms continuous, unannounced updates as standard practice (Google Search Central). AI Overviews appear on nearly half of queries and depress top-result click-through by roughly a third, while pulling citations from well outside the top ten results (seo-kreativ). And the surface increasingly favors named, trusted sources: in May 2026 Google began highlighting links from publications a user already subscribes to inside AI Overviews (Nieman Lab), a trust-and-relationship signal that a one-off asset from an unknown author has no way to earn. Every one of those facts rewards the same thing: consistent, credentialed, first-party publishing over the one-and-done drop.
The Strongest Case Against This
Here is the honest counter, because there is a real one. A retainer costs more than a project, and "just publish video continuously forever" is exactly what a studio that sells continuous production would tell you. If you are a small team, a genuinely excellent pillar asset really can rank and hold for a long time, and plenty of businesses have built durable traffic on a handful of great evergreen pages without a monthly content machine. Cadence for its own sake, publishing thin content on a schedule, is worse than publishing nothing. All of that is true.
So the argument is not "shoot more video." Thin, fast content is precisely what the 2026 updates are built to bury, and a pile of forgettable clips will not save you. The argument is narrower: the durable unit is a system that produces original, experience-heavy content repeatedly, not the raw volume. A single outstanding video can hold, yes, but its odds get worse every quarter the ground shifts under it, and betting your entire visibility on one asset holding forever is a bet against a system Google just told you never stops. The retainer is not the point. The continuity and the originality are the point. If you can get those another way, get them another way. Most teams cannot sustain it off the side of a desk, which is why the system usually has to be somebody's actual job.
What To Do Monday
First, retire the recovery-sprint line item. Stop budgeting for the post-drop rescue project and move that money into a standing production cadence. You are funding a metabolism now, not an emergency room visit.
Second, audit your library honestly. Count the original, first-party assets you published in the last ninety days: real customers on camera, real demos, real founder points of view. If the number is close to zero, no technical fix will save your visibility, because you are not feeding the machine anything only you have.
Third, lead with experience, because it is the one signal that cannot be copied. AI Overviews and the trust regime reward first-hand proof: your customer, your data, your shoot. Original first-party video is the strongest experience signal you can produce, and it is the one thing an AI answer cannot source from anyone but you.
Fourth, structure every asset to be cited, not just watched or read. Answer the implied question in the first two sentences of each section and each video description, keep them self-contained, and put your name and credentials on the work. Under a trust-first, citation-first surface, a clean, attributable, authored answer is what gets pulled into the box.
Fifth, measure quarters, not updates. Stop trying to attribute every wobble to a named event Google may never name. Watch your branded search, your owned-asset count, and your citation share over rolling quarters. The continuous algorithm rewards the continuous operator.
Do those five things and the next unannounced update becomes weather, not a crisis. You cannot out-sprint an always-on algorithm with one perfect video. You can only out-last it with a library, and a library is a system, not a purchase.