Written December 2025. Updated August 2026.
If you run a news or media site and your traffic fell off a cliff, the first meeting is about content. That meeting is usually the wrong one, and it costs weeks.
Nothing in the newsroom changed overnight. What changed is either how Google evaluates your site, or something structural about the site that an evaluation exposed. Both look identical from the dashboard, and telling them apart is most of the work.
The December 2025 core update is the clearest case I have worked on. It rolled out on 11 December and took eighteen days to complete. Across the industry, publishers reported Discover declines between 50 and nearly 100 percent, and some saw impressions disappear entirely within 48 hours. Most of what follows comes from that period, but the pattern is not specific to that update. It repeated in February, March, and again in May 2026.
What follows is the diagnostic lens I use.
1. Discover dependency is the exposure, not the cause
Before anything technical, look at where the traffic was coming from.
Discover became the dominant channel for news publishers without most of them deciding it should. Analysis by NewzDash across more than 400 publishers found Web Search fell from 51 percent of Google traffic to news sites in 2023 to 27 percent by Q4 2025, with Discover accounting for around 67 percent.
On one publisher I worked with, it was worse than that. Discover carried 82 percent of all clicks across sixteen months. When the surface withdrew, it did not taper. It recorded days of exactly zero. Web Search over the same period never reached zero at any point.
That gap is the whole thing. Discover is demand Google chooses to send you. Search is demand that already exists and comes looking for you. A business built on the first has no floor, and December 2025 is when a lot of publishers found that out at once.
This is exposure rather than cause. Discover did not break the site. Discover is where the breakage showed first, because it is the surface with the least patience and the fewest guarantees.
Practically, that means two things. Read your recovery on Search, not Discover, because Discover comes back in volatile steps that flatter the chart. And treat Discover volume as revenue rather than as evidence of health.
2. Editorial intent breaks at the template level, and sometimes below it
Commercial routing embedded in article templates creates ambiguity at scale. A weight-loss calculator next to a headline about university certifications makes a page harder to classify. Each element is legitimate. Repeated across an archive, the combination is not.
This part is not controversial. Glenn Gabe's analysis of the December update flagged UX and ad load directly, and warned one publisher about their user experience the day before the rollout began. Core updates evaluate sites holistically, and what a template communicates about a page's purpose is part of that.
That was my read in December, and it was pointed at the visible layer. It turned out to be incomplete.
Commercial routing does not have to be visible to cause a problem. On one site I later took apart, the CMS applied nofollow to every external link by default, and eight commercial partner domains were written into an exclude list. Links to those eight shipped with no rel attribute at all, across an estimated 330,000 pages.
Nothing in the markup looked wrong. A crawler sees followed outbound links and reads them as normal, because on most sites they are. It has no way of knowing that on this site a followed external link was impossible by default, and that every single one was a deliberate exception written into a configuration field.
The links were in the HTML the whole time. The policy that made them a scheme was not. No audit tool surfaced it, because audit tools read rendered output, not the rules that produced it.
I wrote that one up in full, with the data and the timeline: Reversing a 90% traffic collapse at a 650,000-page news publisher.
If your audits keep coming back clean while your traffic keeps falling, that gap is the first place I would look.
3. Commodity content is a positioning problem
Google now extracts and summarises commodity information directly. Dates, definitions, standard figures, event details. If your value is being the place that has those, you are supplying an answer rather than receiving a visit.
The publishers holding up concentrate on what cannot be summarised without losing something. Original reporting, first-hand data, informed interpretation, a voice a reader returns for.
This is not a content quality problem. It is a question of what you are for.
4. Crawl efficiency is not a ranking factor, and still matters for news
This section needs a caveat before it starts, because it is the one most often argued badly.
Google states plainly that crawling is not a ranking signal, and that improving crawl rate will not by itself improve positions. John Mueller has said crawl budget is overrated and that most sites never need to think about it. If you run a site of a few thousand pages, skip this section.
For a large publisher, the relevant part is narrower and it is not about rank. It is about latency. Google's own guidance says crawl budget becomes relevant for sites publishing rapidly at scale, and Discover and News are time-sensitive surfaces where arriving late is the same as not arriving.
On the publisher environments I have worked on, refresh crawling routinely sits above 95 percent of all activity, with image crawling taking close to a third of the total. When breaking news publishes, the crawler is busy maintaining the archive.
There is a common claim that pruning low-value pages redistributes crawl budget, and a common counter-claim that it does nothing, because pages nobody links to were barely being crawled anyway. I had the logs to check.
On one publisher, 65 percent of verified Googlebot requests in June 2026 went to paths outside the core content sections. By August that was 44 percent, with core sections rising from 35 to 56 percent of all crawl activity. A crawl trap was closed and a few thousand low-value URLs were withdrawn. Nothing was added.
So redistribution is real, at least at that scale and with a genuine trap in play. What I will not claim is that the redistribution caused the recovery. Those are two different statements and only the first one is measured.
5. AMP became overhead
Canonical tags work. They tell Google which URL to prefer. What they do not do is make the other URL free.
At publisher scale, index coverage fills with URLs marked alternate with proper canonical, most from legacy AMP paths. Every article exists twice. Google crawls both, renders both, and reaches the same conclusion it would have reached from one.
AMP did not stop working. It stopped being worth what it costs.
6. Clicks stopped being the primary measure
AI-assisted discovery does much of the research before anyone reaches your site. Users arrive having already gathered context and compared options.
Clicks go down. Intent goes up.
If your reporting is built to maximise sessions, this reads as failure. If it measures engagement depth and conversion, it reads as what it is. Fewer visitors, further along.
7. Text-only is a single point of failure
Images, video, and audio are discovery surfaces now, not decoration. A publishing operation that produces only written articles has all of its distribution risk in one channel.
Spreading across formats is not a content strategy. It is redundancy.
8. Technical SEO moved from access to interpretation
For most modern platforms, crawl access is not the constraint. The constraint is whether machines can tell what your content is.
Structured data, consistent entity naming, predictable templates. These lower the cost of understanding you. The goal is not technical perfection. It is clarity at scale.
What recovery looks like
I have been through this more than once, including Periodista.gr.
Visibility did not come back from new headlines or more output. It came back from removing what was structurally wrong and letting the site be re-evaluated.
One thing worth saying plainly, because the industry is bad at it. Recoveries land inside update windows, which means you can rank your interventions by confidence but you usually cannot prove which one earned it. Anyone showing you a chart and a single cause is either lucky or selling.
The takeaway for leadership
When traffic drops, the first question should not be what to publish next.
Where was the traffic actually coming from, and what happens if that surface withdraws. What does a template tell a machine about a page's purpose. What is the site's value made of, original work or commodity information. And what is the CMS doing that nobody has looked at, because it was configured years ago by someone who has left.
These are system properties. They decide whether your content gets the chance to compete at all.
Update, August 2026. On one of the publisher environments behind this analysis, the eventual cause was narrower than anything I suspected when I first wrote this. A sitewide commercial-link pattern lived in CMS configuration rather than in the HTML, which is why no crawl and no audit tool surfaced it. I have left the original reasoning in place rather than editing it to match what I found later. The full forensic write-up, with the data and the timeline, is here.
Disclaimer: this analysis is based on patterns observed across publisher environments I have worked on, crawl diagnostics, and post-update behaviour. It is a systems-level interpretation, not inside knowledge of Google's ranking systems. Figures are indicative of those environments rather than universal thresholds. Google's systems change continuously and outcomes vary by architecture, content model, and delivery behaviour.
