Check this page with an assistantOpens a chat asking it to summarise this article and name the evidence behind each claim.

Claude opens with the prompt on your clipboard: Anthropic does not support prefilled prompts on the web, and we would rather copy it than ship a button that drops it.

What Is Established, and What Is Inference

Established: Google ran a core update from 27 March to 8 April 2026 and another from 21 May to 2 June 2026, an unusually short gap by recent standards. Also established: since November 2025, site reputation abuse has been enforced algorithmically, targeting third-party content hosted on strong domains to borrow their standing. Inference: that these represent a shift toward assessing pages individually rather than inheriting from the domain.

The distinction is worth holding onto, because the inference is being sold as a finding. Google has never published a page-level authority score, and its guidance has for years described core updates as reassessments of content rather than actions against sites. When a consultant explains your traffic loss using a named mechanism that the search engine has never named, you are being given a story rather than a diagnosis.

What survives the scepticism is the observation itself, and it is widely reported: sites that used to see new pages perform respectably by virtue of the domain they were published on have stopped seeing that. Whatever the mechanism, the practical world has changed in a way that makes the same response correct. That response does not depend on the theory being right, which is the test any recommendation here should pass.

The Subsidy Model

Think of a strong domain as a subsidy. Pages that could not have ranked on their own merits ranked anyway, funded by the reputation of everything else on the site. When the subsidy narrows, those pages return to what they would always have earned alone, and the loss looks sudden because the subsidy was invisible while it worked.

This explains the shape of decline that confuses people most: not a uniform drop across a site, but an uneven one where the best pages are untouched and a long tail collapses. If a site had been penalised you would expect the strong pages to suffer too. If a subsidy narrowed, you would expect exactly what people are reporting, with the pages that most depended on it falling furthest.

It also explains why the usual remedies disappoint. Adding internal links to a page that nobody wants does not create demand for it. Refreshing the date on a page that answers a question nobody asks does not make it answer one. Where a page was riding the subsidy, the only durable fixes are giving it a reason to exist or removing it, and most sites have more of the second category than they expect.

Finding the Pages That Were Riding

Export page-level performance for a period before the update and an equal-length period after, matched for seasonality and day of week. Sort by absolute clicks lost, not percentage. Then look for the trait the losers share, because the cluster is the finding and the individual pages are just its members.

  1. Match the periods properly.

    Twenty-eight days before the update start against twenty-eight days after completion, on the same weekdays. Comparing an arbitrary month to another month imports seasonality as if it were an algorithmic effect, which is how most core update analyses go wrong in the first ten minutes.

  2. Rank by clicks lost, not percentage lost.

    A page falling from four clicks to one is a 75% decline and irrelevant. A page falling from eight hundred to five hundred is where your traffic went. Percentage sorting reliably surfaces the least important pages on the site.

  3. Look for the shared trait.

    Same template, same publishing sprint, same author, same thin-content pattern, same year. Sites rarely lose scattered individual pages; they lose cohorts. Naming the cohort turns two hundred fixes into one decision.

  4. Separate demand loss from ranking loss.

    Impressions falling with position steady means fewer people are searching, or the result is being resolved without a click. Position falling with impressions steady is an assessment change. These have different causes and opposite remedies, and conflating them wastes a quarter.

  5. Check whether AI surfaces absorbed it.

    A query answered inside a generated response can lose clicks without losing ranking. The impressions side of that now has a dedicated view, described in the Search Console generative AI report, and the pattern itself is zero-click search rather than a core update effect.

Merge, Fix, or Remove

Every page in the losing cohort gets one of three verdicts. Merge if several pages circle the same question without any of them answering it fully. Fix if the page serves a real query badly. Remove if it serves no query and no visitor. Most sites need more merging than they expect and less removing than they fear.

SignalVerdictWhy
Several pages competing on one queryMergeConsolidated depth beats split coverage, and it resolves cannibalisation at the same time
Real impressions, poor position, thin answerFixDemand is proven; the page is simply not good enough to meet it
Published to fill a calendar, no query behind itRemove or noindexNothing to improve toward; the page had no reader in mind
Template page with almost no unique dataRemove the template, keep the useful instancesThis is the failure mode described in programmatic SEO
Converts, ranks poorly, low trafficKeep and leave aloneSearch performance is not the only reason a page exists

That last row deserves emphasis, because a cleanup driven purely by Search Console will happily delete pages that quietly close business. Cross-reference against conversions before acting on anything, and preserve links when merging so accumulated equity survives the consolidation. The mechanics for doing that without losing signal are in the content consolidation playbook, and the content refresh checker will sort candidates by decay if you want a starting queue.

The Cadence Change Matters More Than Either Update

Six weeks between core updates instead of three or four months changes the working model. There is no longer a stable period in which to make a change, wait, and read a clean result. Assessment is close to continuous, which makes a quarterly cleanup habit more useful than a reactive recovery project.

The practical shift is from responding to updates to maintaining a site that has less to lose from them. That means auditing the tail on a schedule rather than after a scare, keeping a real record of what you changed and when so that a future decline has a timeline to check against, and resisting the urge to make a dozen simultaneous changes that will be impossible to attribute afterwards.

It also raises the cost of publishing volume without demand behind it. Every page added to the tail is exposure to the next reassessment, and the pages most likely to be reassessed downward are the ones written to a calendar rather than to a question. That is the same conclusion reached from a different direction in whether Google penalises AI content: the problem was never the tool used to write, it was publishing things nobody needed. The recovery sequence itself, if you are in the middle of one, is in recovering from a core update.

Questions People Ask About Page-Level Authority

Does Google use page-level authority?

Google has never published a metric by that name, and you should treat the phrase as practitioner shorthand rather than a documented mechanism. What is documented is that core updates reassess how well individual pages serve their queries, and that the site reputation abuse policy explicitly targets individual sections of otherwise reputable domains. The observable effect people describe as page-level authority is real; the named mechanism is inference.

What happened in the March and May 2026 core updates?

Google ran a core update from 27 March to 8 April 2026, then launched another on 21 May 2026 which completed on 2 June. Roughly six weeks separated the end of one from the start of the next, a noticeably faster cadence than the three to four months that had been typical. Many sites reported uneven impact within a single domain, with some sections losing visibility while others held.

Is a core update a penalty?

No, and the distinction changes what you do about it. Google describes core updates as broad reassessments of how content is evaluated rather than actions against a site, and states there is nothing wrong with pages that decline. A manual action appears in Search Console and names the problem; a core update decline does not, because it is a change in relative assessment rather than a sanction.

How do I find which pages dragged the site down?

Compare page-level clicks and impressions across matched periods before and after the update, then sort by absolute loss rather than percentage. Percentage change flatters pages that had nothing to lose. What usually emerges is a cluster of pages with a shared trait: a template, a topic, an authorship era, or a publishing sprint, and that shared trait is the actual finding.

Should I delete pages that lost traffic?

Rarely as a first move. A page that lost rankings still holds links, history, and sometimes conversions the report does not show you. The usual order is merge related thin pages into one substantial page, improve pages that serve a real query badly, and only remove pages that serve no query and no visitor at all. Deletion is a decision about pages with no reason to exist, not about pages with disappointing numbers.

Primary Sources

SearchHandled Editorial TeamPublished Jul 10, 2026 · Last reviewed Jul 10, 2026. Every factual claim is checked against the linked primary sources; corrections can be submitted through our contact page.