How does probeo help with a content refactor?

Traffic was falling and nothing had broken

Their search traffic had been declining for a long time. Nothing had gone down, nothing was erroring, and pages that used to surface had stopped.

That is what the bar moving looks like from the inside. The pages were fine when they were written. The standard they are judged by changed.

What they could not see

• Which pages were losing ground, and which were holding.
• Whether a page answered the question people were actually asking.
• Whether what a page said was still true, and qualified the way it needed to be.
• Where titles, headings and descriptions no longer matched each other or the page.

What probeo showed them

Every page graded the way Google's quality rater guidelines grade it: what the page is for, the question it answers, and whether what it says holds up. Alongside that, the basics that had drifted: titles, H1s and descriptions.

Then the rewrites. Hundreds of pages, each delivered as a rewritten page for their team to review and publish.

What they did with it

They published the rewrites in one push and kept reviewing in batches after, folding what each review found back into the next round.

What happened

The numbers went down before they went up, which is what rewritten pages do while Google recrawls and re-evaluates them. Then the pages started earning clicks again.

Impressions were lower than a year before, and far more of them turned into clicks. More of those clicks came from people who had never searched for the company by name.

Where it stands

The review is still going, page by page. probeo watches each rewritten page as it settles, shows which queries it surfaces for and what Google now thinks its question is, and that feeds the next rewrite.

What merging actually does

Merging is a canonical decision, and Google already makes one whether or not you do.

Where pages are genuinely retired rather than merged, Google's instruction is to return 404 or 410, and it removes such URLs from the index while crawling them less often.1,2,3

• Google clusters pages it considers the same or very similar and picks the one it judges most complete and useful, then uses that page to evaluate content and quality.
• Your declaration is a hint, not a rule. Google may choose a different canonical, and the factors it weighs include HTTPS, redirects, sitemap inclusion and the canonical annotation.
• The methods differ in strength. A redirect is a strong signal that the target should become canonical, a canonical annotation is a strong signal, and sitemap inclusion is weak. They stack.
• Some duplication is normal and is not a spam violation, though it makes performance harder to track.
• Permanent redirects do not cost you: Google states plainly that 301 and other permanent redirects do not cause a loss in PageRank.

How to judge a page without inventing a score

Google publishes the questions it wants you to ask. Does the content provide original information, reporting, research or analysis. Does it avoid simply copying or rewriting other sources. Does the title give a descriptive, helpful summary. Does it provide substantial value compared with other pages in results.4

It also publishes the warning signs. Producing lots of content on many topics hoping some performs. Mainly summarizing what others said. Writing to a word count, which it notes it has no preference for. Changing dates to look fresh. Adding or removing content mainly to seem fresh, which it says will not work.4

The rater guidelines go further and are unusually concrete. Page quality is assessed by first establishing the page's true purpose, then its potential to cause harm, then how well it achieves that purpose. Quality of the main content is judged on the effort, originality and talent or skill that went into it, and for informational or sensitive topics on accuracy and consistency with expert consensus.5

And a single defect is enough. The guidelines state that any one Low attribute justifies a Low rating, including a mildly misleading title or an inadequate level of expertise for the page's purpose.5

Why one page's problem is not only that page's problem

Ranking works at page level, and site-wide signals contribute to how pages are understood. Google says good site-wide signals do not mean everything ranks well, and poor ones do not mean everything ranks badly.6

The helpful content system became part of core ranking in March 2024. Its original announcement described a site-wide signal, and said that on sites with relatively high amounts of unhelpful content, any content is less likely to perform well, so removing unhelpful content could help the rest.6,7

Recovery is slow. Google says it could take several months for its systems to confirm that a site is now producing helpful, reliable content, and recommends against making changes to content that is already performing well.8

Why your reports have not answered it

The data is not wrong. It reports exactly what it says it reports, and the shape of that is easy to misread when you are looking for a culprit.

• Rare queries are anonymised. They are left out of the table and still counted in the chart, which is one documented reason the totals do not match the sum of the rows.
• The table shows at most a thousand rows, and a direct export is truncated to a thousand representative rows. On a site with thousands of pages, the tail you are looking for is not in the file you downloaded.
• The API is a different instrument. It accepts up to 25,000 rows a request and pages through the rest, which is the difference between a sample and a site.
• Average position means the topmost result from your site. In the chart that is the whole site, in a row it is that page. A falling site average can be a change in which pages appear at all rather than a change in where any single page ranks.
• If you suspect AI answers, the generative AI report combines AI Overviews and AI Mode and reports impressions rather than clicks per page, so it cannot confirm that a specific page lost clicks to them.

How much is anonymised for your property is not published, so the gap between the chart and the table is real and unquantified. That is worth knowing before anyone builds a theory on top of a total.9,10,11,12,13,14,15

The URL in the report is not the page in your head

Search Console assigns performance to the canonical URL. When someone clicks a duplicate in the results, the click counts toward the canonical rather than the URL they actually reached, and the page dimension groups by the final URL after redirects. If you did not choose a canonical, Google chose one for you and consolidated the signals onto it.13,16,2,17

Analytics counts the landing page instead, and its Search Console reports are filtered to landing pages that are also canonical. The two systems are describing different objects before anyone sits down to reconcile them, which is why that reconciliation never quite closes.18,19

This is also why a migration reads as a decline. After a URL change, Google says a medium-sized site can take a few weeks or more before the new URLs replace the old ones, longer for a large site, and advises keeping the redirects for at least a year so the signals transfer. For that whole period the page-level record is split across two sets of URLs.3

Pages rarely lose alone

When you do name the losing pages, they usually turn out to share something. That is not a coincidence, because most of the documented mechanisms operate on groups rather than on individual pages.

• A noindex rule on a page that robots.txt blocks is never seen, so pages can stay in results long after a team believes they were removed.
• A structured data manual action costs a page its rich result without changing how it ranks in web search. That arrives as lost clicks at an unchanged position, which is the hardest pattern to diagnose from a chart.
• Search Console itself groups URLs that share a user experience and applies one status to the whole group.
• Site-wide quality is assessed for the site overall rather than page by page, and Google says it can take several months for its systems to confirm that a site is producing helpful, reliable content again.
• Deleting pages is Google's last resort, for content that cannot be salvaged, not a first move when traffic falls.

One caution about speed, since it is usually the first suspect in the room. Google is explicit that page experience is not a single signal and that good Core Web Vitals do not guarantee rankings, and it says Search still seeks to show the most relevant content even when the experience is poor.20,21,22,23,8

What this cannot tell you

There is no published threshold of similarity at which two pages should be merged, no documented count of how many weak pages is too many, and no mapping from a rater page-quality rating to a ranking outcome. The raters themselves have no control over rankings and their data is not used directly in ranking.5,4

Google also gives no rule for when to redirect a retired page to a related one rather than let it return 410. Its only instruction is to return 404 or 410 for deleted or merged content.3

Anyone offering a numeric content score is inventing it.

Questions