Should I delete thin programmatic pages?

Not before checking what they earn. On one site of mine, a templated matrix that was about 1% unique text earned 46% of search clicks from 13% of impressions. Export clicks by page type first. Prune pages that have earned nothing for months, one by one, and keep pages that earn more than their share of clicks, however thin they look.

Why do thin template pages sometimes rank well?

Because they answer a narrow question precisely. A page titled with the exact place and topic someone searched for answers them in its first line, even when the rest of its text is shared with hundreds of sibling pages. Search engines reward the match between the question and the answer, not originality measured across your own site.

How do I decide which programmatic pages to prune?

Use outcomes, not rules about importance. Start with pages that earned zero impressions over a full quarter, check each is not new or uncrawled, and leave alone any page that earns clicks. Avoid blanket rules, such as hiding pages for small regions, because the smallest ones can rank near the top.

The Thin Pages That Earn 46% of the Clicks

¶ WAR STORYTHE PAGES I WANTED GONE

01Four hundred pages of filler

One of the editorial sites I run has a programmatic matrix: a page for every combination of region and topic, generated from one template with a handful of local figures dropped in. Measured by unique text, the pages were about 1% different from each other. Every audit checklist I have ever read would call that thin content, and so did I. The plan was to collapse the matrix into a few strong hub pages and redirect the rest.

The data I nearly skipped

Before deleting anything I exported three months of search data, mostly to have a before picture for the redirect. The site had 68 clicks from 11,923 impressions, with impressions up 164% from the first half of the period to the second. Then I split it by page type. The matrix pages had drawn 13% of the impressions and earned 46% of the clicks. The pages I was about to delete were the best converting pages on the site.

// KEEP OR KILL~~ONE SITE · THREE MONTHS OF SEARCH DATA~~

SHARE OF IMPRESSIONS

13% matrix

SHARE OF CLICKS

46% matrix

126matrix pages with impressionsKEEP, AND LIFT ANY NOINDEX

330matrix pages with zero impressions in three monthsPRUNE CANDIDATES, ONE BY ONE

2.7 to 6.4positions held by the smallest regions, the ones a rule had hiddenTHE RULE WAS WRONG

The same pages were about 1% unique text. The engine ranked them anyway, and people clicked.

¶ WAR STORYWHY THE FILLER WON

02Specific beats original

The reason is not mysterious once you look at what people typed. Someone searching for a topic in a particular region is asking a narrow question, and a page titled with exactly that region and that topic answers it in the first line, even if the paragraphs below are shared with four hundred siblings. The broad hub pages I wanted to keep were better writing and a worse answer. A search engine does not grade originality. It grades whether the page answers the question that was asked.

The rule that hid the winners

It got worse. A noindex rule on the site hid matrix pages for the smallest regions, on the theory that places with few people would never be searched. The two smallest regions were ranking at positions 2.7 to 6.4 and earning clicks anyway. Somebody had already noticed and hand patched an override to rescue them, which is how I found the rule at all. A rule based on how important a page ought to be will eventually hide a page that is working.

There is an honest caveat. Sixty eight clicks is a small sample, and 46% of it is about 31 clicks, which I would not bet a whole site on by itself. What made it decisive was that everything pointed the same way: impressions growing fast, the matrix converting better than any other page type, the smallest regions ranking near the top, and nothing in the data arguing for the deletion except my own sense of what a good page looks like.

This is the same trap as trusting the traffic chart over the logs: an assumption about what should matter, standing in for a measurement of what does.

EXTRACTED — THE SENTENCE THIS ENTRY EXISTS FOR

Audit checklists measure what a page is. Search data measures what it does.

¶ WAR STORYWHAT I DID INSTEAD

03Prune the tail, not the set

The tail, one page at a time

The matrix was not all gold. Of 456 matrix pages, 330 had not earned a single impression in three months. Those are real pruning candidates, and they get looked at one by one, because a page with no impressions in a quarter can still be new, or still be waiting on a crawl. The other 126 stay, the noindex on any of them that earns impressions comes off, and the population rule is gone. The pages that embarrass you are sometimes the pages doing the work.

The real ceiling was position

The more useful finding was underneath. The site's head terms, the broad searches the hubs were built for, sat at positions 76 to 88. Adding or removing pages was never going to fix that. At that depth the problem is authority and links, which is a different fight from page count, and deleting the pages that do rank would only have made the site smaller while it lost that fight.

When consolidation is right

None of this is an argument for publishing templates without limit. Consolidation is right when pages compete for the same query and split the clicks, when a set of pages has earned nothing for a long time, or when the template says something untrue for some combinations. Consolidation is wrong when the only evidence against a page is that it looks thin. I now run one test before deleting anything: export the clicks by page type first, and if the pages I am about to remove earn more than their share, stop. The same instinct applies to alarming reports in general. Look at outcomes before you act on appearances.

Tomorrow: an audit that found rich structured data everywhere except the one page where money changes hands.