aeo

Thin content in an extraction world

What thin means once boilerplate is stripped, how to find thin pages across a site, and when to consolidate, expand, or noindex. No invented word counts.

A page is thin when, after the navigation, footer, and template are stripped away, what remains does not answer anything on its own. That definition has nothing to do with a word count. A 120-word page that states a store’s hours, address, and phone number answers its question completely. A 1,500-word page that circles a topic without stating anything specific is thin however long it is. What changed with answer engines is not the definition. It is that the page is now judged by the block an extractor keeps, not by the page a person scrolls.

This post covers what thin means after extraction, how to find thin pages by hand and across a site, and the three things you can do with one: consolidate it, expand it, or keep it out of the index. It sits under Answer Engine Optimization: a page-level guide and builds on the boilerplate ratio, which measures how much of a page survives extraction in the first place.

Is there a minimum word count for AI answers?

No published minimum exists, from Google or from any answer engine. Google’s guidance on helpful content asks, as a self-assessment question, “Are you writing to a particular word count because you’ve heard or read that Google has a preferred word count?” and answers it: “No, we don’t” (Google Search Central: creating helpful content). The questions the same page does ask are about substance: whether content provides “a substantial, complete, or comprehensive description of the topic,” and whether you are “mainly summarizing what others have to say without adding much value.”

Length rules do exist inside tools, including ours, and they are useful as filters. They are not standards. A length rule finds candidates for a person to read. It does not decide that a page is thin.

What does thin mean once boilerplate is removed?

After extraction, a page is reduced to its main content block. Thinness is a property of that block, and it shows up in four recognisable forms.

FormWhat the main block holdsTypical cause
Empty shellA heading and one or two sentencesTemplate generated for every item in a database, whether or not there is anything to say
RestatementText that repeats the title and the category description in different wordsCopy written to fill a slot
Near-duplicateThe same block as dozens of other URLs, with one noun swappedLocation, colour, or size variants published as separate pages
Pointer pageA short intro and a list of links elsewhereTag pages, thin category pages, archive pages

The near-duplicate form is the one extraction makes worse. A person sees the full page, including a header and footer that make it look substantial. An extractor removes those and is left with the same paragraph it found on fifty sibling URLs. Google’s spam policies name the extreme versions: “scaled content abuse” is “when many pages are generated for the primary purpose of manipulating search rankings,” and the doorway section gives as an example “having multiple domain names or pages targeted at specific regions or cities that funnel users to one page” (Google Search Central: spam policies). Most thin pages are nowhere near those definitions. They come from templates built for completeness rather than for answers.

The test that cuts across all four forms is the stand-alone test from chunking: take the main block on its own, with the site removed. Does it answer a question someone would ask, with something specific, in a way that would still make sense quoted elsewhere? If not, it is thin, whatever its length.

How do you check one page by hand?

You need a browser and five minutes.

  1. Open view-source: for the URL and confirm the main copy is in the raw HTML. If it is not, the page may be thin only to crawlers that do not render scripts, and that is a rendering problem first.
  2. Open the page in Firefox’s Reader View, or select the main content and paste it into a blank document. That is a reasonable stand-in for what a heuristic extractor keeps.
  3. Read what remains, and write the question the page answers in one line. If you cannot, the page does not have a job.
  4. Write the answer to that question in one sentence, using only facts from the page. If the page does not contain the facts, it is thin.
  5. Search your own site for the same question. If another page answers it better, this page is a consolidation candidate.

The five steps produce a verdict and a reason, which is what the decision in the next section needs.

How do you find thin pages across a site?

At scale, use a length filter to find candidates and a similarity check to find near-duplicates, then send the candidates to a person.

In a crawler with custom extraction. Extract the text of your main content container, such as //main or //article, and a character count with string-length(normalize-space(//main)). Export to a spreadsheet, add your template column, and sort ascending within each template. The shortest pages on a template are the first ones to read. Comparing within a template matters, because a short store-hours page and a short article are not the same finding.

For near-duplicates. Many crawlers report exact and near-duplicate content directly. Where yours does not, extract the first paragraph of the main block and use COUNTIF in the spreadsheet to find openings that repeat across URLs. Pivot by template: one template producing many identical openings is a template finding, not a set of page findings.

In Search Console. Export the Pages table from the Performance report for the last three months. Pages on your thin-candidate list with no impressions at all are the strongest consolidation or noindex candidates. Pages with impressions and clicks are pages people find useful in some form; read them before changing anything.

From a terminal. A developer can extract main content with an open-source library such as trafilatura and count words per URL. It is a faster version of the crawler method and adds nothing else.

GEOBee reports the word count of the main content for every URL in its Words column, and its SEO lens flags pages under a thin-content floor that defaults to 100 words and can be changed in Preferences (the Overview & grid, crawl settings). Its AEO lens scores Extractability at 15 of 100, rewarding enough real text versus HTML noise, with “aim for 300+ words” as guidance (the three lenses). Both numbers are our heuristics for flagging candidates. Neither is a published threshold, and a page under either can be complete.

Consolidate, expand, or noindex?

Every thin page gets one of four outcomes. Decide by asking what job the page should do, not by its length.

SituationActionHow
Another page answers the same question betterConsolidateMerge any unique facts into the stronger page, then 301 redirect the thin URL to it
The page has a real job and real demand, but no substanceExpandAdd the facts that answer its question: specifications, steps, numbers, a definition
The page must exist for users but answers nothing for searchNoindexnoindex meta tag or HTTP header, and leave it crawlable
The page has no job and no demandRemoveReturn 404 or 410, and remove internal links to it

Consolidate is the right default for near-duplicates. Fifty location-variant pages with one swapped city name are usually one page about the service and a short list of areas served. Merge first, redirect second, and update internal links to point at the destination rather than leaving them to pass through the redirect. Status codes, redirects, and canonical drift covers how redirect chains and canonicals behave once you start merging.

Expand only when the page has a question to answer and you have the facts to answer it. Expansion that adds words without facts produces the restatement form above, which is thin at greater length. The additions that change a page are the specific ones: a number with a source, a step someone can follow, a definition that stands alone. What makes a page citable lists them.

Noindex fits pages users need but searchers do not: internal search results, filtered views, account flows. Google’s documentation states one requirement that is easy to get wrong: for noindex to work, the page “must not be blocked by a robots.txt file,” because a crawler that cannot fetch the page never sees the rule (Google Search Central: block indexing with noindex). Blocking and noindexing the same URL cancels the noindex.

Remove is for pages nobody needs. A 404 or 410 is a legitimate outcome, and a smaller set of pages that each answer something is easier to keep current. Freshness signals and honest dateModified is easier to maintain on two hundred pages than on two thousand.

Re-check after the change

Thin-content work is verified the same way it was found. Re-crawl the affected templates, re-run the extraction, and confirm that consolidated URLs return a single redirect to their destination, expanded pages now pass the stand-alone test, noindexed pages carry the rule and remain crawlable, and removed pages return the status you intended. Auditing answer-readiness across 10,000 URLs uses the same extraction-and-spreadsheet loop for section openings, and the two audits share most of their setup.

What this work changes is what an extractor holds when it reaches your pages: fewer blocks that say nothing, and more that answer one question completely. It does not decide whether those blocks are cited. It removes the pages that could not have been.


To list every page’s main-content word count and the thin-content flags across a site, download GEOBee. The 7-day trial includes full crawls and export.

On this page