There’s a persistent belief in SEO that a 2,000-word page outranks an 800-word page because Google rewards length. It’s wrong. Word count is a proxy for something real, but a bad one — and confusing the proxy for the signal is costing people rankings they should own.
What Google is actually trying to infer is whether a page has sufficient topical depth to satisfy the information need behind a query. That’s a different measurement problem than counting words.
What “Topical Depth” Means in Practice
Google’s systems don’t evaluate depth by length. They evaluate it by semantic coverage — specifically, whether a page addresses the sub-concepts, related entities, and expected co-occurring terms that a document on this topic would plausibly contain if written by someone who genuinely understood the subject.
This is downstream of the same vector-space logic that powers passage-level retrieval. A page about “insulin resistance” that never mentions fasting glucose, HOMA-IR, or adipose tissue is missing expected semantic neighbors. That absence is a signal — not a hard penalty, but a legibility gap. Google can’t confidently place that page in the conceptual neighborhood it’s competing in.
The mechanism isn’t mysterious. Language models trained on large corpora develop strong priors about what terms co-occur with what topics. When a page is processed, its semantic fingerprint is compared against those priors. Pages that match the expected fingerprint score higher on topical relevance, regardless of how many words they use to get there.
The Word Count Trap
Here’s where the confusion comes from. Longer pages often happen to cover more sub-concepts — not because length causes depth, but because writers who know a subject naturally surface related entities and edge cases as they write. Word count correlates with depth the same way a wet street correlates with rain. The street didn’t cause the rain.
We’ve seen this pattern clearly in our own work: a 950-word page with tight semantic coverage outranking a 2,400-word page that restates the same three points, then bolts on two FAQ sections padded with synonyms. Higher word count, lower topical depth. Google picked the shorter one.
Stuffing a page to 3,000 words by repeating core concepts, adding transitional filler, or appending a generic FAQ doesn’t expand your semantic fingerprint. It adds tokens that cluster around the same concepts rather than reaching into adjacent but relevant conceptual space.
What Actually Moves the Needle
Covering the Expected Sub-topics
For almost every topic, there’s a set of sub-questions and related concepts that a well-informed document would address. These aren’t necessarily the sub-headings you’d expect — they’re the conceptual neighbors that appear in source material Google already trusts on this topic.
A practical way to find them: read the actual body text of the top-ranking pages for your target query, not just their headings. What entities keep appearing? What processes, qualifications, or edge cases do multiple sources reference? Those are the semantic neighbors you need. Not because of some NLP trick, but because leaving them out genuinely leaves your content less complete.
Specificity Over Breadth
Depth is not the same as breadth. A page that goes one level deeper on a single aspect of a topic — real mechanism explanation, real edge cases — often outperforms a page that touches ten aspects at surface level.
This matters especially for queries where the user has a follow-up question behind their initial search. Google’s passage retrieval can surface a well-specified section of a longer page, but only if that section actually contains the specificity. Vague coverage of many topics gives the passage retrieval system nothing to grab.
Entity Disambiguation
One underused mechanism: explicitly disambiguating the entities on your page. If your page covers “Mercury” the planet rather than the car brand or the element, establishing that through co-occurring entities — orbital period, NASA missions, perihelion, magnetosphere — helps Google anchor the page in the right part of its knowledge graph. A page that surfaces the right co-occurring entities for this sense of a term is demonstrably more specific than one that floats ambiguously.
Where Schema Fits In
Structured data can support topical depth inference, but not in the way most people use it. Marking up an Article with just a name and description is noise. What actually helps is using schema to make entity relationships machine-readable: about, mentions, mainEntity — properties that explicitly assert what this page is about and what it connects to.
Schema is a supporting signal, not a substitute for the underlying content. If the body text doesn’t have the semantic coverage, markup doesn’t conjure depth from nothing. Google can read both and will notice when they diverge.
The Caveat Worth Taking Seriously
None of this means short pages always win. Topic complexity matters. A query like “what is a cookie” can be satisfied in 400 words. A query like “how to implement OAuth 2.0 PKCE flow in a single-page application” cannot — not because Google requires word count, but because the topic genuinely requires more conceptual coverage to be useful. Six hundred tightly-written words on that query is undercoverage, not efficiency.
The test isn’t length. The test is: does this page actually answer the question, including the sub-questions a competent person would anticipate? If yes, you’re done. If not, the fix is more substance, not more words.
A Diagnostic Worth Running
Take a page you think should rank better than it does. Extract all the noun phrases and named entities from the body text. Do the same for the top three ranking pages on the same query. Look for entity classes that appear in theirs but not yours — not keyword matches, entity classes. Process, outcome, mechanism, qualification, counter-example.
What’s missing from your entity set is almost always a better explanation for the ranking gap than word count, title tag, or link profile.
Fix the conceptual gaps first. The length problem usually solves itself.