You expanded a 1,200-word article to 3,500 words by adding background context, definitions, and tangential examples. Rankings dropped. A competitor with an 800-word page covering the same topic held position one. The assumption that more words means more depth fails because Google’s content evaluation systems do not count words. They evaluate topical coverage at the entity and subtopic level, passage-level relevance to query variants, and information gain relative to other indexed pages on the same topic. A 3,500-word page that repeats the same points in different words scores lower on these signals than an 800-word page that addresses distinct subtopics with unique information.
How Google’s Systems Measure Topical Coverage Without Counting Words
Google’s content evaluation operates through entity recognition, passage-level indexing, and topical modeling, none of which measure word volume. John Mueller has stated directly: “From our point of view the number of words on a page is not a quality factor, not a ranking factor.” Danny Sullivan reinforced this in 2023: “The best word count needed to succeed in Google Search is … not a thing.”
Entity recognition identifies the specific concepts, terms, and relationships a page addresses. When Google processes a page about “email marketing,” it identifies entities like open rate, subject line, list segmentation, A/B testing, and deliverability. The breadth of entities covered, not the word count devoted to each, determines the system’s assessment of topical coverage. A 600-word page that mentions and contextualizes 15 distinct entities related to email marketing demonstrates more topical coverage than a 2,000-word page that discusses only 5 entities in extensive detail.
Passage-level indexing, introduced in 2021, allows Google to evaluate and rank individual passages within a page independently of the full page’s topic. This means each section of a page is assessed for its relevance to specific queries. A page with ten well-targeted passages, each relevant to a different query variant, generates more passage-level relevance than a page with ten passages that all address the same core query from slightly different angles.
Topical modeling evaluates whether the page addresses the semantic dimensions of its topic. For a topic like “Kubernetes deployment,” the expected dimensions include configuration, scaling, monitoring, troubleshooting, and security. A page that addresses all five dimensions in 1,000 words achieves stronger topical coverage than a page that addresses only configuration and scaling in 3,000 words. The model does not reward volume. It rewards dimensional completeness.
Information Gain as the Core Depth Signal
Google’s information gain scoring concept, documented in a patent filed in 2018 and granted in June 2024 (“Contextual Estimation of Link Information Gain”), provides the most direct mechanism for evaluating content depth independently of length. The patent describes a system that determines an information gain score based on the unique information a document provides beyond information contained in other documents already presented to the user.
The mechanism operates at the entity level. Google identifies all entities and concepts within a document, then evaluates which of those entities represent new information relative to other documents covering the same topic. A page that introduces entities not found in the competing top-ranking pages receives a higher information gain score. A page that covers the same entities as every other ranking page, regardless of how much more text it uses to describe them, provides zero information gain.
This creates a direct penalty for the “skyscraper technique” when applied naively. If every top-ranking page for a query covers the same 20 entities, writing a longer page that covers those same 20 entities with more words adds no information gain. The competitive advantage goes to the page that introduces entity 21, 22, or 23, which might represent a fringe subtopic, original research finding, or novel application that no other ranking page addresses.
For example, in a competitive SERP for “content marketing strategy,” every ranking page might cover entities like content calendar, buyer personas, SEO integration, and social media distribution. A page that also covers dark social attribution, content decay measurement, or brand affinity modeling introduces entities with high information gain because they are underrepresented in the existing corpus. The length required to introduce these entities may be minimal, but the information gain signal is substantial.
The practical implication: adding 1,500 words of background context that restates commonly known information adds zero information gain. Adding 200 words that introduce a novel data point, an uncommon methodology, or a non-obvious subtopic connection adds measurable information gain. Depth is measured by the uniqueness of the information contributed, not by the volume of text.
Passage-Level Quality Assessment and the Helpful Content System
Google’s Helpful Content System, integrated into the core algorithm in March 2024, evaluates content quality at both the page level and the passage level. This creates a mechanism by which padding, content added to increase word count without adding substantive value, can actively harm a page’s ranking rather than simply failing to help.
The system evaluates whether each section of a page provides “a satisfying experience” for users seeking information on the section’s topic. Passages that restate information already covered elsewhere on the page, provide definitions that the target audience already knows, or offer generic advice that applies to any topic in the domain are evaluated as low-value passages. When a page contains a high proportion of low-value passages relative to its total content, the page’s overall quality assessment declines.
Mueller has stated that “not all pieces of content need to be comprehensive” and that “some questions just require a direct, quick, and simple answer.” This directly contradicts the padding approach, which assumes that every page benefits from being longer. When a topic can be fully addressed in 800 words, the 800-word treatment receives a stronger quality signal than a 2,500-word treatment that achieved its length by adding unnecessary context.
The passage-level assessment also explains why targeted content pruning, removing weak sections from an otherwise strong page, sometimes produces ranking improvements. When low-value passages are removed, the page’s average passage quality increases, and the system’s quality assessment improves even though the page became shorter.
Observable Signals That Distinguish Genuine Depth From Content Padding
Distinguishing real depth from padding requires examining specific content characteristics rather than aggregate metrics. Observable signals differentiate the two.
Genuine depth indicators: Multiple distinct entities per section, with each section introducing concepts not covered in other sections. Specific data points (numbers, dates, study citations, measured outcomes) rather than general claims. Original frameworks, taxonomies, or categorization systems that organize the topic in a non-obvious way. Coverage of edge cases, exceptions, and failure modes that generic treatments omit. Distinct subtopics that address different query intents, allowing individual passages to rank for different long-tail queries.
Artificial length indicators: High semantic similarity between paragraphs, where different sections express the same idea in different words. Definitional content that explains basic concepts the target audience already understands. Repeated conclusions restated at the end of multiple sections. Generic advice applicable to any topic in the domain (“always test your changes,” “monitor your results”) that does not contribute topic-specific insight. Extensive historical background that provides context without contributing actionable information.
A quantifiable test: count the distinct entities introduced across the full page. If a 3,000-word page introduces 12 unique entities and a 1,000-word page on the same topic introduces 10, the longer page provides only marginal entity coverage advantage despite being three times longer. The information density, measured as unique entities per word, is substantially lower in the longer page.
Practical Framework for Evaluating Content Depth Before Publication
Content teams can assess depth versus padding before publication using a three-step audit process.
Step 1: Subtopic mapping. Before writing, identify the distinct subtopics the page must address based on analysis of top-ranking pages and user intent research. Each subtopic should correspond to a different dimension of the topic. After writing, verify that each subtopic has a dedicated section and that no subtopic’s section merely restates another.
Step 2: Entity coverage check. Extract the key entities (concepts, terms, tools, processes, people, data points) from the page. Compare against the entities covered by the top 5 ranking pages. Count entities that the page covers that competitors do not (information gain entities). If the page introduces zero unique entities, it provides no information gain regardless of length.
Step 3: Passage independence audit. Read each H2 section in isolation. If removing any section would not reduce the page’s ability to satisfy the primary query, that section is a candidate for removal. If every section could be removed without damaging the core answer, the page is structured as accumulated padding rather than layered depth.
The decision criterion for when a page is complete: the page is complete when every expected subtopic has been addressed, the page introduces at least 2-3 entities not covered by competitors, and every section contributes a distinct point that no other section on the page makes. Additional length beyond this point adds words, not depth. For diagnosing whether underperformance is caused by insufficient depth versus other factors, see Content Depth Underperformance Diagnosis. For the misconception that longer content universally ranks better, see Content Depth Underperformance Diagnosis.
Does Google’s information gain scoring penalize pages that cover the same entities as competitors without adding new ones?
Google’s information gain scoring does not actively penalize pages for covering common entities. It assigns higher scores to pages that introduce entities not found in other documents covering the same topic. A page covering only the same 20 entities as every competitor receives a lower information gain score but is not penalized. The competitive disadvantage emerges because pages introducing entities 21, 22, or 23 score higher, making common-entity-only pages rank lower by comparison rather than by penalty.
Can removing low-value sections from a long page improve its ranking even though the page becomes shorter?
Removing low-value sections can improve rankings. Google’s Helpful Content System evaluates passage-level quality, and a high proportion of low-value passages relative to total content reduces the page’s overall quality assessment. When weak sections are removed, the page’s average passage quality increases, producing a stronger quality signal. This explains why targeted content pruning on individual pages sometimes produces ranking improvements despite reducing total word count.
How many unique entities should a page introduce beyond competitors to achieve meaningful information gain?
Introducing 2-3 entities not covered by competing top-ranking pages provides a measurable information gain signal. These entities might be fringe subtopics, original data points, uncommon methodologies, or novel application examples. The length required to introduce these unique entities may be minimal, but the signal is substantial. Pages introducing zero unique entities provide no information gain regardless of how much additional text they contain covering the same topics.
Sources
- Google Says Word Count Not a Quality Factor – Search Engine Journal
- Google’s Information Gain Patent – Search Engine Journal
- What Is Information Gain in SEO and Does Google Measure It? – SEMrush
- Creating Helpful, Reliable, People-First Content – Google Search Central
- Information Gain Score in SEO – Sara Taher
