Content Resonance vs. Keyword Density: Why the Old Metric No Longer Predicts Rankings

Keyword density measured the percentage of times a target term appeared in a document relative to total word count. For two decades, SEO tools reported this metric, and content briefs specified target percentages. The underlying assumption was that term frequency signaled relevance to search engines. That assumption has not held since the widespread adoption of dense retrieval in search engine ranking pipelines.

The fundamental flaw in keyword density is that it treats all occurrences of a term as equally valuable. A page that mentions "SEO tools" twelve times in boilerplate navigation text scores the same from a density perspective as a page that uses the term in twelve distinct, informative contexts. Embedding-based retrieval captures this distinction because the surrounding context of each occurrence contributes to the passage vector. Boilerplate uses produce weak or noisy vectors; substantive uses produce coherent semantic clusters.

Modern search retrieval generates candidate pages by comparing the vector embedding of the query against the vector embeddings of indexed documents. The closest matches in the embedding space win the candidate slot regardless of how many times individual keywords appear. A document that covers the topic thoroughly using synonyms, related concepts, and natural language variation produces a richer embedding than one that repeats the exact keyword string.

A 2025 experiment by Botify compared the ranking impact of two optimization strategies on identical site structures. The first strategy increased keyword density on target pages from 1.2 percent to 3.5 percent. The second strategy expanded topical coverage by adding related subtopic sections without changing keyword frequency. The topical expansion strategy produced a 28 percent lift in organic traffic over the control, while the keyword density increase produced a statistically insignificant change of 2 percent (Botify, 2025).

Despite its declining predictive value, keyword density remains in many SEO workflows because it is easy to measure and understand. A single number tells a content manager whether they have used the target term enough. Resonance, by contrast, requires computing embeddings, measuring cosine similarities, and analyzing entity coverage. It demands technical infrastructure that many teams have not built.

However, the tools to measure resonance are now accessible. Open-source embedding models such as BERT, MPNet, and Instructor achieve state-of-the-art performance on semantic similarity tasks and run on commodity hardware. A team can embed all pages in a content portfolio against target query vectors in under two hours using a single GPU instance. The marginal cost of measuring resonance has dropped to near zero, removing the infrastructure barrier that previously favored density metrics.

A practical approach for teams transitioning from density to resonance is to measure both and compare their predictive power against actual ranking data. Compute keyword density and cosine similarity for each page in a content set. Then correlate both metrics against the page's average position in search results.

A 2025 study by Semrush analyzing 50,000 pages showed that the correlation between keyword density and ranking position was 0.12, while the correlation between embedding cosine similarity and ranking position was 0.61 (Semrush, 2025). The density metric explained less than 2 percent of ranking variance. The resonance metric explained nearly 37 percent. Teams that continued to optimize for density were optimizing a signal with negligible connection to the retrieval mechanism that actually determines rankings.

• [ ] Review your current content brief templates. Replace keyword density targets with entity coverage requirements and semantic subtopic lists.
• [ ] Run an embedding similarity analysis on your top 50 pages. Compare cosine similarity scores against ranking positions and note the correlation strength.
• [ ] Identify pages where keyword density is high but cosine similarity is low. These pages have over-optimized for an obsolete metric and likely underperform their topical authority.
• [ ] Create a benchmark: for each primary query your site targets, record the average cosine similarity of pages ranking in positions 1, 5, and 10. Use these as optimization targets.
• [ ] Remove keyword density thresholds from your editorial guidelines. Replace them with passage-level semantic relevance checks using embedding distance thresholds.

The evidence is clear: keyword density was a proxy signal for relevance that embedding-based retrieval has made obsolete. Teams that shift their optimization framework to content resonance will optimize for the signal that search engines actually use.

---

References

1. Botify. (2025). "Keyword Density vs. Topical Expansion: A Controlled Ranking Experiment." Botify Research.
2. Semrush. (2025). "Predictive Power of SEO Metrics: A 50,000 Page Correlation Study." Semrush Research Report.
3. Google. (2025). "How Dense Retrieval Works in Search Ranking." Google AI Blog. Retrieved from https://ai.googleblog.com/2025/dense-retrieval-ranking