What are the SEO ranking factors that truly impact Google positions?

Out of 150,000 pages analyzed, the number of images (0.95) and semantic optimization score (0.93) dominate ranking correlations, far ahead of exact query in title (0.69). Tables, lists, structured data, and videos have virtually no measurable impact. Caution: correlation is not causation.

What does the analysis of 150,000 pages reveal about ranking factors?

This study analyzes the correlations between 23 on-page factors and positions in Google results, across more than 150,000 French pages.

The major finding upends common beliefs: the number of images shows the strongest correlation (0.95), ahead of semantic optimization score (0.93) and word count (0.85). The number of products on e-commerce pages follows at 0.79.

In contrast, practices considered important show weak correlations: exact query at the beginning of title (0.75), presence of an H1 (0.69), or exact query in title (0.69). Tables, lists, videos, and structured data have virtually no measurable impact.

Titles with numbers even show a negative correlation (0.77). Only 25% of top 10 pages contain the exact query in the title, questioning the dogma of keyword stuffing.

The median content of first results is 736 words versus 619 for tenth positions, but 28% of pages are under 300 words, suggesting that quantity alone is not enough once in the top 10.

What are the methodological limitations to understand for expert interpretation?

This analysis relies on statistical correlations, not direct causal relationships. A strong correlation can mask a confounding factor.

The number of images comes first, but no satisfactory explanation emerges. E-commerce pages with many products naturally have more images, yet the number of products only correlates at 0.79. The hypothesis: images could be a proxy for other signals (content richness, editorial effort, user engagement).

The weak correlation of exact queries in titles is surprising, but explainable: on 3-4 word queries, Google understands semantic variations. The algorithm prioritizes contextual relevance over exact matching.

Structured data shows zero impact on ranking, confirming Google's official position: they serve rich snippet display, not positioning. Their presence may correlate with CMS age rather than quality.

Major limitation: the study excludes off-page factors (backlinks, domain authority) and behavioral metrics (CTR, time on page), which heavily influence rankings. The results concern only on-page in an already competitive context (top 10).

What are the debatable points worth critical analysis?

[Opinion] In my view, the correlation of image count remains suspect. Across 150,000 pages, such consistency suggests either an unidentified hidden factor or selection bias. My experience shows that massively adding images without editorial relevance doesn't boost positions. I lean toward confusion: sites that invest in iconography also invest elsewhere (UX, content, technical).

[Generalization] The claim "having an H1 has little impact" deserves nuance. In the top 10, 91% of pages have an H1. So it's not a differentiating factor at this level, but probably a prerequisite to get there. I would nuance by saying that the absence of H1 excludes you from the top 30, but its presence won't move you from 8th to 3rd place.

[Experience Feedback] The weakness of exact query in title matches my field observations. For 2-3 years, semantic variations rank as well as exact keywords. Google understands that "best online casinos" and "top casino sites internet" target the same intent.

[To Verify] The negative correlation of numbers in titles (-0.77) seems counter-intuitive against clickbait best practices. This could reflect a bias: listicles perform in CTR but not necessarily in pure algorithmic relevance. Commercial queries without list expectation would be penalized by this format.

[Opinion] My view on semantic optimization (0.93): it's the only strong actionable factor. Unlike images whose impact remains mysterious, boosting semantic score has a verifiable mechanical effect. However, 50% of first results exceed the recommended score by only 10%, suggesting diminishing returns beyond a threshold.

• Does multiplying images become an unexpected on-page priority? The strongest correlation (0.95) suggests a poorly understood but statistically robust signal. Integrate 10-15 relevant images per long-form content, using AI generators like Ideogram to produce quickly.
• Does semantic optimization beyond recommended score offer marginal advantage? 50% of first results exceed the recommended score by 10%. Aim for 10-20% above the threshold to differentiate without falling into over-optimization.
• Does reaching 1200+ words remain structural but not differentiating in top 10? The median at 736 words and 28% of pages under 300 words show that quality prevails. Produce dense rather than long content to stand out once well-positioned.
• Should we abandon the obsession with exact query at title beginning? Only 25% of top 10 use it, weak correlation at 0.75. Favor natural phrasing that captures intent and optimizes CTR.
• Should we drastically increase product count per e-commerce page? Correlation at 0.79, one of the few direct actionable levers. Move from 8-16 to 32-64 products per category to reduce unnecessary pagination and enrich semantic field.
• Should we ignore structured data for pure ranking? Near-zero correlation (0.04). Implement them only for rich snippets and CTR, not to climb organic positions.
• Should we avoid wasting time on tables and lists as ranking factors? Null correlations polluted by old CMS. Use them for user readability, not for hypothetical algorithmic boost.
• Does shortening URLs maintain medium correlation (0.77) probably indirect? Home pages and pages high in architecture rank better and have short URLs. Optimize crawl depth rather than URL length in isolation.