SEO Trends

The data: Semrush: 0.47 correlation (highest individual factor). Our own July 2026 re-baseline puts semantic vocabulary breadth, the closest measurable proxy for relevance depth, at r=-0.29, the strongest single on-page correlation in our dataset and up from -0.22 in April. Every study puts content relevance #1 or #2.

Content that directly matches search intent and demonstrates comprehensive topical understanding consistently outranks content optimised purely for keywords. Semrush's 300K SERP study found text relevance correlates at nearly double the strength of any authority metric. Our own data now agrees more strongly than it did six months ago: the breadth of genuine topic vocabulary on a page is the single best on-page predictor of position we measure.

The data: Backlinko: #1 results have 3.8x more backlinks than #2-#10. Our July 2026 re-baseline: referring domains r=-0.26 (up from -0.16 in April) and total backlinks r=-0.26 (up from -0.16), both now inside our top 5 factors overall. Page 1 pages carry a median 54 referring domains. Nofollow links correlate at -0.24, nearly as strongly as dofollow.

This is the finding that surprised me most this cycle, and it runs against what I have been writing. Most public studies report backlink weight declining year over year. In our own data it moved sharply the other way in six months. Two readings are possible: link signals genuinely regained weight, or our query sample rotated toward more competitive terms where link profiles separate winners more sharply. I cannot yet tell you which. What I can say is that the quantity-to-quality shift still holds: referring domain diversity and topical relevance are what move, not raw link count, and 10 referring domains is an eligibility floor rather than a competitive target.

The data: Every source lists intent alignment as a prerequisite for ranking. ClickRank and Semrush both position it as the gateway signal: if intent is wrong, nothing else matters.

Google evaluates whether your page satisfies what the user actually wants, not just whether it contains the right keywords. A perfectly optimised page targeting the wrong intent will not rank. This is especially true post-AI: Google's systems now understand query intent at a much deeper level.

The data: Our July 2026 re-baseline: load time r=-0.25, strengthened from -0.24 in April but no longer the strongest individual on-page factor, having been overtaken by semantic vocabulary breadth (-0.29) and referring domains (-0.26). It remains the strongest performance factor we measure. Page 1 pages average roughly 2.2s full load. INP has replaced FID as the interactivity metric.

Page speed did not get weaker. It got stronger in absolute terms and still fell down the table, because content depth and authority signals strengthened faster. That distinction matters for how you spend: speed is a top-tier factor and a hard floor, but it is no longer the single biggest on-page lever, and stripping images and copy from a page to chase a speed score is now a net loss against the content-depth signals that overtook it.

The data: Search Engine Land frames this as the absolute prerequisite: if Google cannot crawl it, nothing else matters. Every technical SEO source lists HTTPS, clean URLs, XML sitemaps, and robots.txt as baseline requirements.

This is the unsexy factor that gates everything else. Pages that are blocked, slow to render, or riddled with redirect chains never get the chance to compete on content or links. Our client data confirms it: pages implementing more baseline structural elements consistently outperform those with gaps.

The data: Google's Quality Evaluator Guidelines treat E-E-A-T as a page quality framework, not a direct ranking signal. However, multiple studies show strong correlation between E-E-A-T indicators (author bios, citations, credentials) and rankings, especially for YMYL topics.

E-E-A-T is not a single score Google calculates. It is a framework reflected across multiple ranking systems: site quality scores, author entity recognition, consensus scoring, and trust signals. The extra 'E' for Experience (added 2022) rewards first-hand knowledge, not just credentials.

The data: Our client data: dwell time 0.84 correlation (strongest individual engagement signal). Bounce rate: -0.68 correlation. Backlinko: average time-on-site for page 1 results is 2.5 minutes. Engagement weight growing fastest year-over-year (~12% and climbing).

Google's NavBoost system measures clicks, long clicks, short clicks, and last longest clicks. The correlation data is striking: dwell time at 0.84 is the strongest individual engagement correlation we see across our client base. User engagement is no longer a tiebreaker; it is a primary signal.

The data: Surfer SEO: #1 on-page factor (topical coverage depth). Semrush: semantic relevance dominates their correlation data. Our July 2026 re-baseline: unique semantic vocabulary r=-0.29, now the single strongest on-page factor we measure, up from -0.23 in April. Page 1 pages carry roughly 250 distinct topic-related terms.

Google's API leak confirmed siteFocusScore, siteRadius, and topPetacatTaxId as real signals measuring how focused a site is on its core topic. Sites that demonstrate deep expertise in a niche outperform generalist sites, even with fewer backlinks. Our correlation data now backs this harder than it did in April: semantic vocabulary breadth went from the third-strongest on-page factor to the outright first. The mirror image is just as useful: over the same period, keyword-in-title-tag factors fell sharply down our table. Depth in the body is being rewarded where title-tag keyword engineering is not.

The data: Ahrefs 75K brand study: brand web mentions 0.664 correlation with AI visibility (3x stronger than backlinks at 0.218). YouTube mentions: 0.737 correlation. Google patent US 9,031,929: brand search ratio as a site quality signal.

This is the most divergent factor in the consensus. Traditional SEO studies underweight brand signals. But Ahrefs' AI visibility data shows brand mentions are the single strongest predictor of AI search citation. Brands mentioned frequently across the web (especially on YouTube) are 6.5x more likely to be cited in AI-generated responses.

The data: Our July 2026 re-baseline: specific JSON-LD types with 3+ properties correlate at r=-0.21, ahead of generic 'has schema markup' at -0.16. The notable change this cycle is which properties matter: datePublished and dateModified entered our top 200 from outside it, while product metadata (sku, gtin, Product) dropped out. Surfer SEO: too many schema types on a single page correlates negatively with rankings.

The nuance matters here. Having schema is good. Having specific, detailed JSON-LD with the right @type is better. But over-implementing every possible schema type actually hurts. The new wrinkle in our July data is a rotation from commerce metadata to publishing and freshness metadata: date fields, headline, WebPage, WebSite and ImageObject all rose while product identifiers fell. Keep Product schema on product pages regardless, because rich result eligibility is a separate benefit from position correlation.

The data: Our July 2026 re-baseline: number of shared on-page factors used r=-0.26, up from -0.23 in April, with all three of our breadth counters now inside the top 7 factors. Notably, the focused counters (top 50 and top 100 factors used) correlate more strongly than an undifferentiated count of all factors used. Most successful sites scored high across ALL major factors. No single silver bullet.

Pages that are well-rounded across multiple signal categories (structure, keywords, entities, schema, links, meta) outperform pages that over-index on any single factor, and that effect strengthened this cycle. The shape of the data adds a caveat worth having: breadth across the factors that actually matter beats breadth for its own sake. Ticking 300 boxes is not better than ticking the right 50.