On-Page SEO: Complete Optimisation Framework (2026)

Google processes 8.5 billion searches per day using semantic understanding — and the pages that win are no longer those with the most keyword repetitions, but those that best demonstrate relevance, authority, and genuine helpfulness (SEO Score Tools). A landmark 2024 study of 320 websites found that pages optimised for keyword density experience a 28% higher bounce rate and 14% lower conversion rate compared to entity-first optimised pages (SearchAtlas). The playbook has changed.

This guide covers every lever you can pull directly on the page — title tags, headings, content structure, internal linking, E-E-A-T signals, structured data, and optimising for AI Overviews — with specific benchmarks, source-backed statistics, and a prioritised checklist you can act on today.

For the technical foundations that sit beneath on-page work, see Technical SEO.

Quick answer:

On-page SEO is the practice of optimising individual web pages to rank higher in search results. In 2026 this means: matching search intent in your title and content, building entity salience around your primary topic, structuring content with clear H2/H3 passages that target featured snippets, keeping internal links active and descriptive, demonstrating E-E-A-T through author credentials and original data, and using JSON-LD structured data to help AI systems understand and cite your content.

---

Title Tags: The First Signal Google Reads

Title tags remain "probably the most important ranking factors on a page" — they're the first contextual signal Google encounters (Redevolution). Court testimony from the U.S. v. Google antitrust case confirmed that query terms in title and headings are "actually kind of crucial" for topicality scoring.

The rewrite problem

76% of title tags were rewritten by Google in Q1 2025 (Search Engine Land). Google pulls replacements from H1s, body text, or external anchor text when it judges your title doesn't match page content or user intent. Common triggers:

• Keyword stuffing or unnatural phrasing
• Length that truncates badly in SERPs
• Mismatch between title promise and actual content
• Brand name dominating at the expense of topic clarity

What works

• 50–60 characters for reliable SERP display without truncation (Search Engine Land)
• Include target keyword naturally — no forced exact-match repetition
• Match intent, not just keywords: informational intent needs a different frame than transactional
• Well-crafted title tags can lift CTR by 20–50% (Medium / Rank Website SEO)
• Monitor Google Search Console for title rewrites and adjust within two to three weeks of detecting a change

---

Headings: Hierarchy, Passage Indexing, and Featured Snippets

Headings serve three jobs simultaneously: they signal topical structure to Google, they enable passage indexing of specific content sections, and they determine whether your content earns featured snippets.

What the data shows about H1 usage

A 2026 Rankability study of 5 keyword sets across the top 40 Google results found (Rankability):

• 93.5% of top-ranking pages have a single H1 tag
• 13.5% have the exact keyword phrase in H1 — with a weak negative correlation to rankings (-0.2670)
• 88.5% partially match the keyword in H1 — with negligible correlation

The takeaway: use one H1 per page, include a natural partial match to your primary topic, but don't chase exact-match cramming.

John Mueller has said Google gives "a slight boost if we see a clear heading on a page" — but headings alone are not the ranking factor. The structure and semantic clarity they provide matter more than the presence of specific keywords (Greenlane Marketing).

Passage indexing: the 40–100 word opportunity

Google's passage ranking system indexes and ranks specific 40–100 word passages from within long-form content independently of the overall page (Data Enriche). This is powered by BERT and MUM language models.

A restructuring case study showed: retrofitting 20 articles with proper H2 passage structure moved 17 of 20 articles into the top 5, increased average traffic by 220% per article, and captured 31 new featured snippets (Data Enriche).

The content architecture implication: write the first 40–60 words under each H2 as a direct, self-contained answer to the question the H2 poses. Then provide supporting detail.

Featured snippet targeting

Featured snippets increase page CTR by more than 25% even from positions 2–10, and 70% of featured snippets come from pages ranked positions 2–10 (Niumatrix). 32.3% of featured snippets come from content placed immediately after an H2 or H3 tag.

Snippet distribution:

• Paragraph snippets: ~70% of all snippets (target 40–50 word answers)
• List snippets: ~19%
• Table snippets: ~7%

Heading best practices for 2026

• One H1 per page, include primary topic near the start
• Add a header approximately every 300–400 words (Digital Applied)
• Aim for 3–8 H2 sections per article
• Phrase H3s as questions — they're highly effective for featured snippet capture (Conductor)
• Never skip heading levels (W3C specification: "documents should not skip levels")
• Use question-style H2s for AI Overview targeting

---

Content Optimisation: Entity Salience Over Keyword Density

"By 2026, the gap between keyword-centric and entity-centric optimisation has become a chasm" (SearchAtlas). Google's Knowledge Graph now contains over 500 billion facts about 5 billion entities — and its NLP systems evaluate content through entity recognition, not keyword counting.

How entity salience works

Entity salience measures how central a recognised entity is within a document, scored 0 to 1. Google's NLP documentation treats a score of 0.5+ as indicating primary topical relevance (SearchAtlas). If your target entity scores below 0.10 on a page ostensibly about that entity, you have a content focus problem.

Five factors determine entity salience:

1. Position prominence — where the entity appears in the document structure
2. Frequency patterns — how often, weighted by context
3. Syntactic relationships — how entities connect grammatically
4. Semantic context — surrounding information reinforcing importance
5. Co-occurrence patterns — which related entities appear nearby

Missing co-occurring terms is a signal problem. When human experts write about a topic, they "naturally use a predictable, highly correlated cluster of related terms." If those terms are absent, "the NLP model will struggle to confidently classify the page as being an expert resource" (Contadu).

Content depth and word count

Refreshed content generates 106% more traffic compared to publishing new articles on the same topic (Upward Engine). But word count alone is not the lever: pages recovering from Helpful Content Update penalties averaged 1,400 words, while non-recovering pages averaged 1,650 words. Information density beats volume.

Content freshness now accounts for approximately 23% of Google's ranking factors as of Q1 2025, and user engagement metrics (bounce rate, time on page) account for roughly 12% of algorithm influence (Upward Engine).

AI-generated content: the risk profile

A 16-month study of 4,200 articles across 140 domains found (Digital Applied):

• Pure AI content had a 3.2× higher deindexation rate following spam updates
• The ranking gap between pure AI and human content grows from 14% at 3 months to 31% at 16 months
• AI-assisted content (with 30%+ rewriting, original data, named expert attribution) achieved only a 4% median ranking position difference at 16 months

Sites publishing more than 50 AI-only articles per month were disproportionately hit by deindexation events. The March 2026 Core Update confirmed this: pure AI content saw a -18% average ranking decline while human-written content with strong E-E-A-T saw +11%.

Google does not automatically penalise AI-generated content, but explicitly states "AI or automation disclosures are useful for content where someone might think 'How was this created?'"

---

Internal Linking: The Highest-ROI Low-Cost Tactic

25% of web pages receive zero internal links (Digital Applied). That means a quarter of your site is invisible to crawlers and starved of PageRank — a fixable problem that costs nothing but audit time.

The sweet spot

A Zyppy SEO study of 23 million internal links across 1,800 websites found that pages with 45–50 internal links show a significant organic traffic boost — with traffic declining above that threshold (Inblog). The recommended density: 2–5 contextual links per 1,000 words.

Critical pages should sit within three clicks of the homepage — pages at depth 1–3 generate 9× more SEO traffic than those at depth 4 or beyond.

The pillar-cluster model

A hub-and-spoke content architecture is the foundation for topical authority and internal link flow:

• One pillar page covers a broad topic comprehensively
• 8–15 cluster pages each handle a specific subtopic
• Cluster pages link back to the pillar; the pillar links to each cluster
• Closely related cluster pages cross-link to each other

Clustered content drives roughly 30% more organic traffic than isolated keyword posts, and cluster rankings persist about 2.5× longer than standalone pieces (Single Grain).

Anchor text distribution

Pages with at least one exact-match anchor had roughly 5× the traffic of pages without one. Recommended distribution (Upward Engine):

Anchor Type	Target %
Branded	30–50%
Naked URL	20–30%
Generic ("click here")	10–20%
Partial match	5–15%
Exact match	1–5%
LSI / related	5–10%

Never use "click here" or "read more" as anchor text — it eliminates the topical signal entirely (LinkBoss).

Link equity priority tiers

• Tier 1 (conversion pages): 15+ inbound internal links
• Tier 2 (pillar pages): 8–15
• Tier 3 (cluster posts): 3–8
• Tier 4 (admin / thin content): 0–2

---

E-E-A-T: What Actually Signals Quality

E-E-A-T — Experience, Expertise, Authoritativeness, Trustworthiness — is not a direct ranking factor but a framework Google's Quality Raters use to evaluate content. "Using a mix of factors that can identify content with good E-E-A-T is useful" (Google Search Quality Rater Guidelines). Trust is the most important of the four.

Google's March 2024 Helpful Content integration resulted in a 45% reduction in low-quality, unoriginal content in search results.

Practical implementation

Experience: Include case studies with real outcomes and dates. Use screenshots, data extracts, and process maps. Share personal insights with specifics — not generic observations.

Expertise: Author bios should list relevant credentials, years of experience, notable projects, and links to external profiles (LinkedIn, conference talks). "Clear authorship information, such as bylines to content where readers might expect it" is strongly encouraged by Google (Gravatar Blog).

Authoritativeness: Inbound links from reputable industry sources. A single link from a respected industry site carries more weight than dozens of generic directory listings.

Trustworthiness: Clear privacy disclosures, verified reviews on G2 or TrustRadius, robust contact routes, and transparent pricing or trial terms.

Schema markup for E-E-A-T

• Person schema for authors with sameAs links to Wikipedia, LinkedIn, or Wikidata helps Google verify identities
• Article schema with author and reviewedBy properties clarifies creation and editorial oversight
• Organization schema establishes consistent publisher identity
• Speakable schema flags your most citable passages for AI synthesis

The sameAs property is "the most direct way to tell Google exactly which entities your content is about" (SEO Score Tools / Contadu). JSON-LD in the <head> remains Google's preferred format.

---

AI Overviews: The New SERP Reality

AI Overviews (AIOs) now trigger for approximately 30% of US desktop keywords (seoClarity), with zero-click searches rising from 56% to 69% between May 2024 and May 2025 (Similarweb via IDEAVA).

The CTR impact

An Ahrefs study of 300,000 keywords found that position 1 CTR for AIO-triggering queries dropped from 0.073 in December 2023 to 0.016 in December 2025 — a 58% reduction (IDEAVA). For publishers, this is an existential shift: Google search traffic to publishers declined globally by 33% in the year to November 2025.

Being cited in AIOs is better than not

Queries where an AIO appears and your site is cited: CTR is 35% higher than cited competitors and 91% higher on paid (Seer Interactive via IDEAVA). The top 20 organic results appear in AIO sources 97% of the time — so ranking traditionally is still the primary path to AIO citation.

Optimising for AIO citation

• Put direct, concise answers in the first 100 words of each section
• Use clear question-based H2/H3 subheadings
• Structure content with lists and tables where appropriate
• Add Speakable schema to flag citable passages
• Build tight semantic clusters — entity-linked content gets cited more often
• For internal links: place them near definitions and data points, not mid-paragraph (ZCMarketing)

For a deeper look at how content structure signals work across the full SEO stack, see Content & On-Page.

---

On-Page SEO Optimisation Checklist

Element	Target	Priority
Title tag	50–60 characters, natural keyword, matches intent	High
H1 tag	One per page, partial keyword match, 50–60 chars	High
H2/H3 structure	Question-phrased, 300–400 word intervals, 3–8 H2s	High
Passage answers	40–60 word direct answers under each H2	High
Entity salience	Primary entity in title, H1, first paragraph, schema	High
Internal links	2–5 contextual per 1,000 words, descriptive anchors	High
Pillar-cluster architecture	Critical pages within 3 clicks of homepage	High
Author bio	Named author, credentials, external profile links	Medium
E-E-A-T signals	Firsthand data, case studies, dates, methodology	Medium
JSON-LD schema	Article + Person + Organization + BreadcrumbList	Medium
Content freshness	Update cycle every 3–6 months	Medium
Featured snippet targeting	40–50 word paragraphs, list formatting	Medium
Core Web Vitals	LCP < 2.5s, CLS ≤ 0.1, INP < 200ms	Medium
Mobile-first	Full functionality on mobile	High
HTTPS	SSL on all pages	High
Alt text	Descriptive, keyword-relevant where natural	Low
Orphan pages	Target: zero	Medium
Heading levels	No skipping (H1 → H2 → H3, never H1 → H3)	Medium
Speakable schema	Tag top 10 informational pages	Low
AI/automation disclosure	Add where users might reasonably expect it	Low

---

Frequently Asked Questions

How long should my content be for good rankings?

Word count is not a direct ranking factor. A Helpful Content Update recovery study of 847 affected pages found recovering pages averaged 1,400 words while non-recovering pages averaged 1,650 words — more words did not help (Upward Engine). Aim for comprehensive coverage of your topic with high information density. Match the depth of what the top-ranking pages provide, but prioritise completeness over length targets.

Does Google penalise AI-generated content?

Not automatically. Google's documentation states it evaluates content by quality and helpfulness, regardless of how it was produced. However, a 16-month study across 4,200 articles found pure AI content had a 3.2× higher deindexation rate following spam updates and a 31% ranking gap vs human-written content at 16 months (Digital Applied). AI-assisted content with substantial editing, original data, and named author attribution performs near-parity with human-written content.

How many internal links should a page have?

Research from a study of 23 million internal links across 1,800 websites found that 45–50 internal links per page correlates with peak organic traffic — with returns diminishing above that threshold (Inblog). For contextual density, target 2–5 links per 1,000 words of body content. Always use descriptive anchor text that reflects the topic of the linked page.

What is E-E-A-T and does it affect rankings directly?

E-E-A-T stands for Experience, Expertise, Authoritativeness, and Trustworthiness. It is a framework from Google's Search Quality Rater Guidelines — not a direct algorithmic signal. Google has confirmed "using a mix of factors that can identify content with good E-E-A-T is useful" for ranking, but there is no single "E-E-A-T score." Practically, implementing E-E-A-T means adding named author bios with credentials, including firsthand experience and specific data in content, earning mentions from reputable sources, and maintaining transparent trust signals.

Should I still optimise for featured snippets now that AI Overviews exist?

Yes — featured snippet optimisation and AI Overview optimisation use nearly identical signals. Both reward direct, concise answers placed immediately after question-phrased H2/H3 headings, in 40–50 word paragraphs. Pages already structured for featured snippets are better positioned for AIO citation. Being cited in an AI Overview delivers 35% higher CTR than non-cited organic results at the same position (IDEAVA).