On-page Optimization Guide

On-page optimization is everything you control inside a single page — the HTML, the content, the structure, and the signals that tell Google and AI search engines what the page is about and how well it answers a query. It’s not a fixed percentage of SEO work; it’s the foundation everything else builds on. Off-page signals (links, brand mentions, E-E-A-T) compound the work of a well-optimized page, but they rarely save a badly-optimized one. This guide covers what on-page SEO actually means in 2026 and the specific elements worth getting right.

What counts as “on-page” in 2026

On-page SEO spans three layers:

1. Content — the words, images, and media on the page. Topic completeness, user-intent match, and E-E-A-T signals live here.
2. HTML elements — title tag, meta description, headings, alt text, canonical, structured data, Open Graph, hreflang. All the markup Google and AI crawlers read directly.
3. User-experience signals — Core Web Vitals, mobile rendering, accessibility, and the interaction quality that determines whether people stay on the page or bounce.

The modern ranking systems — Google’s Helpful Content classifier (integrated into core 2024), E-E-A-T signals, BERT, MUM, and the AI Overview generation stack — all read signals from these three layers. Get them aligned with actual user intent and rankings follow. Stuff them with keywords and hope for the best, and you’ll lose ground.

Title tag

The title tag (<title>) is still one of the most important on-page signals, because it’s both a ranking input and the clickable link in search results. Modern best practice:

• Keep it under 60 characters — Google’s desktop SERP truncates around 580 pixels, which lands near 50–60 characters for most fonts. Longer titles get trimmed with an ellipsis.
• Lead with the primary topic — put the most important words near the start, where they’re most visible and carry the most weight.
• Write for clicks, not just crawlers — a compelling title that earns CTR consistently outperforms a keyword-stuffed one, because CTR in SERPs is itself a ranking signal.
• Match user intent — a how-to query wants a how-to title; a comparison query wants a comparison title. Mismatched intent hurts rankings even with otherwise perfect on-page signals.
• Include the brand where it fits — usually at the end (“How to Fix X | YourBrand”). Some high-authority brands lead with it; most should trail.
• Make each title unique. Duplicate titles across pages tell Google you don’t know which page should rank for the topic.

Google sometimes rewrites the title shown in search results using the page’s H1, internal link anchor text, or on-page headings if it judges the original title unhelpful. That rewriting was formalized in 2021 and continues today — so even if you write a perfect title, the best insurance is consistency between the <title>, the H1, and the opening content.

Meta description

The meta description isn’t a direct ranking factor, but it’s the preview text shown under the title in search results — and therefore a direct lever on click-through rate, which is a ranking signal. Current guidance:

• 120–155 characters fits desktop SERPs consistently; longer descriptions get truncated. Mobile SERPs are tighter (~120 characters).
• Include the primary keyword naturally — Google bolds matching query terms in the snippet, which increases CTR.
• Write it as a promise or value statement — describe the value the reader gets if they click, not just a summary of the page.
• Unique per page. Duplicate meta descriptions are flagged in Search Console’s HTML enhancements report.
• Accept rewrites. Google rewrites meta descriptions about 60–70% of the time based on the query. Write one anyway — when Google uses yours, you get the framing you chose.

Header hierarchy (H1–H6)

Headings do three jobs: they outline the page for scanning humans, they signal topic structure to crawlers, and they help AI summarizers extract quotable sections for answers.

• Exactly one H1 per page, describing the page’s primary topic. Should closely match the <title> but doesn’t need to be identical.
• H2s structure the main sections. Think of them as chapter titles; each should describe a complete subtopic a reader might search for independently.
• H3s handle sub-sections under H2s. Don’t skip levels (an H2 followed by an H4 breaks the outline).
• Use natural, descriptive language. “How to set up your account” reads better than “Account Setup” and gives AI Overviews something to quote.
• Include keywords in headings where natural. Forced keyword insertions hurt readability and rarely improve rankings; natural use helps both.

URL structure

URLs are read by crawlers, AI models, and users. A good URL is short, descriptive, and stable:

• Short and meaningful: /guide/on-page-seo beats /p?id=8472.
• Hyphens between words. Underscores don’t count as word separators; hyphens do.
• Lowercase, unreserved characters. Stick to letters, digits, hyphens, periods, and tildes per RFC 3986 — anything else needs percent-encoding.
• Stable URLs survive redesigns. Every URL change costs rankings during re-crawl; 301-redirect the old URL to the new one to preserve link equity.
• Avoid deep nesting. Pages 4+ folders deep are harder to discover and crawl; see our website architecture guide for more on site structure.

Content: helpful, specific, trustworthy

Google’s Helpful Content system — standalone 2022, integrated into the core algorithm in March 2024 — explicitly demotes content that reads as written-for-search rather than written-for-humans. The ranking signals it uses for “helpful” include:

• First-hand experience. The “E” in E-E-A-T — Experience, Expertise, Authoritativeness, Trust — was added in December 2022. Content that demonstrates actual use (“we tested this across 12 sites”), original data, and specific details outperforms generic coverage.
• Topic completeness. Does the page genuinely answer the question, or does it circle around the topic while padding word count? Cover what a real user needs — no more, no less.
• Clear primary purpose. Is the page obviously for users, or obviously for ranking? Auto-generated content, thin content, and keyword-stuffed pages are demoted by design.
• Trust signals. Author bylines with credentials, publication dates, updated dates, sources cited, contact information, clear site ownership. For YMYL topics (health, finance, legal), these signals carry more weight than they do for casual content.

The shift away from keyword-centric SEO toward topic-and-intent-centric SEO accelerated with the launch of BERT (2019), MUM (2021), and AI Overviews (2024). Modern on-page content should target a question or need, cover it with specific and credible information, and structure the answer so both humans and AI summarizers can extract value.

A note on “keyword density”

Old guides recommending “1–2% keyword density” (or “3% for Yahoo”) have no basis in Google’s guidance, and Yahoo hasn’t been a distinct search engine since 2009. Google’s John Mueller has repeatedly confirmed keyword density is not a ranking factor. The modern signal Google actually uses is topical coverage — whether your page semantically addresses the query and related subqueries — not how many times a specific phrase appears. Use natural language and include the primary keyword and related terms where they make sense.

Image SEO

Images are on-page elements with real ranking and accessibility implications:

• Alt text describes the image for screen readers and crawlers. Write it as a real description of what the image shows, with natural keyword inclusion where appropriate. Decorative images (pure background, presentational) should use empty alt="" so assistive tech skips them.
• Filename should be descriptive and hyphenated: on-page-seo-checklist.webp, not IMG_4729.jpg.
• Modern formats — WebP (universal browser support since ~2020, 25–35% smaller than JPG) for most web images; AVIF (even smaller) for hero images; SVG for logos and vectors; PNG only when you need lossless transparency.
• Explicit width and height attributes prevent Cumulative Layout Shift (a Core Web Vital) as images load.
• Native lazy loading (loading="lazy") on off-screen images — supported in every evergreen browser since 2020.
• Responsive images via srcset and <picture> serve appropriately-sized images per viewport.

Internal linking and anchor text

Internal links are the single most under-used on-page SEO tool. They distribute PageRank inside your site, tell Google which pages are most important, and help AI summarizers discover related content to cite.

• Link contextually within body text — a link inside a paragraph carries more weight than the same link in a footer.
• Use descriptive anchor text. “our crawl budget guide” beats “click here” because the anchor text itself is a topic signal. Avoid over-optimizing exact-match anchors; natural variation is fine.
• Build topic clusters. A pillar page on a broad topic links out to narrower cluster pages, and each cluster links back to the pillar. This distributes authority and signals topical expertise.
• Keep important pages shallow. Aim for 3 clicks or fewer from the homepage for any page you care about ranking.
• Implement breadcrumbs with BreadcrumbList structured data so they appear in search results and AI-generated answers.

Outbound links and link attributes

Linking out to authoritative sources is a trust signal — it shows you did your research and gives readers a path to deeper information. Don’t avoid outbound links thinking you’re “hoarding link equity”; that’s not how modern PageRank flow works.

Google supports four link attributes, significantly updated in September 2019:

• Default (no attribute) — a normal “dofollow” link that passes PageRank.
• rel="nofollow" — tells Google not to follow the link for ranking purposes. Since March 2020, Google treats this as a hint rather than a strict directive.
• rel="sponsored" — introduced September 2019 for paid/affiliate links. Required by Google for FTC-disclosed partnerships.
• rel="ugc" — introduced September 2019 for user-generated content links (forum posts, blog comments).

Multiple attributes can stack: rel="sponsored nofollow" is valid. Use the attribute that matches the link’s nature, and don’t stress about “link juice” — focus on linking to sources that genuinely help the reader.

Canonical tags

The <link rel="canonical"> tag tells Google which URL is the authoritative version when duplicate or near-duplicate pages exist (filter URLs, UTM-tagged URLs, trailing-slash variants, paginated pages). In 2026:

• Self-reference every indexable page — every page should include a canonical pointing to its own clean URL.
• Use absolute URLs (https://example.com/page), not relative ones.
• Don’t canonical every paginated URL to page 1 — that tells Google the other pages are duplicates and cuts off the links underneath. Each paginated URL should canonical to itself.
• Don’t point canonicals across different content. The canonical page should be genuinely the same or very similar to the source; otherwise Google ignores the tag.

Structured data (schema.org)

Structured data is metadata that tells search engines exactly what a page is about and how its content is organized. In 2026 it powers rich results, AI Overview citations, and voice-search extraction. Common schemas by content type:

• Article / BlogPosting — headline, datePublished, dateModified, author, publisher. Table stakes for editorial content.
• FAQPage — for pages with a FAQ section. Can win featured-snippet real estate.
• Product — for e-commerce pages (name, price, availability, review).
• Recipe, Event, Video, HowTo — for their respective content types.
• BreadcrumbList — on every non-homepage.
• Organization — on the homepage. Name, logo, sameAs social links, contactPoint.

Deliver structured data as JSON-LD in the page <head> (Google’s preferred format). Test with the Rich Results Test — the successor to the Structured Data Testing Tool (deprecated August 2021).

Open Graph and Twitter cards

These meta tags don’t affect Google rankings directly, but they control how your page looks when shared on social platforms — which drives click-through and referral traffic. At minimum, set og:title, og:description, og:image, og:url, og:type for Open Graph, plus twitter:card (typically summary_large_image) and matching Twitter meta. Share images should be 1200×630 pixels (minimum 600×315). A missing OG image is the most common cause of ugly-looking social shares.

Core Web Vitals

Three user-experience metrics are formal ranking signals. Target these at the 75th percentile of real-user measurements:

• Largest Contentful Paint (LCP) — under 2.5 seconds. Time until the biggest above-the-fold element renders.
• Interaction to Next Paint (INP) — under 200ms. Responsiveness to user input. INP replaced First Input Delay (FID) in March 2024.
• Cumulative Layout Shift (CLS) — under 0.1. How much content shifts as the page loads.

Measure field data (real users, not lab scores) in Search Console’s Core Web Vitals report. Our page speed tools guide covers the measurement stack in detail.

Mobile-first rendering

Google completed mobile-first indexing for the entire web in September 2020. The mobile version is what Google crawls and ranks — and AI crawlers follow the same pattern. If content or markup differs between mobile and desktop, the mobile version wins. The universal 2026 answer is responsive design (one HTML, CSS adapts per viewport) rather than separate mobile sites or adaptive serving.

Optimizing for AI Overviews

Since May 2024, Google has rolled out AI Overviews across the US and beyond. Pages cited in AI Overviews earn roughly 35% more organic clicks than uncited pages — making “being the source” a new on-page SEO objective. Practical moves:

• Answer the query directly early in the page. AI summarizers prefer pages that answer the question in the first 1–2 paragraphs.
• Use clear headings and structure. Extractable chunks are easier to cite.
• Back claims with specifics and sources. AI models weight fact-dense, sourced content.
• Include FAQ sections with FAQPage schema. Featured-snippet and AI Overview extraction both reward well-structured Q&A.
• Keep freshness current. Update dateModified when content genuinely changes; AI Overviews tend to favor recent sources for time-sensitive queries.

On-page SEO checklist

Before hitting publish:

1. Title under 60 characters, primary keyword near the front, compelling to click.
2. Meta description 120–155 characters with clear value promise.
3. One H1; H2s structure major sections; H3s for sub-sections.
4. URL short, lowercase, hyphenated, descriptive, stable.
5. Content answers a real need with specific, sourced, first-hand information.
6. Natural keyword use — no density targets.
7. Images with alt text, modern format (WebP/AVIF), explicit dimensions.
8. Internal links with descriptive anchor text; breadcrumbs with schema.
9. Outbound links to authoritative sources, appropriate rel attributes.
10. Self-referencing canonical with absolute URL.
11. Structured data in JSON-LD (Article, FAQPage, BreadcrumbList, Organization).
12. Open Graph and Twitter meta tags with 1200×630 share image.
13. Core Web Vitals pass at 75th percentile (LCP, INP, CLS).
14. Mobile rendering verified via URL Inspection.
15. AI Overview readiness: direct answer early, clear structure, cite-worthy specifics.

Frequently asked questions

Is keyword density still a ranking factor?

No — and it hasn’t been for years. Google has repeatedly said there is no target density. Modern ranking systems assess topical coverage and semantic relevance rather than word-frequency ratios. Write naturally, include the primary keyword and related terms where they make sense, and forget the percentages.

How long should a page be for good on-page SEO?

There’s no fixed length. Match the depth to the query: informational queries that need thorough coverage get 1,500–3,000 words; transactional or navigational queries get far less. The top-ranking pages for your target query are a reasonable benchmark, but length alone doesn’t rank — quality and completeness do.

Does Google punish thin content?

The Helpful Content system (integrated into core algorithm 2024) demotes content that offers little value or reads as written-for-search-engines. “Thin” in the modern sense isn’t about word count alone — it’s about whether the page genuinely helps a user. A 400-word answer that solves the problem beats a 2,000-word page that pads around it.

Should I add FAQ schema to every page?

Only where you have a genuine FAQ section answering real questions. Misusing FAQ schema to win SERP real estate was a Google target in the August 2023 update — sites using FAQ schema on pages without actual FAQ content saw rich results disappear. Keep it honest.

How often should I update on-page optimization?

Audit pillar and top-traffic pages every 6–12 months. Update the dateModified when you genuinely update content; don’t bump the date without changes. For fast-moving topics (tools, prices, statistics, Google guidance), quarterly checks are appropriate. Stable evergreen content can go longer.

What matters more: on-page SEO or backlinks?

Both, for different reasons. On-page SEO determines whether your page can rank — whether Google can understand it, whether it genuinely addresses the query, whether the user experience is adequate. Backlinks (and broader E-E-A-T signals) determine how competitively it ranks. Strong on-page with weak backlinks tends to rank in long-tail, low-competition queries. Weak on-page with strong backlinks often ranks but underperforms what it could. Both matter.

Bottom line

On-page SEO in 2026 is about aligning content, markup, and user experience so that Google, AI summarizers, and real users all reach the same conclusion about what your page is and who it’s for. The fundamentals haven’t changed — good titles, good meta descriptions, clean headings, useful content, sensible internal links — but the 2016 version of this work (keyword density, exact-match stuffing, sitemaps and subdomains listed as on-page elements) is actively counterproductive now. Write for the person searching, mark up what you wrote so machines can parse it, measure the experience with Core Web Vitals, and layer in structured data and E-E-A-T signals. That’s the whole playbook.