On-page SEO is the discipline of optimising what is on a web page — and in its underlying code — so that retrieval systems can confidently extract, trust, cite, and recommend it. The 2018 definition (title tags, headings, keyword density) is still partly true, but the surface that on-page work now needs to serve has expanded: Google AI Overviews, ChatGPT search, Perplexity grounding, and Copilot’s Bing-driven retrieval each select sources rather than rank them. Modern on-page SEO is judged on both ranking and selection eligibility, simultaneously.
What on-page SEO actually means in 2026
The 2018 definition of on-page SEO was a list. Title tags. Meta descriptions. Heading hierarchy. Keyword placement. Image alt text. Internal linking. Content structure. All still true. None of it is wrong. But the audience for that work has expanded, and the audience expansion changed the standard.
Until roughly 2023, on-page SEO had exactly one audience: Google’s ranking algorithm. Optimising a page meant signalling relevance and authority to one system that returned an ordered list of ten blue links. Get into the top ten, capture clicks proportional to position, move on.
The audience now includes Google AI Overviews, Google AI Mode, ChatGPT Search, Perplexity, Microsoft Copilot, and a growing list of agentic systems that crawl, index, and cite the open web on behalf of users who never type their query into a traditional search box. Each of those systems uses different extraction logic. Each makes different decisions about which sources to trust, paraphrase, or quote verbatim. None of them publish a complete specification of how they decide. What they share is a hard requirement: the page has to be structurally clean enough that an automated process can lift its claims, attribute them correctly, and reuse them with confidence.
This is not the death of traditional SEO. Google’s traditional ranking system still drives the majority of click-through traffic for most commercial queries. Title tags still influence CTR. Headings still signal structure. Keywords still indicate topical relevance. The discipline has not been replaced; it has been extended.
The distinction worth holding in mind throughout the rest of this guide: AI systems do not rank pages. They select them. Ranking is ordinal — first, second, third, in a list. Selection is binary — either the page is in the cited set or it is not. A page can be ranked #4 organically and never appear in the AI Overview for the same query, because the Overview’s selection criteria are different from the ranking algorithm’s. On-page SEO in 2026 has to serve both surfaces: the ranking surface where ordering still matters, and the selection surface where eligibility is the only currency.
On-page SEO and the four floors of AI visibility
The most useful way to think about where on-page SEO fits in 2026 is through the Four-Floor Model of AI visibility. Each floor represents a distinct dependency in the chain that leads from your page being published to your business being recommended — and ultimately transacted with — by an AI system.
On-page SEO does its primary work on Floor 1 and Floor 2.
Floor 1 (Entity Foundation) is where on-page work declares what the page is and who published it. Page title. H1. Author byline with named credentials. Schema markup naming the publishing organisation and the human author. Canonical URL. These signals tell every system — search engines and AI alike — that this is a real page about a defined topic published by a real entity. Get this floor wrong and nothing above it loads.
Floor 2 (Content Extractability) is where most on-page SEO actually lives. Intent matching. Heading hierarchy that maps to how a user thinks about the topic. Definition-first openings that an AI system can lift and cite. Declarative claims with named sources. Structured data that backs up visible content. Internal linking that signals topical relationships. This floor determines whether a page can be cited, not just ranked.
Floor 3 (Trust and Selection) is partly supported by on-page work but ultimately decided off-page. The author bio, the named-entity treatment, the citation-worthy claims on the page all contribute. But what closes Floor 3 is third-party validation: editorial pickup, earned media, mentions in authoritative sources the page itself does not control. On-page work is necessary for Floor 3; it is not sufficient.
Floor 4 (Agentic Execution) is the floor under construction. AI systems are beginning to act on the user’s behalf — booking, comparing, purchasing — rather than just answering questions. On-page work that prepares a page for agentic interaction (clean structure, semantic markup, accessible forms, machine-readable booking surfaces) is what makes a business eligible when the lift arrives at this floor. Most businesses currently occupy Floors 1 and 2 reliably and reach Floor 3 inconsistently. The temporary ladder to the fourth floor is being built right now; the on-page foundations done well on the lower floors are what determine whether a business can climb when the lift opens.
Where on-page SEO actually lives
The phrase “on-page” is genuinely misleading. It suggests every on-page SEO element is visible on the page itself, which is not the case. A meta description never appears on the page; it appears in the search engine result snippet. Schema markup never appears on the page either; it lives in the underlying HTML where only machines can read it. Canonical tags, robots directives, and Open Graph metadata are all on-page SEO elements that the user never sees.
“On-page” is best understood as “living on this URL’s HTML response” — whether visible to humans or only to machines. The table below clarifies which on-page elements show up where, and which audience each is primarily designed for.
| Element | Where it lives | Who sees it |
|---|---|---|
Page title (<title>) | Source <head>, browser tab text and SERP headline | Users (in tab and SERP), search engines, AI crawlers |
| Meta description | Source <head>, shows as SERP snippet | Users (in SERP only), AI systems as a snippet candidate |
| H1, H2, H3 headings | Visible in page body | Everyone — primary content hierarchy signal |
| Body content | Visible in page body | Everyone — the actual answer to the query |
| Schema markup (JSON-LD) | Source <head> or <body>, invisible to users | Search engines and AI systems only — the machine-readable layer |
| Image alt text | Source <img> attribute, surfaced by screen readers | Visually impaired users, search engines, AI image description |
| Internal links | Visible in page body as anchor text | Users (clickthrough), search engines (crawl signal), AI systems (entity relationships) |
| Canonical tag | Source <head>, invisible to users | Search engines and AI systems (deduplication signal) |
| Robots meta directive | Source <head>, invisible to users | Crawlers (indexing instruction) |
| Open Graph and Twitter Card | Source <head>, invisible on page | Social platforms generating link previews |
| URL structure | Visible in browser address bar and SERP | Everyone |
| Structured data attributes | HTML element attributes (itemprop, aria-*) | Search engines, AI systems, assistive technology |
Two practical consequences follow from this. First: an audit of on-page SEO has to look at both the rendered page and the source HTML, because half the work happens in places the browser doesn’t show. Second: the machine-readable layer (schema, canonical, structured data attributes) has to match what is visibly on the page. Schema claiming a 4.9-star rating when no reviews are visibly displayed is a Google manual action waiting to happen, and AI systems treat the mismatch as a trust failure.
On-page SEO by audience
What on-page SEO means in practice depends on who is reading. An enterprise procurement team and a Southampton sole trader both need on-page work done well, but the priority order is materially different. Use the four sections below to jump to the audience that matches your situation.
Enterprise
You want governance, scale, and risk reduction across thousands of pages.
The on-page challenge at enterprise scale is consistency, not invention. Title tag templates that work for one content type fail for another. Schema architecture that handles legal pages doesn’t map cleanly to product pages. Internal linking that distributes authority across a 50-page site falls apart at 5,000 pages. The work is programmatic consistency, schema architecture standardisation, governance for AI-readable content, structured data quality control, and internal link equity distribution that doesn’t collapse under its own weight. Brand integrity matters at scale: one section publishing AI-generated thin content can devalue an entire domain’s citation signal across AI systems. See the CITATE framework for the operational standard, and schema architecture for the technical foundation.
B2B SaaS
You want comparison-intent pages that win against competitor-owned content.
B2B SaaS prospects research extensively before contacting a vendor. Comparison pages (“Vendor X vs Vendor Y”), alternative pages (“alternatives to Vendor X”), integration pages, and feature pages do disproportionate work in the funnel because they intercept the prospect at the comparison stage. On-page SEO for SaaS is about owning the specific decision queries: making sure your comparison content is structurally clean enough to be cited by ChatGPT, your integration documentation is detailed enough to be referenced by Copilot, and your ROI calculators generate engagement signals that compound into citation eligibility. Coviant Software’s ROI calculator and competitor alternative pages generated 200+ enterprise leads attributable to the calculator-anchored on-page architecture. Further reading: SaaS SEO.
B2B services (law firms, professional services, consultancies)
You want named-author authority that prospects can verify before they ever fill in your contact form.
B2B services buyers do extensive pre-purchase research because the trust threshold is high and the cost of a wrong choice is significant. On-page SEO for this audience is essentially trust architecture: named-author bylines with verifiable credentials on every substantive page, Person schema declaring who the authors actually are, service pages structured around the specific decisions a prospect is trying to make (not generic descriptions of what the firm does), FAQ content that answers what a prospect would ask in the consultation. Internal linking reinforces practice-area authority by clustering related content under named expertise. Olliers Solicitors’ Operation Soteria page ranked #1 organic within days of publication and earned a Google AI Overview citation with the Olliers logo in the right-panel sources; the mechanism was named author with verifiable credentials, standalone extractable opening, and structured content meeting all six CITATE criteria. Further reading: law firm SEO.
Local SME and B2C
You want to be the answer Google shows when someone searches your service plus your town.
Local commercial intent is decided in seconds. The on-page priorities for a local business are service pages that mention the geographic area naturally (not as keyword stuffing), LocalBusiness schema declaring location and category clearly, fast mobile performance because most local discovery happens on a phone, and content structured to be pulled into both the local pack and AI Overview answers. On-page SEO for local SMEs matters less for the local pack itself (the Google Business Profile dominates that surface) and more for organic visibility on commercial-intent queries like “boiler repair Southampton” or “family solicitor Winchester”. See SEO agency Southampton for the buyer-tier breakdown showing where on-page work fits in the wider channel mix, and local business SEO for the broader local strategy.
Worked example: B2B intake friction
The clearest way to see what on-page SEO actually does is to follow a real buyer journey. Consider a corporate prospect who needs to instruct a specialist solicitor on a complex matter — say, a serious fraud investigation. The prospect knows the type of firm they need. They search Google for “serious fraud solicitor London”.
The firms that rank on page 1 all look broadly similar: a service page describing “fraud and financial crime”, a generic contact form, a careers page, a couple of news articles, an “our team” page with photos. None of them name a specific lawyer the prospect can identify with the matter. The prospect, who has already made several decisions about who they trust before they ever clicked, opens a second tab and searches LinkedIn for individual fraud solicitors at each firm. They find a named partner who has handled a similar matter, message them on LinkedIn, and bypass the firm’s contact form entirely.
The firm lost the attribution. The on-page SEO drove the discovery, but the conversion happened on a different surface, in a way the firm cannot measure and cannot replicate at scale. The intake form is a friction layer the prospect routed around because the on-page content gave them no reason to use it.
What on-page SEO could have done differently:
- Named-lawyer pages with Person schema, biographical detail, areas of practice, and verifiable credentials
- Practice-area pages that internally link to the specific lawyers who handle each matter type
- Calendar booking embedded per lawyer, removing the “wait for a callback” friction
- FAQ content answering “what happens at the consultation” so the prospect arrives prepared
- Case study content (anonymised where required) demonstrating the type of work each lawyer actually does
Each of those is on-page work. None of them are exotic. Together they convert the search-intent prospect at first contact instead of losing them to a workaround. They also do something AI systems care about: they make the firm’s named expertise extractable. A query to ChatGPT or Perplexity asking “who is a specialist serious fraud solicitor in London” now has structurally clean source material to draw from, with named entities, verifiable credentials, and dated case experience. The same on-page work that fixes the intake friction also makes the firm recommendable.
The on-page SEO elements that matter most in 2026
What follows is the practical list. Twelve elements, each treated as plainly as possible. The order is rough priority for most commercial pages; specific audiences (per the segmentation above) may reorder for their context.
1. Page titles
The page title is the single highest-leverage on-page element. It is the SERP headline, the browser tab text, and the most common source of social link previews. AI systems use it as a primary signal of what the page is about and frequently quote it in citations. Keep titles under 60 characters where possible to avoid SERP truncation. Lead with the topic the page actually covers, not the brand name. Avoid duplicate titles across the site — identical titles signal cannibalisation and dilute click-through rate. Match the title to the search intent; a comparison query needs a comparison-format title.
2. Meta descriptions
Meta descriptions are not a ranking factor. They are a click-through factor and an AI citation signal. ChatGPT specifically uses the meta description as one of the inputs when selecting which source to cite for a response. Keep them under 160 characters, expand on what the title promises rather than repeating it, and write them for the human who has just scanned the SERP and is deciding which result to click. Generic meta descriptions get rewritten by Google to match the query; specific, well-written ones survive intact more often.
3. Heading hierarchy (H1 to H6)
One H1 per page, used for the main page title. H2s for major sections. H3s for sub-sections within an H2. The logical hierarchy is more important than the specific words. Headings serve three audiences: human readers scanning to find the relevant section, screen readers navigating accessibly, and AI systems segmenting the page into extractable chunks. Descriptive headings outperform clever ones — an H2 that reads “Where on-page SEO actually lives” tells both the reader and the AI system what the section covers; an H2 that reads “The hidden architecture” tells neither.
4. URL structure
Short, descriptive, lowercase, hyphenated. Avoid dates in evergreen URLs (locks the content into a specific year and looks stale on update). Avoid IDs and query parameters where the content is meaningful (/blog/123 tells a user nothing). Avoid changing URLs after publication unless the redirect strategy is properly handled; broken canonical history is one of the most common ranking regressions following site rebuilds. The URL itself is a weak ranking signal but a strong user-comprehension signal — it appears in the SERP, in shared links, and increasingly in AI citations.
5. Content depth and intent matching
Word count is not a ranking factor; intent matching is. An informational query (“what is X”) needs definition, context, and breadth. A commercial query (“best X for Y”) needs comparison, structured options, and a recommendation. A navigational query (“X support contact”) needs the answer fast and visible. A transactional query (“buy X”) needs trust signals, pricing, and a clear purchase path. The same word count produces wildly different outcomes depending on whether the content matches the intent of the query that brought the visitor.
6. Internal linking
Internal links distribute authority across a site, signal topical relationships to both search engines and AI systems, and guide users toward related content. The two consistent failures: linking with generic anchor text (“click here”, “read more”) which gives zero topical signal, and linking inconsistently so high-priority pages end up undersupported relative to their commercial value. Anchor text should be specific and natural — describe what the linked page is about, vary the phrasing across instances, and avoid exact-match keyword stuffing. The internal link is also a citation signal for AI systems building entity relationships across a domain.
7. Schema markup
Schema is the machine-readable layer of on-page SEO. FAQPage schema lost its rich result in May 2026 but remains valid structured content that AI systems parse. HowTo schema lost its rich result earlier in September 2023 with the same machine-readable status. Article schema declares authorship and publication date. Person schema names authors with their credentials. LocalBusiness schema declares geographic relevance. The fundamental rule: schema must match the visible content of the page. Claiming a rating in schema without displaying review content on the page is a Google manual action category. See the schema markup guide for the implementation specifics across platforms.
8. Image optimisation and alt text
Alt text exists primarily for accessibility — visually impaired users rely on it to understand images via screen readers. Search engines and AI systems use it as a secondary signal for image content. Keep alt text under 125 characters, describe the image accurately, include relevant context (not keyword stuffing), and leave decorative images with empty alt attributes rather than fake descriptions. Compress images for performance: WebP for general use, JPEG for photography, PNG for transparency. Descriptive filenames (commercial-glazing-installation.jpg not IMG_4593.jpg) reinforce topical relevance for both search and AI image indexing.
9. Mobile and Core Web Vitals
Technically Core Web Vitals sit in technical SEO rather than on-page SEO, but the boundary is artificial — a page that loads slowly, shifts layout on render, or fails on mobile is on-page-broken regardless of how the elements are categorised. The three metrics: Largest Contentful Paint under 2.5 seconds, Interaction to Next Paint under 200ms, Cumulative Layout Shift under 0.1. Google uses mobile-first indexing for almost all sites, so the mobile rendering is the rendering that matters. Speed and stability are page experience signals that Google has confirmed as ranking inputs.
10. Content freshness
Some content benefits from regular updates; some doesn’t. Time-sensitive topics (algorithm changes, software versions, regulatory updates) need dated content with explicit revision history. Evergreen topics (foundational explainers, methodology guides) can hold value for years without updates as long as the underlying claims remain accurate. The failure mode is dressing evergreen content in artificial datestamps (“2026 ultimate guide to X”) when nothing meaningful has changed — users notice, AI systems weigh the substance not the headline. Genuine updates with substantive revisions are worth marking; cosmetic ones aren’t.
11. E-E-A-T signals (Experience, Expertise, Authoritativeness, Trustworthiness)
Google’s quality rater guidelines describe E-E-A-T as a measurement framework for content quality, particularly on YMYL (Your Money or Your Life) topics where the stakes of bad information are high. Named authorship with verifiable credentials. Evidence of hands-on experience. Original photography rather than stock imagery. Specific case examples rather than generic claims. Author bios with links to professional profiles. For B2B services and YMYL content (legal, medical, financial), E-E-A-T signals are functionally a prerequisite for AI citation eligibility — systems consistently weight named expertise over anonymous content.
12. Structured answers and declarative openings
The single highest-leverage on-page intervention for AI citation eligibility is a standalone, extractable answer to the page’s primary query in the first 100 words. AI systems generating responses scan the opening of candidate sources for an extractable claim. A page that opens with a definition or declarative answer is materially more citable than one that opens with context-setting prose. The opening sentence should answer the query directly, in a form that could be quoted standalone without surrounding context. The rest of the page can develop the nuance.
How to do on-page SEO properly
Five steps. The order matters because each step builds on the previous one. Skipping the first two and starting at step three is the most common reason on-page SEO work fails to compound.
Common on-page SEO mistakes
The six failure patterns that cause measurable damage to organic visibility and AI citation eligibility. Each looks like best practice and is actively counterproductive.
Keyword density chasing
Still happening in 2026 despite a decade of guidance against it. Tools that report a “keyword density” figure as a ranking metric are measuring nothing useful. The page either matches the query intent semantically or it doesn’t; repeating the keyword five extra times doesn’t move the needle and may trigger over-optimisation signals. Write naturally; cover the topic; the keyword density takes care of itself.
Templated FAQ schema on every service page
Common pattern: three or four generic FAQ questions bolted onto every service page to generate FAQ rich results. The rich result was deprecated for most commercial sites in 2026. The schema is still valid markup that AI systems parse; the rich result is gone. A templated three-question FAQ that exists only to capture SERP real estate now produces no SERP feature and looks structurally weak to AI systems comparing genuinely useful FAQ content. Keep FAQ content where it answers real prospect questions; remove it where it was only there for the rich result.
AI-generated content with no expertise grounding
Google’s helpful content updates specifically target thin, generic, AI-paraphrased content that adds no original value to the existing corpus. The diagnostic: if the article could have been written without anyone at the publishing organisation having direct experience of the topic, AI systems treating other AI systems as primary sources will eventually create a citation collapse. Original research, case examples, named expertise, and dated evidence are what differentiate citable content from AI noise.
Identical title tags across multiple pages
Caused by templating systems that generate Service Name | Brand Name for every service variant. Multiple URLs with identical titles cannibalise each other in the SERP — Google has to choose one to display and the others lose impression share. Each title should reflect the specific page’s topic. Where pages genuinely cover the same topic, consolidate them rather than competing against yourself.
Internal links with “click here” anchor text
Every internal link is a topical signal: this page is related to that page, and the anchor text describes the relationship. “Click here”, “read more”, and “learn more” describe nothing. They waste the topical signal entirely and produce no benefit. Replace with anchor text that describes what the linked page is about. Accessibility benefits too: screen reader users navigating by link list get useful destinations instead of a list of identical “click here” entries.
Schema markup that contradicts visible content
Most common variants: AggregateRating in schema with no visible review content, prices in schema with no visible price on the page, FAQ questions only present in JSON-LD with no equivalent visible content. Google treats the mismatch as a manual action category and will remove the rich result. AI systems treat it as a trust failure and weight the page lower in selection. The rule: every schema claim must be visibly substantiated on the page. If the visible content doesn’t back it up, the schema shouldn’t be there.
