A crawler reads your page top to bottom. So does the extractor that builds an AI Overview. Most on-page SEO guides ignore that order and give you a flat checklist of "factors". This one walks the page the way a machine does — URL down to verification — with the real HTML, schema and code at every layer.
On-page SEO is the set of changes you make on a single page — content, HTML, structure, performance — to help it rank in search and be cited by answer engines. It sits between technical SEO (the plumbing — crawlability, sitemaps, redirects) and off-page SEO (the links and brand mentions you earn elsewhere). In 2026 the job has two outputs: win the classic SERP, and get extracted by AI Overviews and AI Mode. The same nine layers serve both, but the optimisations differ.
The page we're walking through
Take a typical service page targeting the keyword seo audit for a small UK marketing agency. It lives at /seo-audit/. Domain authority is modest. The page currently sits on the bottom of page two. Below is roughly how it appears in the SERP today — and how the same nine layers, fixed, would lift it.
SEO Audit Services | Example Marketing
We provide professional SEO audit services for businesses. Contact us today for a quote.
What Is On-Page SEO? — Moz
Learn how to optimize a page for search using on-page SEO best practices.
On-Page SEO: The Definitive Guide + FREE Template (2026)
The complete guide to on-page SEO in 2026 with a downloadable checklist.
The example.co.uk listing is the running case. Three of its nine layers are doing real work. The other six are vague at best, broken at worst. Let's walk them.
1. The URL
A clean URL is short, lowercase, hyphen-separated, and uses the primary keyword phrase once. /seo-audit/ already does that. The two things to actively get right at this layer in 2026: the canonical declaration and the slash policy.
If your site serves a page under more than one URL (with and without trailing slash, with and without ?utm params, on www and non-www), declare which one wins. AI engines treat canonicalisation as a trust signal — pages that contradict themselves don't get cited.
<link rel="canonical" href="https://www.example.co.uk/seo-audit/" />
Pick a trailing-slash policy and apply it consistently across canonical, sitemap, internal links, and the Location: header of any 301. Mixed slash usage is the single most common low-effort on-page bug.
2. The <title> tag
Still Google's strongest on-page signal. Lead with the keyword phrase, end with the brand, write for the dominant intent of the SERP. Google rewrites a large share of titles it finds in HTML when they don't match intent — losing control of your title is a sign Google disagrees with your framing.
SEO Audit Services | Example Marketing
SEO Audit — Live SERP Review in 7 Days | Example Marketing
The "after" version includes the keyword, signals what the reader gets (a live SERP review), and gives a concrete time-to-deliverable. It's also under 60 characters so it doesn't get truncated in the SERP.
<title>SEO Audit — Live SERP Review in 7 Days | Example Marketing</title>
3. The head block — meta, OG, schema
Three things matter inside the <head> after the title: the meta description, the Open Graph block, and the structured data. In 2026 the schema is the most important of the three because it's the part AI engines read most closely.
The meta description is not a ranking factor on its own, but it owns the click-through rate. Write it for the searcher's question, include the keyword once, stay inside 155–160 characters.
<meta name="description" content="A 7-day SEO audit covering technical, on-page, and content gaps. Live SERP review, prioritised fix list, and the actual code to ship." />
For the structured data, ship at least Article schema (or Service schema for service pages) plus FAQPage schema if the page answers questions. FAQ schema is the highest-leverage single addition for AI Overview citation — it gives answer engines structured Q&A pairs they can quote directly.
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [
{
"@type": "Question",
"name": "What does an SEO audit cover?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Technical crawl, on-page review of the top 30 templates, content gap analysis against the live SERP, and a prioritised fix list with the HTML to ship."
}
},
{
"@type": "Question",
"name": "How long does an SEO audit take?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Seven working days end-to-end. Day one is the crawl, day seven is the readout call."
}
}
]
}
</script>Validate every schema block in Google's Rich Results Test before it ships. A broken schema fails silently — the page still renders, but the rich result and the AI citation opportunity both disappear.
4. The first 50 words
Forty-four percent of the citations an LLM hands out come from the first 30% of a page's text. The implication is unambiguous: the opening paragraph is the most load-bearing prose on the page. Write it as a direct, standalone answer to the page's title question — no setup, no history, no "in this article we'll cover".
The example page's current opener — "We provide professional SEO audit services for businesses…" — fails this. It tells the reader nothing they didn't already know. Replace it with: "An SEO audit is a structured review of the technical, on-page and content layers of your site, producing a prioritised list of fixes ranked by their effect on ranking. Ours runs over seven days."
5. The H1 and the heading hierarchy
One H1 per page, written close to a phrase real people type. The H1 is the strongest topic signal after the <title> and is what AI engines treat as the canonical name of the page's subject. Below the H1, use H2s as the questions a reader would actually ask — pull them from the People Also Ask block before drafting.
<h1>SEO Audit Services for UK SMBs</h1> <h2>What does an SEO audit cover?</h2> <h2>How long does an SEO audit take?</h2> <h2>What does the deliverable actually look like?</h2> <h2>How much does an SEO audit cost?</h2>
Each H2 phrased as a real PAA question doubles as an FAQ schema candidate and an extractable answer block for an AI Overview. The first sentence under each H2 should answer the H2 in 30–50 words.
6. The body
The body of the page is where most of the visible value lives, and where most pages waste it. Three rules, in order of impact.
Lead every section with the answer. Inverted pyramid. The reader who scans only the first sentence under each H2 should still know what you're claiming. Saving the verdict for the last paragraph is a habit from a previous decade of SEO writing — it actively hurts AI Overview citation now.
Use scannable units. Short paragraphs (2–4 sentences). Bullet lists where three or more items are parallel. Tables for any comparison of three or more attributes. The extraction logic in AI Overviews favours dense, structured chunks over wandering prose. Walls of text don't get cited.
Cite outbound. When you make a claim, link to the source. Sites that cite their sources are cited more often in return — AI engines treat outbound citation the same way humans treat referenced research papers, as a trust signal. Don't pad citations; one credible primary source beats five blog-roll links.
Core Web Vitals is the floor, not a tiebreaker
Pages that fail Core Web Vitals get filtered out of ranking consideration before the body content is evaluated. Target LCP under 2.5s, FCP under 1s, INP under 200ms — and test on a 3G mobile profile in Chrome DevTools. Fixing speed beats every other on-page change on a slow site.
7. Internal links
Three or four contextual internal links per piece, capped at roughly one per 200–300 words. Each one should be the next thing the reader would naturally click — not a dump of "related services" at the bottom. Internal links distribute equity and help AI engines map the topical neighbourhood of your site.
<p>An audit is the first step in any SEO engagement. The next is execution — see how we <a href="/seo-services/">deliver on-page SEO</a> once an audit identifies the priorities.</p>
Anchor text should describe the destination, not the source. "Click here" and "learn more" are wasted signal. "On-page SEO services" tells the crawler what the destination is about.
8. Images
Every image earns its place by adding information the prose can't. Diagrams, screenshots, original photography. Stock photos of teams shaking hands are content fluff and AI engines treat them as such. At the markup layer, four things to get right.
<img src="/img/seo-audit-deliverable.webp" alt="A sample page from an SEO audit report showing the prioritised fix list and technical scores" width="1200" height="800" loading="lazy" />
The alt describes function, not appearance. width and height prevent cumulative layout shift. loading="lazy" defers off-screen images. The .webp extension (or .avif) compresses better than .jpg for the same visual quality. None of these is new in 2026 — they're just the rules nearly every CMS still ships broken.
9. The verification step
The last layer of on-page SEO is the one most teams skip: checking that machines see what you think they see. Three commands, in order. Run them on every page you publish, the day you publish.
# 1. Does the page render server-side? curl -sA "Googlebot" https://www.example.co.uk/seo-audit/ | grep -E "<h1>|<title>" # 2. Does the page pass Core Web Vitals on mobile? npx unlighthouse --site https://www.example.co.uk/seo-audit/ # 3. Does the schema validate? open "https://search.google.com/test/rich-results?url=https://www.example.co.uk/seo-audit/"
If the first command returns empty tags, your page is JavaScript-rendered and AI crawlers are seeing an empty shell. Brief your dev team on server-side rendering before doing anything else on-page. If the Lighthouse score fails, fix speed first. If the schema test errors, fix the JSON-LD before re-deploying.
Where to start tomorrow
Pick one page that should be ranking better than it is. Walk it through the nine layers above, in order. Stop at the first layer that fails — fix that one — and ship. Most pages reveal three or four fixable issues in under an hour, and a single corrected title or schema block can move rankings inside a fortnight.
"On-page SEO has always been about helping your content appear in search results. Today, the same fundamentals boost your visibility in agentic search — where AI tools scan the web, pull sources, and generate answers for users.
SourceSemrush blogSemrushCategory leader, on its own social channels
The fundamentals didn't move. The audience did — and half of it now reads HTML instead of prose.
FAQ
Does the order of elements inside the head matter?
Not for ranking, but it matters for resilience. Put <meta charset> first, then <title>, then <meta name="viewport">, then everything else. Some browsers stop parsing the <head> on the first script that crashes, and you want your title and meta description to land before that point.
Should canonical URLs include or exclude the trailing slash?
Either is fine — pick one and apply it consistently across the canonical tag, the sitemap, every internal link, and the destination of any 301. Mixed slash usage tells search engines you don't know which page is canonical, which is a quiet ranking dampener on most CMSs.
Where should I place FAQ schema — head or body?
Either works. The <head> is the cleaner default for purely structured data. The <body> is fine if your CMS only lets you inject content there, as long as the JSON-LD itself is valid. AI engines parse both. What matters is that the questions in the schema match the questions actually rendered on the page.
Can a page have more than one H1?
Modern HTML5 technically allows it. SEO practice still says no. Multiple H1s dilute the topic signal and break the heading hierarchy AI engines use to extract structure. One H1 per page, full stop.
How do I check what an AI crawler actually sees on my page?
Curl the URL with an AI-bot user agent (-A "Googlebot", -A "ChatGPT-User") and grep for the elements you expect — title, H1, key paragraphs, schema. If the response is an empty shell, your content is JavaScript-rendered and you need server-side rendering or static generation before any on-page work matters.
How often should I refresh on-page SEO on an existing page?
Cornerstone pages: every 12 to 18 months as a baseline, sooner if the SERP for the target keyword shifts substantially. Refresh means real updates — new data, restructured sections, refreshed code samples — not a bumped date stamp on otherwise identical content. AI engines spot the difference, and citation rates decay fast on stale pages.
Want a second pair of eyes on your on-page SEO?
We walk priority pages through the nine layers above and hand back a prioritised fix list with the actual HTML to ship. Most clients see early movement inside the first month.
Request an audit
