Table of Contents
TL;DR
- On-page SEO is the set of changes you make to the page itself to help Google and the AI chat surfaces understand it. Title, headings, body, links, and images.
- The page title still carries the heaviest single-element weight. Google rewrites about sixty-two percent of meta descriptions when they do not match the page intent.
- The H1 written as a question that mirrors what the customer would type cites at meaningfully higher rates than the H1 written as a topic label.
- The meta description still earns its slot. The four out of ten Google keeps are the four where the description matched the page intent.
- Open Search Console. Find the highest-impression page. Rewrite the H1 to match what the customer would type. Add a forty-to-sixty word answer block right under it. The same change moves both Google rankings and AI chat citations.
Most on-page SEO advice you can read this month was written for the search engine of 2018.
The advice is not always wrong. The page title still matters. The H1 still matters. The internal links and the alt text still matter.
What changed is which surfaces read the page now. In 2018 a page was written for Google to rank. In 2026 the same page is also read by ChatGPT, Perplexity, and the AI Overview that sits above the regular results.
The good news is that the on-page work that wins on chat surfaces is mostly the same on-page work that wins on Google. One rewrite, two engines. The bad news is that the small-business owner who follows 2018 advice ships pages that read clean to a 2018 Google. The same pages read invisible to a 2026 AI chat.
This piece walks through the on-page elements that earn the credit on a small website in 2026. The page title. The H1 as a question. The H2-and-answer-block structure.
The meta description that survives Google’s rewrite filter. The image alt text that does three jobs at once. The internal-link policy. And the single one-hour change worth making first.
What does on-page SEO mean for a small website in 2026?
On-page SEO is the set of changes you make to the page itself. The work helps search engines and the AI chat surfaces understand what the page is about.
Title, headings, body text, links, and images. The list looks the same as it did in 2018. Two changes since then. The chats now read the page too, and the structural shape of the answer matters as much as the keywords inside it.
The mechanical reality of on-page SEO has stayed remarkably stable. Search engines still parse the title, the H1, the H2s, the body text, the links, and the image data. Customers still scan a page in a few seconds and decide whether to read it. The fields and the human reading patterns have not changed.
What changed is which engines read the same fields. ChatGPT, Perplexity, AI Overviews, and Gemini now extract passages from the same body text that Google used to rank. The small business owner who optimises only for Google in 2026 is making the page work harder than it needs to. The same edits that win on Google win on the AI chat surfaces too.
The site article on structuring content for featured snippets in 2026 covers the structural shape of the body in depth. This piece covers the elements above and around the body that work alongside it.
Why does the page title still carry the heaviest single-element weight?
The page title is the line Google shows in its search results.
It is the line most engines hash when they decide what the page is about. About sixty characters before truncation. The customer’s exact wording earns the click. Primary keyword first, benefit second, brand last is the safe template that still works in 2026.
The page title is a single sentence with three jobs. It tells the engine what the page is about. It tells the customer whether to click. It earns or loses the visit before the page even loads.
A title that names the customer’s question wins over a title that names your service. "How do you choose a freelance copywriter?" beats "Copywriting Services" for almost any small site. The first mirrors the search query. The second mirrors the org chart.
Google has been known to rewrite page titles that do not match the page intent. The rewrite happens silently on the search results page. The page in your CMS still shows the title you wrote.
The title shown to searchers is something Google generated. The fix is to write a title Google does not need to rewrite.
Around sixty characters keeps the title from truncating in the search result. Google measures title length in pixels rather than in characters, but sixty is the safe rule of thumb. A title that runs to seventy or eighty characters is a candidate for the rewrite filter.
Why is the H1 better written as a question than as a topic label?
The H1 written as a question cites at higher rates than the H1 written as a topic label.
The question mirrors how the customer would type the query. Pages with a 0.90 or higher cosine match between H1 and search query cite at around forty-one percent versus thirty percent below 0.50. The customer types a question. The H1 should answer the question.
The query-heading match is the single strongest content signal for AI chat citations on the page-level rules so far measured. The Indig analysis of 1.2 million ChatGPT responses (February 2026) showed a clear pattern. Pages whose H1 closely matched the customer’s query cited at meaningfully higher rates. Pages whose H1 named the topic instead cited at lower rates.
The customer does not type the noun-phrase shape of the topic into a search box. Real typing looks more like a how-do-I question. The page that mirrors the customer’s wording is the page the engine surfaces. The page that names the org-chart label sits invisible.
Question-shaped H1s are not a magic format. The point is the match. The H1 takes the shape of the question the customer would actually type.
A what-question H1 for what-questions. A how-question H1 for how-questions. Mirror the query, not the concept.
How do H2s and answer blocks structure a page for both reading and extraction?
Each H2 carries a question the reader actually wonders about.
Right below the H2, an answer block of forty to sixty words gives a clean self-contained answer. A supporting paragraph adds one specific fact or example. The total section sits in the one-hundred-twenty to one-hundred-eighty word band the engines reward for extraction.
The pattern is mechanical. The reader who scrolls scans the H2s first. The H2 that names the reader’s question earns the eye. The answer block right below the H2 gives the answer in one short paragraph the reader can absorb at a glance.
The same shape that pulls a reader’s eye pulls a chat-surface citation. The Indig analysis of 1.2 million AI chat responses found the answer-block pattern in 72.4 percent of cited blog posts. Around forty-four percent of citations come from the first thirty percent of the page.
The supporting paragraph after the answer block is where the depth lives. One specific fact, one small example, one concrete number. The answer block stays clean for the engine to lift. The supporting paragraph rewards the reader who wants more.
Sections shorter than 100 words feel thin and rarely win citations. Sections longer than 250 words tempt the engine to skip them in favour of a tighter source. The middle band is wide enough for most natural section shapes.
What does the meta description do in 2026 if Google rewrites half of them?
The meta description still earns its slot.
Google rewrites around sixty-two percent of meta descriptions when the description does not match the page intent or runs too long. The four out of ten Google keeps are the four where the description matched the page. A definition-plus-benefit-plus-one-number pattern survives the rewrite filter most often.
The 62 percent rewrite rate is the headline figure most small business owners cite as a reason to skip the meta description entirely. The full picture is different. The 38 percent that survive are the meta descriptions that actually match what the page is about.
Writing your own meta description is a bet that yours will be one of the four out of ten Google keeps. The pattern that survives is concrete. One sentence, around 120 to 160 characters. A definition or direct answer at the start.
The trap is letting an AI tool write the description for you. It tends to pad, to hedge, and to drift from the page. That drift is exactly why AI meta descriptions get rewritten the moment the words stop matching the page.
A benefit or specific number in the middle. The primary keyword used once, naturally. No question marks, no clickbait, no urgency.
The honest framing for a small site is that the meta description is hygiene rather than a competitive edge. It does not move rankings. It does move click-through rate when Google keeps it. The five minutes per post still earn their return.
Why do image alt text and filename still matter (and why for a different reason now)?
The original purpose of image alt text was accessibility.
The 2026 reason is broader. Alt text feeds image search, multimodal extraction in the AI chat surfaces, and screen readers, all from the same field. A descriptive sentence that names what the image shows wins all three.
The filename matters at a smaller scale. A meaningful filename adds a small extraction signal at near-zero cost.
The alt text field on every image carries three jobs in 2026. Accessibility for screen readers. Indexing for image search.
Multimodal extraction for the AI chats that read images alongside text. The same descriptive sentence covers all three — a generic placeholder covers none.
A good alt text sentence reads like a caption a sighted reader would also understand. "A small business owner reviews a Search Console performance report on a laptop screen" tells every reader what the image shows. "image1" tells nobody anything.
The filename carries a smaller signal. "google-search-console-performance-report.png" earns a small extraction credit. "IMG_4391.png" earns nothing.
The CMS lets you rename the file once at upload. That is the only moment the work needs to happen.
What internal links does a single on-page section need?
A well-shaped section carries zero links inside the answer block and one link inside the supporting paragraph.
A typical site article of two thousand words ships with two or three internal links across the whole post. The links point to topically-adjacent posts on the same site. The link inside the supporting paragraph sends the curious reader to deeper context without breaking the citation.
The link policy is mechanical and matches what the engines reward. Links inside an answer block reduce citation likelihood. The engine reads the link as a signal that the real answer lives elsewhere. Links inside the supporting paragraphs that follow the answer block do not carry the same penalty.
Two or three internal links across a two-thousand-word post is the practitioner band. Below two the post sits underconnected to the rest of the site. Above five on a small post the page begins to read like a link directory rather than an article with a purpose.
The site article on internal linking for small websites covers the link policy in detail, including the orphan-page check and the anchor-text habit. This section keeps the on-page focus on link placement within the section.
Other questions worth answering
How does schema markup fit a site’s workflow if independent data suggests it barely moves citations?
Schema is hygiene work, not a competitive edge. Vendor studies report 2.7x to 3.2x citation uplift from schema. Independent 2026 analyses found domain authority outweighs schema roughly 3.5 to 1 and detected a slight negative ChatGPT correlation.
Add FAQ schema only where the visible Q and A already lives in the body copy. Skip it when visible content does not match the schema.
What URL slug pattern earns the click on a customer search result?
Three to six lowercase words separated by hyphens, in the customer’s own wording. The shortest noun-phrase that names the query wins the click in a 2026 Google result and a ChatGPT citation footer. Skip stop words. Skip publication years that age the URL inside a year.
How do you stop hedging phrases from creeping into your draft?
Three hedging phrases account for most drift. ‘It appears that.’ ‘Some experts suggest.’ ‘One might argue.’
Print a ban list. Pin it above your screen. Read every sentence. If the verb is conditional, rewrite the verb.
Cited content uses definitive statements 36.2 percent of the time versus 20.2 percent for non-cited content in 2026 ChatGPT data.
When does a heading hierarchy with skipped levels hurt citation rates?
Whenever the customer’s question lives below a skipped level. A heading hierarchy that jumps from H1 straight to H3 reads as broken to an extraction parser. The 2026 ChatGPT parser walks the tree, hits a missing branch, and skips the orphaned content underneath. Perplexity penalizes the orphaned section even when the writing is clean.
How does an AI-drafted batch keep brand voice consistent across many posts?
Two prompts, not one. A system prompt encodes brand voice, banned words, sentence-length caps, and the closing CTA shape. A user prompt supplies the per-post target keyword and the draft body.
Voice drift accumulates batch by batch. A 2026 spot-check on every fifth post catches drift before it ships. Pin a short voice sample to the system prompt to slow drift on ChatGPT or Claude.
Which on-page change should you make first if you have one hour?
Open Google Search Console.
Sort pages by impressions for the last twenty-eight days. Pick the top page. Read its H1. Rewrite the H1 in the customer’s words.
Add a clean forty-to-sixty word answer block right under it. Save and publish. One H1 rewrite with an answer block under it typically moves more than three new posts.
The high-impression page is the page the engines have already decided is worth showing. The work is to convert the impressions into clicks and citations. The H1 rewrite tells the engine the page now answers the customer’s question in the customer’s words.
The answer block under the rewritten H1 gives the engine something clean to lift when it cites the page. Forty to sixty words. No hedging.
No links inside the block. The answer the customer would walk away with after reading just the H1 and the next paragraph.
The next month’s impressions and citations tell you whether the rewrite landed. The data takes about two weeks to settle and another two weeks to read. One H1 rewrite per month is enough to compound.
The rewrite goes on the page earning the most impressions. After a year the site earns its place on both Google and the AI chat surfaces.
If you want a second pair of eyes on which H1 to rewrite first, you can contact me here. Send the URL of the page and the top three queries it ranks for from Search Console. I will name the H1 I would rewrite and the answer-block draft I would write under it. No pitch.