On-page SEO: what matters and what does not

On-page SEO checklists have a structural flaw: they present forty items as though each mattered equally. Keyword in the title, keyword in the H1, keyword in the first hundred words, keyword density, image alt text, meta keywords tag, LSI terms, word count.

They do not matter equally. A small number of things determine whether a page can rank at all, a few more help at the margin, and a long tail exists mostly because it is easy to put on a checklist and tick off.

This is the same material sorted by consequence. It is a sub-pillar of SEO basics, which covers the wider picture.

The three that actually decide it

1. Whether the page matches the search intent

Not a keyword question. A "what does the searcher want" question.

A page can have perfect titles, headings and structure and rank nowhere because it answers a different question from the one being asked. If the results for your target query are all comparison tables and you wrote a how-to guide, no amount of on-page tidying fixes that.

This is first because it is the only item that can invalidate everything else. The method for checking it is in search intent.

2. Whether the page actually answers the question, completely

The unglamorous one that decides most outcomes.

A page that answers the question better than the pages currently ranking has a reason to displace them. A page that says the same things less specifically does not, however well optimised. This is what "quality" means in practice — not writing flair, but whether a reader leaves with what they came for.

Concretely, better usually means: more specific, more current, includes the thing everyone else omits, and shows evidence of first-hand knowledge rather than assembled summary.

"Completeness" sounds unmeasurable, so here is a method that takes twenty minutes. Open the top five results. Write down every distinct question each of them answers — not headings, questions. You will end with a list of perhaps fifteen, of which the top pages each cover eight or ten. Now mark the ones your page answers.

Two things fall out. The questions nobody in the top five answers, which you almost certainly know the answer to from doing the work, are your differentiation. The questions everyone answers and you do not are your gaps. A page that closes its gaps and adds two things nobody else says has a genuine claim to be the better answer; a page that covers the same eight questions less specifically does not, however it is formatted.

3. The title tag

The one traditional on-page element that genuinely earns its reputation, for two reasons: it is a strong statement of what the page is about, and it is what people decide to click.

That second job is often worth more than the first. A page at position four with a title people choose outperforms a page at position three with a title they skip. There is a full treatment in how to write a title tag that earns the click.

The ones that help, in proportion

Headings. Not because a keyword in an H2 carries weight, but because headings structure an argument, and a well-structured page is easier to read and easier to extract answers from. Use them to reflect the actual shape of the content. One H1, which on most systems is the title, then H2s for sections.

URL. Readable and stable matters more than keyword-stuffed. /blog/local-seo is better than /blog/post-4471, and both are better than /blog/best-local-seo-services-local-seo-tips-2026. Changing URLs on existing pages carries risk and rarely pays.

Internal links. Genuinely important, and usually filed under on-page. Links from other pages on your site help a page be found, and the anchor text describes what it is about. Treated properly in internal linking: a system, not a habit.

Meta description. Does not affect ranking. Does affect whether your result is clicked, which affects everything downstream. Worth writing, not worth agonising over — Google frequently rewrites it anyway.

Images. Compressed so the page is not slow; alt text that genuinely describes the image for people who cannot see it. Both are worth doing and neither is a ranking lever.

The ones that are essentially decoration

Said plainly, because time spent here is time not spent on the first three.

Keyword density. There is no target percentage. Writing to hit one produces text that reads as though it was written to hit one.

Meta keywords tag. Unused by Google for many years.

Exact-match keyword in the first hundred words. If it belongs there, it will be there naturally. Forcing it is visible.

Word count as a target. Google's own position is that there is no preferred length. Length is an output of covering a subject properly, not an input. Whether more articles or longer ones help is covered in how many articles before SEO starts working.

LSI keywords. Not a thing Google has ever described. Writing naturally about a subject produces related vocabulary without a list.

Sprinkling synonyms to "signal relevance". Reads exactly as it sounds.

Where the advice changes: commercial pages

Almost every on-page guide is written about articles, and then people apply it to their service and product pages, where several of the rules invert.

Length stops helping earlier. Someone on a "plumber in Leicester" page wants to know that you cover their area, what you charge roughly, and how to reach you. Two thousand words about the history of plumbing pushes all three below the fold.

The title's job shifts from interest to reassurance. On an article, the title competes for a click among ten similar promises. On a commercial page, it confirms the reader has arrived in the right place — the service, the place, and nothing clever.

Proof outranks prose. Real photographs of real work, actual prices or a price range, named service areas, genuine reviews. These do more for a commercial page than any amount of on-page tuning, because the visitor's question is "can I trust this" rather than "is this thorough".

One page per thing you sell, per place you sell it — where the pages are genuinely different. Where they are not, you are producing near-duplicates, which is the failure examined in local landing pages that are not thin doorway pages.

What about E-E-A-T?

It comes up in every on-page discussion, so worth being plain about.

It is not a score, not a tag, and not something you can add to a page. It describes the kinds of signals Google's guidance says quality raters look for — evidence of experience, expertise, authoritativeness and trust — and there is no field to fill in.

What it means in practice is unglamorous and mostly not on-page at all: say who wrote the page and why they would know; show the work rather than describing it; cite what you claim; make it possible to find out who you are and contact you. On sites offering medical, legal or financial advice this matters more, because the standard applied is higher.

You cannot optimise for it. You can be a site that visibly has it.

A ten-minute pass that catches most of it

For an existing page, in order:

1. Search the target query. Do the top five look like your page? If not, stop — the problem is intent, not on-page.
2. Read your page against the weakest of those five. Is yours clearly better? If not, that is the work.
3. Rewrite the title so someone scanning ten results would choose it.
4. Rewrite the description to match what the page delivers.
5. Check the headings tell the story without the body text.
6. Add two or three internal links from newer pages to this one.
7. Check it is readable on a phone.

Items three to seven take twenty minutes. Item two takes an afternoon and is where the result comes from, which is why it is skipped so often.

What on-page cannot do

Worth stating because on-page work is where people go when a page is not ranking, and it is frequently not the cause.

On-page optimisation cannot make an unwinnable query winnable. If the pages above you are strong, specific and well-established, a better title will not displace them. The check for that is in judging whether you can realistically win a keyword, and the broader diagnosis in why your website is not ranking.

It also cannot compensate for a site nobody has heard of. On-page work makes a page the best answer it can be; it does not create the standing that lets a site compete for contested terms. Where that standing is missing and the timeline matters, the honest alternatives are narrower queries, or paying for placement outright — the prices for which are on the rankings board.

Common questions

Should the keyword be in the first sentence? Only if the sentence is better for it.

How many H2s should a page have? As many as it has sections.

Does bolding keywords help? Bold what a reader should notice. That is the whole rule.

Should I optimise old pages or write new ones? If the old ones are close and stuck, optimise. If the subject is not covered, write. Content refreshes covers the triage.