Run any site through an audit tool and you get several hundred issues in three colours. Most of them do not matter. A handful matter enormously, and they are usually decided by how the site was built rather than by anything you can fix afterwards with a plugin.
This is the short list, in the order we would fix it.
1. Your content must be in the HTML
This is the one that quietly ruins otherwise good sites. If your page arrives as an empty shell and fills itself in with JavaScript afterwards, you are relying on every system that reads your site to run that JavaScript.
Google generally will, eventually, with a delay. Plenty of other things will not: Bing is less reliable at it, social preview crawlers do not, and the AI assistants that increasingly sit between a customer and a recommendation mostly fetch the raw HTML and read what is there.
There is a simple test. View source on your most important page and search for a sentence from the middle of the content. If it is not there, that page is invisible to a growing share of the systems that decide whether anyone finds you.
The fix is server rendering or static generation. It is a build-time architectural decision, which is why it belongs first on this list and why retrofitting it is expensive.
2. One page per thing you want to be found for
A single-page site with everything on it cannot rank for more than roughly one topic, because you have one title, one description, and one URL to work with. A page can only be the best answer to one question.
If you offer four services, that is four pages. If you have six projects worth talking about, that is six pages. This is not a trick, it is arithmetic: search results are per URL, so a URL is the unit you have to think in.
The caveat is that each page must be genuinely worth reading. Ten pages of real substance beat a hundred thin variations, and the thin variations can actively harm you.
3. Titles and descriptions written per page
The title tag is still the single highest-leverage line of text on any page. Write it for a human scanning results, put the useful word near the front, and keep it under about sixty characters so it does not get truncated.
Meta descriptions do not affect ranking directly and heavily affect whether anyone clicks, which affects everything downstream. Write them as a promise about what the page contains, not as a slogan.
The common failure is templating both across the whole site, so forty pages share one title with the site name appended. Every page needs its own.
4. Speed, specifically the three that are measured
Core Web Vitals are a genuine ranking input and a much larger conversion input. Three numbers matter: how long until the main content appears, how quickly the page responds to a tap, and whether things move around while it loads.
For most small sites the wins are unglamorous and identical every time: serve images in a modern format at the size they are actually displayed, give images explicit dimensions so nothing jumps, load fonts without a flash of invisible text, and delete the JavaScript that is not doing anything. Third-party scripts are usually the single largest cost, and each one should have to justify itself.
Test on a mid-range phone on a normal connection, not on your laptop on office wifi. The gap between those two is where most of the disappointment lives.
5. Structured data, on the things that have a type
Structured data tells machines what a page is rather than making them infer it. It has become more valuable as AI assistants have started answering questions directly, because an explicit machine-readable statement of your organisation, services, and FAQs is easier to trust and reuse than a paragraph.
For most sites the useful set is small: Organization once, Service on service pages, FAQPage where you have real questions and answers, Article on posts, and BreadcrumbList sitewide. Mark up what is genuinely on the page. Marking up things that are not there is the fastest way to lose the benefit entirely.
6. Internal links that mean something
Internal linking is the most underrated item on this list. It tells search engines which of your pages matter and what they are about, and it keeps people on your site.
The practical rule: every important page should be reachable within two clicks of the homepage, related pages should link to each other in both directions, and the link text should describe the destination. "Read more" tells a machine nothing. "Mobile app development" tells it exactly what it needs.
7. Canonical URLs and one version of the site
Pick one hostname, with or without www, redirect the other, force HTTPS, and be consistent about trailing slashes. Set a canonical URL on every page. This is boring and it prevents your own pages competing with each other, which is a self-inflicted problem and surprisingly common.
8. A sitemap generated from your content
A hand-maintained sitemap is a sitemap that is wrong within a month. Generate it from the same source as your pages so a new case study or post appears in it automatically, submit it once in Search Console, and forget about it.
What changed now that AI assistants read your site
A growing share of buying research now happens inside an assistant rather than a results page. Someone asks for a recommendation and gets three names and a summary, without ever seeing ten blue links.
That shifts the emphasis rather than replacing the fundamentals. Content has to be in the HTML, because most assistants fetch rather than render. It has to state facts plainly, because a page that says what it does, where it operates, and what it costs can be quoted, while a page of atmospheric copy cannot. And clear structure with real headings and question-and-answer sections is easier to extract from than a wall of prose.
The useful mental model: write the page so that a stranger who reads only it could accurately describe your business to someone else. That is now a literal requirement rather than a metaphor.
What to ignore
- Keyword density. It has not been a real factor for well over a decade.
- Meta keywords. Ignored by every major search engine since the 2000s.
- Your overall score in an audit tool. It is a marketing number for the tool, not a measure of anything Google uses.
- Chasing a perfect Lighthouse score. Going from 70 to 90 matters. Going from 95 to 100 is usually a waste of a day.
- Buying links. It has a downside and no reliable upside.
Fix the eight above, publish things worth reading, and let the audit tool complain about the other three hundred.