SEO tips ranked by how often the thing is actually broken

Almost every list of SEO tips is somebody’s opinion, presented flat, with no indication of which items are commonly broken and which are already fine on your site. That ordering is the only part that saves you time.

So this list is sorted differently. We ran the same audit over 88 professional marketing sites that returned HTML to our fetcher, and separately over a much larger sample of 1,048 sites for crawler access. The tips below are ordered by how often the thing was actually broken, which is a fact rather than a preference.

1. Check that your canonical resolves, not redirects

The most broken single thing we measure, on roughly a third of professional sites. It is almost never deliberate. The canonical points at the apex while the site serves www, or at http while everything redirects to https, or at a URL without a trailing slash on a platform that adds one.

The effect is that you have handed a consolidation instruction that costs a redirect hop to follow, on every page, forever. Google’s own guidance on consolidating duplicate URLs is explicit that the link element is a hint rather than a directive, which is the part that makes a wrong one so quiet: nothing breaks, you are simply ignored.

How to check it: request the canonical URL itself and confirm it returns 200 directly. If there is a hop, your canonical is wrong, not your redirect. The canonical checker does the request for you, and the longer version of the rules is in what a canonical tag is and when it gets ignored.

2. Put a date in the markup

Nearly a third of the sites we audited carry no date a machine can find. A date rendered by JavaScript is not a date. “Updated recently” is not a date. A date in an image is not a date.

This is one of the cheapest fixes on the list and one of the most consequential, because freshness is one of the few signals anything assessing your page can evaluate without understanding your industry.

How to check it: view source, search for the year. If it is not there, you do not have one. The content freshness checker does it across the page and tells you which of the several places a date can live you are actually using.

3. Say who wrote it, in the markup

A quarter of the sample has no author in the HTML and no Person or Organization schema. A byline styled onto the page in a div is invisible to everything that is not laying out the page visually.

Do both: a visible byline in the markup and an author field in your structured data. They are read by different things and neither substitutes for the other. The Article block is four lines and it is the half most often skipped.

One qualifier, because it is the difference between doing this and gaming it. The author has to be a real person with a profile that resolves. A Person node naming somebody who does not exist is a fabricated entity claim, and it is one of the things our own audit flags on other people’s sites.

4. Stop shipping more bytes than your content justifies

A quarter of the sample fails our page weight check. This is a crawl budget problem as much as a speed problem: a crawler working to a budget fetches fewer of your pages when each one is expensive, and Google’s own guidance for large sites says making pages cheaper to serve is one of the few levers you actually control.

The usual cause is not images. It is markup: enormous inline SVG, duplicated component chrome, and framework output that ships the same thing several times. We have fought this on our own site and the honest finding was that deduplicating markup recovered far less than expected, and the only real lever was having enough actual content to justify the wrapper. The page speed checker reports document weight separately from render time, which is the split that tells you which of the two you have.

5. Check what your robots.txt does to AI crawlers specifically

Across 1,048 sites, 28.3% block at least one major AI crawler while only 1.5% block Googlebot. Nearly 27% do the first while permitting the second, and in most cases nobody chose that: it arrived with a security product, a platform default, or a copy and pasted block from a template.

If you have decided to block them, fine, that is a legitimate position. The problem is the sites that have not decided anything and are blocked anyway.

How to check it: name the agents explicitly. A blanket Allow is not the same as naming GPTBot, ClaudeBot, PerplexityBot and Google-Extended, because group selection and longest match mean a bot with a group of its own is judged only on that group. The AI crawler checker evaluates your live file once per agent, and which crawler belongs to which assistant covers the ones worth naming.

6. Make sure your structured data parses

Only about a third of the larger sample has JSON-LD present and parsing at all. Among those that have it, the quiet failure is that a malformed block is discarded entirely and silently: one trailing comma or one smart quote pasted from a document, and the whole block is ignored with no warning in any interface.

How to check it: extract every application/ld+json block and parse it. Then read it back against what the page actually displays, because markup describing things the page does not show is a policy problem rather than a technical one. The structured data checker does the mechanical half, and working examples of the six types most sites need covers the rest, including the FAQ rich result Google withdrew in May 2026 and what that changes about why you would still ship the markup.

7. Serve your content without JavaScript

In our agency sample this failed on nobody, which is the finding rather than an absence of one: professional marketing sites are mostly static. The failure concentrates hard elsewhere, in app shells and single page applications built without server rendering.

If that is you, the consequence is not subtle. Anything reading your page without executing scripts receives an empty shell, and a large share of the retrieval ecosystem does exactly that.

How to check it: disable JavaScript and reload. What you see is what a lot of machines see. The render gap checker does the same comparison without you having to change a browser setting, and why AI crawlers see a blank page on your React site has each major crawler’s documented position on whether it runs scripts at all.

8. Write sections that can be lifted out on their own

The newest item on the list and the one with no legacy equivalent. Assistants cite passages, not pages. A paragraph that only makes sense after the three above it cannot be quoted, no matter how good it is.

The test takes ten seconds: take any single paragraph out of context and read it. If it still answers something on its own, it is liftable. The AI content readiness checker scores this across a whole page.

Be careful about how far you take it, because there is a version of this advice that is wrong. Google’s guide to optimizing for generative AI search explicitly lists “chunking” content into tiny pieces as something you can ignore, and says there is no ideal page length. The useful reading is not cut everything short. It is do not write a paragraph whose meaning depends on the three above it, which is an editing standard rather than a formatting rule.

This is also the one piece of advice here that improves the page for humans at the same time, which is usually a sign that it is real.

9. Confirm your server does not refuse crawlers

18% of sites in the larger sample answered 401, 403 or 429 to a declared crawler, and a share of those served a perfectly normal page to a Chrome user agent from the same address a moment later.

This is the most invisible failure in SEO. Your browser works. Your analytics look fine. And a meaningful set of machines simply cannot fetch your site.

How to check it: fetch your own homepage with curl and no user agent, and compare it with what a browser gets. The indexability checker reports the status a declared crawler receives, and the study itself has the full breakdown and the CSV.

10. Then, and only then, worry about the rest

Everything else that fills these lists, keyword placement, internal link counts, heading phrasing, image formats, is real and none of it is the constraint on a site that is failing anything above.

The reason to order tips by failure rate rather than by importance is that importance is an opinion and a rate is not. A check can be enormously important and almost never broken, which makes it a poor use of an afternoon. The canonical target check is the opposite: modest weight, and the most broken thing we measure anywhere.

If you want the whole list run against your own site rather than read as advice, that is what our free audit does, and the technical SEO checklist is the same material as a list you can tick off by hand. The free tools each run one dimension of it if you would rather start with the single thing you suspect.

Sources

Read on 15 August 2026. The failure rates are ours, measured on the dates given in each study; everything else is somebody’s published documentation and you should check it rather than us.

• Visibility100x, the GEO agency audit, for the 88 site sample and every failure rate quoted in items 1, 2, 3, 4 and 6.
• Visibility100x, who blocks the AI crawlers, for the 1,048 site sample, the 28.3% and 1.5% figures, and the 18% refusing a declared crawler.
• Google, Search Essentials, for the technical requirements and spam policies underneath most of this.
• Google, consolidate duplicate URLs, for canonical being a hint rather than a directive.
• Google, managing crawl budget for large sites, for what page weight costs you in crawling.
• Google, creating helpful, reliable, people-first content, for the self assessment behind item 10.
• Google, optimizing for generative AI search, for the position on chunking, page length and AI text files.
• IETF, RFC 9309: Robots Exclusion Protocol, for the group selection rules in item 5.