This website uses cookies

Read our Privacy policy and Terms of use for more information.

🔎 Focus: Tech SEO
🔴 Impact: High
🟢 Difficulty: Really easy man

Sponsored by Ahrefs

Does AI Recommend Your Brand?

Track and grow your brand’s visibility across AI answers, YouTube, and Reddit.
Turn SEO into AEO and reach new audiences.

Most of my week looks like a Jira ticket.

A developer on the other side of a Jira board, reading three paragraphs I wrote late at night, deciding whether it's worth a slot in their sprint. That's where technical SEO actually happens. Not in the 200-page audit or deck. In the single ticket.

I've been thinking about this a lot lately because I'm putting together a talk on an idea I call Quality Density and Quality Cost. The short version goes like this. Every website has a price - The price a search engine or an AI system pays to crawl it, render it, and pull out what it actually says.

Lower that price and, over time, you get rewarded. Raise it, and you get quietly ignored. Nobody sends you an email about it.

So instead of theory, this issue is four things I actually worked on over the last few months. Clients anonymized, details kept. Enjoy

I. The typo that deleted every review

The first one is a watersports retailer. One product page, an inflatable SUP board, the kind of page that should be doing heavy lifting.

The page had customer reviews. Real ones. They rendered fine for humans, stars and all. If you looked at it in a browser, you'd think everything was working.

Then I opened the JSON-LD.

The review property was misnamed. One word, review versus reviews. That's it. And because the structured data didn't match what Google expects, every single customer review on their product pages was being discarded. The shop had done the hard part, actually getting customers to leave reviews, and a property name was throwing all of it in the bin.

A simple typo broke structured data

We also found an issue in the data.

While I was in there, I found a second problem. The product description said one thing about how the board was constructed. The spec table further down the same page said something else.

I think people underestimate this one. A machine trying to extract facts from the page can't. It has to pick one claim, or trust neither. Either way, the page just got more expensive to understand.

The fix was two parts. Correct the structured data so the reviews are actually read. Then rewrite the content using an entity-first approach, so the copy and the spec table say the same thing about the same product.

II. Twenty-three ways to spell a URL

The second one is a Polish electrical e-commerce site I've worked with for years. Huge catalog. The kind of site where small problems don't stay small.

I ran an analysis of the URL slugs the CMS generates from product names. It turned up 23 distinct special character encoding anomalies.

I want to be careful about what that number means, because it's easy to misread. It's not 23 broken URLs. It's 23 different patterns of things going wrong. Each one gets multiplied across every product whose name happens to trigger it.

Polish doesn't make this easy. Diacritics, plus product names full of slashes, symbols, and model numbers. Every one of those characters is a decision the CMS has to make when it builds a URL.

Make that decision inconsistently, and you get URLs that look different, behave differently, and sometimes point at the same thing.

Here's what I'd argue is the real takeaway. You don't fix this URL by URL. If you do, you'll be doing it forever, because the CMS will keep generating new ones while you work.

You fix the rule. Audit how slugs are generated, write down what should happen to each character, and get that into the generator. Then deal separately with what happens to the URLs that already exist.

The URLs were also created differently for internal linking and differently for sitemaps, creating tens of thousands of redirects, which was insane.

One ticket on the rule saves you hundreds of tickets on the symptoms.

III. Shopify makes crawl waste for you

The third one is a multilingual Shopify store.

Shopify is a good platform. I'd also say it generates crawl waste by default, and most store owners have no idea it's happening.

On this store, crawl was being spent on three things that didn't need it.

The first was pr_prod_strat parameters. These get appended by product recommendation blocks, so every "you might also like" link carries a parameter, and every parameter is a new URL to crawl.

The second was _pos parameters, another set of variations hanging off product links.

The third was the legacy /collections/*/products/* URLs. Shopify lets a product live at its collection path as well as its clean /products/ path. Same product, more than one address.

The usual response is "but the canonical tag handles it." And sure, the canonical tells Google which version you prefer. But a canonical is a hint about what to index. It doesn't stop Googlebot from following every link you hand it. If your internal links point at the parameterized or collection-path version, that's what gets crawled.

The number of pages in Google's database is steadily dropping

So the fix wasn't more tags. It was pointing internal links at the canonical product URLs in the first place and blocking the unwanted parameters. We cleared the database by hundreds of thousands of pages, and now Google has a denser, higher-quality website to analyse.

IV. Move into the house that already has the furniture

The fourth one is a Dutch car leasing site, and this one is a plan rather than a finished result, so I'll be honest about that.

The site had vehicle pages split by leasing type. The same car, effectively, spread across separate pages depending on how you wanted to lease it. That means the authority for one car is divided, and a search engine has to work out which version matters.

The plan is to move vehicle pages into one folder, with the leasing type options living on each vehicle page instead of across separate pages. One car, one page, every option in one place.

The collection pages are where it got interesting. The obvious move in a restructure is to build a clean new folder and move everything into it. Fresh start.

I pushed against that. There was already a folder with collection pages that ranked well. So we keep those exactly where they are, and move the other collection pages into that folder, rather than uprooting the ones that are already working.

I think of it like moving house. You don't move out of the place that already has the furniture just because the new place has nicer walls. You move everything else in there.

The principle: when you consolidate, consolidate into the location that already holds equity. Minimize the number of things that have to be rediscovered. Every URL that moves is a URL a search engine has to learn again.

What these four have in common

If you squint, these look like four unrelated jobs. A schema bug, a slug problem, a Shopify quirk, and a site restructure.

I don't think they are. Every one of them is a Quality Cost problem.

Reviews that exist but can't be read. Product facts that contradict each other. URLs that multiply from one rule nobody looked at. Crawl spent on parameters instead of products. One car's authority split across several pages. In each case, the site was making itself more expensive for a machine to understand than it needed to be.

And this matters more now than it did a few years ago. It's not only Googlebot paying that cost anymore. Every AI system trying to extract a clean answer from your site pays it too. The cheaper you are to read, the more likely you are to be the one that gets read.

The other thing they have in common is how they were delivered. None of them went into a PDF.

Each one became a dev ticket. A specific problem, the specific fix, what done looks like, sitting on the same board as everything else the developers are working on. I've written plenty of audit documents in my life. I've also watched a lot of them sit in a shared drive, untouched.

A PDF tells you what's wrong. A ticket gets it fixed. That's most of the difference.

The review bug took one line to explain and one line to fix. It had probably been there for a long time. Not because nobody was capable of fixing it. Because nobody had turned it into something a developer could pick up on a Tuesday.

That's the job, I think. Not finding more problems. Making the right ones cheap enough to fix.

Are you struggling with growing your site?

If you'd like this kind of work done inside your own dev workflow, embedded with your team and shipped as tickets rather than reports, just book a call below

Seeeeee you soon 👋

oh that’s a human

—