Want to automate the grunt work?

Meet Agent A, the AI teammate with full access to your Ahrefs data. From fixing keyword cannibalization to shipping technical reports straight into Notion or Google Docs, it does the work you’d rather not.

Google has moved from evaluating what a page says to evaluating how it's put together. Layout, components, structure, function. Those are called visual semantics, and they are positioned as the missing piece of topical authority.

In this newsletter, I present you with a shortened version of the research done by Koray Tugberk Gubur from his latest Search Engine Land article.

I think the way our industry will receive this is wrong, and predictably so. It'll get filed under design. It'll end up in a deck for the brand team. Someone will make a carousel about "designing for AI search."

That's not what this is. Almost every claim in that article resolves into something a developer has to change in a template and it impacts organic performance. Which makes it a Tech SEO thing.

Martin Splitt has talked publicly about it:

Documents from the antitrust case showed the same concept being used to classify and rank news, capped at roughly 400 characters.

Here's the detail that stopped me. Those same documents show the extraction being interrupted by share buttons. Facebook, email, Twitter, Google+, sitting in the middle of the DOM, cutting a sentence in half. Google's own example of the thing failing is a boilerplate widget in the wrong place.

Read that again, because it's not a design problem.

Nobody's typography was bad.
Nobody chose the wrong shade of blue.
A template put junk between the words, and the machine picked up the junk.

If the first substantive thing in your DOM is a nav, a cookie banner, a share widget, and a breadcrumb, you are handing Google a 400-character summary of your nothing.

The industry has been busy talking about chunking as if it were purely a linguistic operation. Split the text, embed the chunks, done. Koray's point, and I'd push it harder than he does, is that chunking is structural before it's linguistic.

Something has to decide where the boundaries are. That something reads your HTML.

So you can have perfect entities, clean triples, a beautiful semantic network, and none of it lands if the machine can't tell where one idea stops and the next begins.

The part of the article I keep coming back to is retrieval cost, because it reframes everything else.

Koray's version of the line is that ranking a document can't cost more than ignoring it. Pandu Nayak said something adjacent under oath during the antitrust trial:

Google doesn't run its expensive systems on every page. It checks core topicality first and decides whether you've earned the compute. RankBrain-class algorithms get reserved for results with at least one click and strong signals already in place.

The Content Warehouse leak points the same way. Documents get truncated. Quality gets predicted from early signals. Fail the first check, and you never become a candidate for the second.

Then, in December's core update, the HTML size limit dropped to 2MB and a lot of pages quietly stopped being indexed.

Put those together, and you get an uncomfortable picture. Google's ranking stack behaves like a decision tree. Your best content lives at layer four. If layer one can't work out what your page is, layers two through four never run. Your brilliance is never evaluated. It's not rejected, it's skipped.

Which brings me to the distinction in that article I'd put on a wall.

Low-quality publishers scale text. High-quality publishers scale systems.

Everybody now has a machine that writes four hundred blog posts a month. Almost nobody is investing in front-end work that makes a document cheaper for Google to understand. The gap between those two groups is about to get very wide.

Here's the part e-commerce people need to sit with.

Google appears to classify a website by type before it judges the content on it. Affiliate, aggregator, retailer, SaaS. Same words, different site type, different outcome. Koray describes moving identical content from an affiliate site onto an e-commerce site with an integrated topical map and watching rankings improve almost immediately. Nothing about the text changed. The context around the text changed.

So how does Google know what type you are? It can't ask. It reads your layout. What a page can do is written into its components.

That's where relevance and responsiveness split apart, and it's a distinction worth stealing. A page can be relevant to a query and still not be responsive to the job behind it.

If the task is compare, filter, buy, book or review, and your page only talks about those things, you're relevant and useless.

Google added misleading functionality to its spam policies after the helpful content updates. A page that looks like it compares but doesn't. A page that promises a calculator that isn't there.

I'd argue most e-commerce sites are closer to that line than they think, not through deceit but through neglect.

Filters that only exist in JavaScript.
A comparison module hidden in a tab that never renders.
Reviews in a third-party iframe.

Every one of those is a page claiming a function it can't demonstrate to a machine on a budget.

Your collection page is not an article with products underneath it. If that's what it looks like structurally, that's what it'll be classified as, and it'll be competing in the wrong lane.

Because there are lanes. The antitrust documents show Twiddlers with functions like max_total and a BlogCategorizer, capping how many results of the same type appear together. The Content Warehouse leak has a module assigning categorical weight. Google isn't just ranking documents - it's composing a results page.

So if a query wants two listicles, three product grids, and a video, being the third-best listicle is a losing position no matter how good you are. Page type isn't a default you inherit from your CMS. It's a strategic choice, and most people never make it consciously.

The practical version of all this is small and slightly boring, which is how you know it's real.

Koray's air conditioner example is the clearest thing in the article, so I'll use it as he laid it out.

"How do I repair my AC" is an experience query, and it wants a forum shape.

"AC installation in Rotterdam" is local and service-led, and it wants a directory with providers, ratings and contact routes.

"AC installation prices" is a hybrid, and it wants an answer at the top with providers underneath.

"How to install an AC" is instructional, and every local element you bolt on makes it worse.

Four queries about one object. Four different page types.

Most topical maps I see, including a few of ours from a couple of years back, have a topic column, a query column, an intent label and a priority. None of them have a column that says: this page needs a filterable list, a price table and a quote form, and if we can't build those, we shouldn't publish it.

That's the change. Not more topics. Fewer pages, each one shaped like the job it needs to do.

We need to track the evolution of the topical authority formula across the site,

It started as historical data times topical coverage. Then it became that, divided by cost of retrieval. Now we add a multiplier for having the right visual annotations.

A multiplier is the right operator, I think. If the annotation is wrong, everything upstream gets multiplied by something close to zero. Perfect coverage, perfect accuracy, years of click history, and layer one still doesn't know what it's looking at - is the page an article or a collection page?

I want to be honest about the evidence, because this is an area where confidence outruns proof.

Patents are not shipped systems. Google files things it never builds. The case studies in that article are self-reported, with real names attached, which I respect, but they're still one practitioner's account.

What makes me pay attention isn't any single source. It's that four independent ones point the same direction.

When those agree, the direction of travel is probably real even if the mechanics aren't exactly as described.

So, three things I'd do this week, none of which require believing every word from those sources.

The thing I find quietly funny about all of this is that it puts technical SEO back in the room at the exact moment everyone decided content was the whole game.

For the last two years, the industry's answer to every question has been more words, faster. And the response from search, as far as I can read it, is to get better at ignoring words it can't cheaply place.

A page is letters, pixels, and bytes. We got very good at the letters. We outsourced the pixels to whoever owned the design system. Nobody was watching the bytes.

I don't think layout is the new ranking factor. I think it's the old one that never got a name, and now it has one.

If you want to talk about what that means for your website, we do this all day at organichackers.com.

Reply to this email with “Layout” and your domain.

I’ll take a look at your site and check if your layout is quietly killing your rankings every single day and show you exactly how to fix it.