The web isn’t URL-shaped anymore
30th July, 2025
Half of your website’s audience isn’t human.
Search engines, crawlers, bots and agents likely already consume more of your site than people do. And they don’t experience it the way humans do. Humans load a page at a URL and absorb the whole thing: design, content, messaging, intent. Machines treat that URL as an envelope – something to tear open, strip apart, and mine for meaning.
For decades, SEO has been built around the idea that the URL is the thing. We’ve optimised containers ‑because Google indexed and ranked at the URL level, and because humans needed that structure to navigate. But the systems that now mediate most discovery have already moved past this model. They don’t care about the envelope; they care about the assertions inside it and how those assertions connect to everything else they know.
That shift changes everything about how optimisation works.
To understand what’s at stake, we first need to look at the mental model we’ve been stuck in – and why it no longer fits the web we’re actually dealing with.
The web we optimised for vs. the web we have now
For as long as we’ve been doing SEO, we’ve been operating inside a URL-shaped worldview.
Google’s early systems indexed and ranked URLs. PageRank flowed along links between them. Analytics tools counted visits per URL. Keyword research mapped to URLs. Content strategies were literally drawn as sitemaps. The URL wasn’t just an address – it was the atomic unit of the web.
That model shaped everything:
- We thought in terms of “optimising this page” to rank for that keyword.
- We judged success by whether a URL appeared in search results.
- Even our language – this page ranks, that page needs links – reinforced the assumption that the URL was the thing that mattered.
But the web machines see today isn’t URL-shaped anymore.
Google has been breaking the page into pieces for years. Passage ranking lets it surface a paragraph buried deep in a document. Featured snippets lift individual claims straight out of context. The Knowledge Graph doesn’t “rank pages” at all – it extracts facts and relationships from across the web.
Even LLM-powered agents, which still request pages at URLs, don’t consume them like humans do. They strip out scripts, styling, and noise, and break the content into chunks to extract meaning and assertions. To these systems, the page is just a source file – the raw material they transform into something they can reason over.
The systems that matter now don’t treat a URL as an indivisible object.
To them, a page at a URL is just a container – a source of assertions to extract, evaluate, and connect. And that means our page-first, URL-first mental model no longer matches the environment we’re optimising for.
Machines care about assertions, not pages
Modern systems no longer evaluate your site as a set of documents. They break it down into assertions – discrete statements about the world that they can extract, interpret, and integrate into their models.
These assertions are often represented as triples:
- subject → predicate → object
- Product X → has price → $99
- Author Y → wrote → This Article
Search engines and knowledge graphs process these triples as symbolic facts. They’re stored explicitly, connected to other nodes, and used to answer queries or populate search features.
LLMs and agents do something different.
They don’t store assertions as facts – they encode them into high-dimensional vector spaces. Meaning is compressed, patterns are generalised, and retrieval is probabilistic. The assertion doesn’t live as a record – it lives as part of the model’s learned representation.
In both cases, the unit that matters isn’t the page – it’s the assertions inside it.
What counts is whether those statements are clear, extractable, and connectable. This is the layer machines care about, and it’s where SEO needs to shift its focus: away from optimising documents, toward optimising the clarity and connectivity of the claims your brand makes.
Trust is graph-shaped
Assertions don’t stand alone. A single claim, no matter how well-structured, doesn’t automatically earn trust. Machines evaluate it in the context of everything else they know. They look at how that claim connects: where it came from, what supports it, and how it relates to other signals in their models.
This is where the graph matters.
Search engines like Google use explicit relationships: links between URLs, corroborating schema, and references from trusted nodes. The Knowledge Graph literally encodes trust through connectivity – facts supported by multiple reliable sources gain weight, while contradictory or isolated claims are discounted.
LLMs and agents do something similar in a more implicit way.
They rely on the statistical density of patterns across their training data. Assertions that appear consistently in high-quality contexts become stronger in the model’s representation. Outliers – claims that don’t fit or aren’t well-supported – carry less influence or risk being ignored.
This graph-shaped evaluation is what should prevent “assertion spam” from working.
But right now, many machine learning systems are still naïve. LLMs and emerging agents lack the decades of spam-fighting infrastructure that Google has built – they don’t yet have the equivalent of a “link graph” to moderate quality signals.
To compensate, many of these models lean heavily on search engines’ own results as a proxy for trust. When ChatGPT, Perplexity, or others surface answers, they’re often drawing from Bing or Google SERPs, inheriting their quality filters (and their biases). That means the search engines’ defences still shape how LLMs perceive the web – but it also means that if those results are polluted, the pollution propagates.
Over time, as LLMs develop their own ways of weighting relationships and filtering noise, coherence and connectivity, not volume, will decide what’s trusted. The challenge for brands is to play the long game: build assertions that can stand up to scrutiny as these systems mature.
Authority, for machines, emerges from coherence and connectivity. Assertions gain strength not from volume, but from how well they’re reinforced and corroborated across the wider web. This is the layer where machines decide what to believe – and it’s where your competitive edge is forged.
And crucially, that network isn’t limited to what you publish. Machines are learning from everything – competitor sites, third-party aggregators, marketplaces, YouTube, Reddit, scraped product feeds. If a claim about your product, pricing, or positioning exists somewhere, it becomes part of the model.
That means your assertion strategy isn’t just about structure and markup on your site. It’s about managing how your brand is described, referenced, and understood across the web. If your competitors are making stronger, clearer, or more consistent claims – or if you’re absent entirely – the graph may not tip in your favour, no matter how polished your own pages are.
Worse, you have to assume that parts of the web are actively working against you. Competitors, affiliates, and bad actors aren’t just competing – they’re trying to pollute the models that learn from the web.
This hostile corpus – polluted SERPs, manufactured claims, weaponised contradictions – can distort your narrative, drown out your signals, and weaken your authority. You need to actively defend your assertions against a web that’s being manipulated in real time.
In a graph-shaped web, you’re not just publishing assertions. You’re defending them in an environment where every node and edge can be polluted, manipulated, or weaponised.
Google’s old mantra is no longer enough
For decades, Google has told marketers to “write for humans”.
It was smart advice when search engines were mostly matching queries to documents. Humans were the end audience, and optimising for their needs aligned neatly with how Google ranked results.
But the web isn’t human-first anymore.
Machines now sit between you and your audience, mediating discovery, filtering information, and making decisions on users’ behalf. They’re not just passively indexing; they’re actively interpreting and prioritising what matters.
The crucial shift:
- Humans judge the experience – how the design feels, how persuasive the story is, how the page resonates emotionally.
- Machines judge the structure of meaning – the individual assertions you make, how clearly they’re expressed, and how well they connect to other sources in their models.
This doesn’t mean abandoning human-focused content. It means recognising that Google’s mantra is incomplete.
You need to write for humans and engineer for machines – making your claims explicit, structuring relationships, and publishing in ways that feed both symbolic systems (search engines) and learned models (LLMs).
The brands that thrive won’t be those who only create great content; they’ll be the ones who encode their meaning in ways machines can trust, reuse, and amplify.
The assertion-first future is structural
If machines care about assertions, then how you express those assertions becomes a strategic choice.
Google gives us a narrow window into this: schema.org markup powers rich results, product listings, and event features. But structured data only works where Google has built functionality for it. Beyond that, its impact is limited. You can’t just “markup everything” and expect magic.
But that doesn’t mean structure doesn’t matter – far from it.
Machines learn from patterns, whether they’re formalised in schema.org or not. LLMs and agents don’t use schema.org directly, but they still encode the patterns they find in your content. Clean, predictable structures – at the HTML level, in your copy, in your layout – give them more reliable signals to work with.
This is where semantic HTML still matters. Not because models parse tags “semantically” in the W3C sense, but because semantic markup enforces clarity, hierarchy, and consistency.
- A
<section>with a clear heading creates a pattern that’s easy to learn. - Consistent phrasing (“Price: $99”) reinforces meaning more than prose buried in design.
- Hierarchical markup mirrors relationships, making those relationships easier to encode.
So while schema.org might be a narrow, feature-driven tool, structure in the broader sense – patterns in your HTML, language, and layout – is how you make meaning legible.
And in a world where machines mediate discovery, that legibility is what will decide whether your assertions are learned, trusted, and surfaced.
Stop optimising pages. Start optimising the graph.
This is why the old SEO playbook, where we optimised pages at URLs, no longer holds up.
Machines aren’t judging you by individual documents; they’re modelling the web of meaning around your brand. Your job isn’t to make pages rank – it’s to make your assertions clear, consistent, and so well-connected that machines can’t ignore them.
All of this leads to a simple but uncomfortable truth: the old SEO playbook – optimising pages at URLs – isn’t enough anymore.
Pages are still necessary. URLs still matter. Humans still need something to click on. But machines, which now mediate almost every discovery journey, aren’t judging you at the page level. They’re judging the network of assertions you’ve published and how it connects to everything else they know.
In this environment, authority isn’t something you sprinkle onto a page with keywords and links. It’s something that emerges from how your assertions fit into the graph:
- Are they consistent everywhere they appear?
- Are they reinforced by trusted nodes?
- Do they connect cleanly to other concepts and sources?
- Are they clear enough to be encoded and reused?
This is the real battleground.
Google, LLMs, and agents aren’t scoring individual documents; they’re building models of the world and deciding which nodes in that model they can trust. Your job is to make sure the nodes that represent you – your products, your brand, your expertise – are strong, connected, and coherent.
That’s what an assertion-first strategy looks like.
It’s not about creating more pages; it’s about engineering a network of claims that machines can rely on. The brands that thrive will be the ones who stop obsessing over how each page ranks, and start focusing on how their meaning is learned, reinforced, and surfaced across the machine web.
How to Optimise the Graph
If the web you’re competing in is no longer URL-shaped, then your strategy has to evolve with it.
Optimising individual pages was enough when search engines ranked documents. But in a graph-shaped web, what matters is how your assertions are learned, reinforced, and connected across the entire ecosystem.
So what does that actually look like in practice? It’s not about abandoning content creation or rewriting everything as JSON. It’s about making the things you say clear, consistent, and hard to ignore – wherever machines might encounter them.
Here are some tactics to get there:
- Avoid assertion spam: Don’t flood the web with contradictory or low-quality claims. Today’s naive models may still absorb some of it, but this strategy will backfire as systems mature.
- Design every page as a bundle of extractable assertions: Use clear, repeated patterns for key facts (prices, features, authorship, dates) instead of burying them in prose.
- Reinforce claims through multiple trustworthy sources: Make sure your products, pricing, and brand attributes are consistent everywhere: marketplaces, aggregators, partners, Wikipedia, YouTube descriptions, and press releases.
- Structure your content for learning, not just for humans: Use semantic HTML to give hierarchy and relationships. Add redundant context (“this is an SEO consultancy…”) so machines have multiple ways to pick up key claims.
- Publish machine-friendly endpoints: Offer APIs, clean feeds, and structured exports (JSON, XML) that expose your data clearly. This gives machines a source that doesn’t rely on scraping.
- Monitor and manage the hostile corpus: Track how competitors and bad actors describe you. Counteract misinformation with strong, corroborated claims on trusted properties.
- Leverage third-party validation: Encourage citations, reviews, and mentions on authoritative sites. Machines weight corroboration from diverse nodes more than self-assertion.
In other words, optimising the graph isn’t about gaming algorithms. It’s about making sure the things you say – and the things said about you – form a clear, connected network that machines can trust, today and as they get smarter.
The web after URLs
The URL was the foundation the web was built upon. It shaped how we designed sites, how we thought about SEO, and how we understood discovery itself. But foundations don’t last forever – they get built over, abstracted, replaced.
We’re already standing on a different layer.
Machines no longer navigate the web like we do. They traverse graphs, encode patterns, and build models. The page at a URL is still there, but it’s just a skin stretched over something deeper – a network of assertions that lives outside the document.
This is the web we’re optimising for now:
- Not a collection of pages, but a fabric of meaning.
- Not a structure of links, but a structure of relationships.
- Not a search engine serving documents, but a set of systems deciding what humans ever get to see.
So here’s the real challenge. Stop thinking about how to rank a page. Start thinking about how to exist in the models that are replacing the page.
Because those models are already shaping what the web is, and whether your brand matters in it.
Domains aren’t just web addresses — they’re brand assets, trust signals, and ownership anchors. Even if discovery shifts, having a domain means having control.
This is a fantastic article that highlights the big shift away from “URL” as the mainEntity and unique identifier of every topic. I’m going to have chatGPT summarize it for clients. 😉
Thank you! I was struggling how can I explain this shift to clients, this really helped!
Good job
Thanks