I’ve been doing SEO since 2018. In those years, a lot has changed but one element has gone through a surprisingly quiet evolution in terms of influence: the URL.

It was never the star of the show. Alongside meta descriptions, URLs have always been treated as second-tier signals: not as decisive as H1s, meta titles, or backlinks. But the way they matter has shifted over time, and the AI search era is making them relevant again in a way most people aren’t paying attention to (at least not yet).

A brief history of URL structure

Before SEO, URLs reflected the physical file structure on the server. /products/product-1.html was about organizing folders, not communicating meaning. The goal was to make the site work.

When SEO took over (around 2005), the assumption became that keywords in URLs helped rankings. Practitioners started building structures like /brand-name/womens-sunglasses and /brand-name/mens-sunglasses. The rule of thumb: keep it short, use hyphens, include the main keyword.

Then came the disinterest era (roughly 2020–2025). Google repeatedly stated that URLs were a minor signal compared to content, links, and other elements. Many teams stopped caring. Shorter was better. Product codes were fine.

Now we’re in the AI era. And the calculus has changed again.

What an AI sees before reading your page

LLMs and AI search engines don’t just read page content but they build a representation of a page from multiple signals at once. The URL is one of them, and it gets processed early.

Here’s a real example from an eyewear e-commerce store. Same product, two different URLs:

Old URL: /products/pr--a10s-black-16k08g

New URL: /products/prada-sunglasses-man-pr-a10s-black-16k08g

Before a crawler reads a single word of the page, here’s what it can infer from the URL alone:

Signal	Old URL	New URL
Brand	❌ Unknown (pr could be anything)	✅ Prada
Category	❌ Unknown	✅ Sunglasses
Gender	❌ Unknown	✅ Man
Model	⚠️ Partial (a10s)	✅ PR-A10S
Color	✅ Black	✅ Black

With the old URL, the only thing a bot can extract before loading the page is the color. Brand, category, gender — all require parsing the HTML.

With the new URL, the bot arrives already knowing it’s looking at men’s Prada sunglasses, PR-A10S model. The URL acts as a pre-context signal.

Why this matters for AI systems

A well-structured URL works like a document title: it sets expectations before the content loads. For AI systems processing thousands of pages at scale, this pre-context matters in two concrete ways:

Relevance reinforcement. A URL that matches the query intent strengthens the content match. /prada-sunglasses-man reinforces relevance for “men’s Prada sunglasses” even if the title and H1 say the same thing — because it’s one more consistent signal pointing in the same direction.

Entity extraction. Brand names, product categories, and attributes in the URL are easy named entities for a model to pick up and index. A product code like pr--a10s forces the model to resolve the entity from page content instead. More work, more room for ambiguity.

This isn’t a ranking factor in the old SEO sense. It’s about reducing ambiguity and cognitive load for the model — which translates to more reliable extraction and potentially better representation in AI-generated answers.

What to actually change

A few principles that hold up in the AI search era:

Lead with the brand or category, not the product code. /prada-sunglasses-man-pr-a10s tells a story. /pr--a10s doesn’t.

Keep it human-readable. If a person can’t parse the URL at a glance, neither can an AI. Avoid double dashes, truncated codes, and internal naming conventions that mean nothing outside your CMS.

Don’t keyword-stuff. /buy-cheap-prada-sunglasses-online-men-black is the old SEO playbook. It doesn’t help AI systems — it creates noise. Stick to what’s genuinely descriptive: brand, category, gender, model, color.

Be consistent across your catalog. If your URL structure varies across product types or brands, AI systems will struggle to learn the pattern. Consistency improves parsability at scale.

Don’t change established URLs just for this. If you have URLs with backlinks and traffic, the redirect cost outweighs the benefit. Apply these principles to new pages and migrations — not as a retroactive cleanup project.

---

Like llms.txt and most things in the AEO and GEO space, this is still early. We don’t have hard data proving that cleaner URLs move the needle in AI search rankings. But here’s the thing: there’s no evidence they hurt either — and the underlying logic is sound.

My take is that URL structure will become one of those signals that everyone ignores until it’s too late to fix at scale. I’ve already started cleaning up ours. Have you?