By Joseph Mas
This article explains why full URLs behave very differently for search rankings versus LLM understanding.
Recent research shows that large language models can generate and use URLs as explicit retrieval identifiers rather than treating them as ordinary text. This supports the idea that URLs function as resource pointers inside retrieval and grounding systems. You can find the original research here: Large Language Models are Built in Autoregressive Search Engines.
Note: The paper supports URLs as retrieval identifiers. It does not directly test Google ranking behavior.
The research paper is seventeen pages long, so I broke it down to cut through the noise:
Google’s view (ranking layer)
- Naked URLs are generally neutral to weak signals
- Google prefers descriptive anchor text
- Naked URLs pass link equity but add little semantic relevance•
- Overuse can signal auto generated content, low editorial quality, or poor user experience
Its imperative to also consider the context and surrounding text matter a lot to search engines too, not just the specific anchor text. Naked links in good contextual sentences still pass semantic clues. Actually, surrounding context really needs to be considered in all cases – in my humble opinion.
For ranking I believe it can safely be said; Google prefers language, not raw pointers.
LLM ingestion view (grounding layer)
- Naked URLs are strong signals
- They function as explicit document identifiers
- They create hard source boundaries
- They act as unambiguous reference anchors
- They reduce entity confusion, attribution drift, and blended or inferred sources
For ingestion, models want certainty, not prose.
Anchor Text and Naked URL Functionality Delta
Anchor text helps systems decide what something means. Naked URLs help systems decide what something is.
Meaning versus identity.
Real World Application
How this is applied is ultimately a design choice. For consumer facing content, descriptive anchors usually make sense for readability and user experience. For scientific, technical, legal, or professional documentation, full URLs can be valuable as explicit, unambiguous references where clarity, attribution, and long term machine interpretation matter more than presentation.
My summarized take on this
- Naked URLs are weak for Google ranking
- Naked URLs are strong for LLM ingestion
- Using both intentionally enables layer aware optimization
Here is my professional opinion: Naked URLs are not universally better or worse. They simply serve a different purpose, and knowing when to use them is part of modern optimization.
Honestly, I just did two hours of research just to reach the conclusion I already knew before I started. I’m trying to convince myself I didn’t just waste time finding empirical evidence… This was actually a very boring topic for me, but it’s been rolling around in the back of my head for over a decade. Now I can let it go 🙂