Google’s new BlockRank research (covered in Search Engine Journal) shows how large language models can rank web pages by reading a query alongside multiple documents and determining which one best answers what the user’s looking for.
DeepMind calls this In-Context Ranking (ICR), and it represents a real shift toward evaluating pages based on meaning, not just keyword matching.
But this isn’t the death of traditional SEO. Not even close.
The Practical Reality
Even the most powerful LLM can’t analyze billions of pages in real time. So BlockRank has to work with what the research calls “candidate documents” — a shortlist that’s already been filtered before the model ever sees it.
And how do pages make that shortlist? Almost certainly through the same traditional signals we’ve always worked with:
- Can Google actually crawl and index your page?
- Does your content match the language and intent of the query?
- Do you have decent authority and link quality?
- Is your content fresh and relevant?
If you’re not in that initial candidate pool, it doesn’t matter how semantically perfect your content is. The LLM never sees it.
Where Things Change
Once pages make it into the candidate set, BlockRank evaluates them differently. While the research doesn’t spell out every mechanism, we can infer from how LLMs work and from Google’s other neural ranking systems (PassageRank, BERT, MUM) that it’s likely looking at:
- Entity salience: which entities are central to the topic and how well you cover them
- Passage-level relevance: which specific sections directly answer the query
- Semantic precision: whether your content means what the searcher needs, not just matches their words
This is optimization for understanding, not signals. Does your content actually resolve the question behind the search?
It’s a New Layer, Not a Reset
Think of it like this:
- Retrieval layer: traditional SEO gets you into the room
- Reasoning layer: semantic quality determines your seat at the table
Technical fundamentals still matter. Crawlability, site architecture, authority — these open the door. Semantic clarity, entity coverage, and contextual relevance decide who ranks first.
What This Means
If BlockRank or something like it goes live, it’s not the end of traditional SEO. It’s the evolution of what many of us already think of as “semantic SEO” — writing for meaning, not just matching.
You’ll need both layers working: a technically solid site that can get indexed and retrieved, plus content that an LLM would immediately recognize as the best answer.
The bar just got higher.
Will Scott 