Thread by @Marie_Haynes

#LearnSEO is going to be different this week. Here's a long thread with my thoughts on

➡️entities
➡️semantic search
➡️E-A-T
➡️content quality

There will be some questions in the thread, but this isn’t a typical Q1/A1 chat.

I don’t claim to have all the answers here, but want to generate discussion. My opinion is that SEOs who understand semantic search have a big advantage. We should not overlook it!

I’ll end the tweet thread by asking for discussion on how we can practically use semantic SEO to help our clients better.

“Semantic search” sounds like a buzzword. I think many SEOs paid no attention to semantic search bc of the negative talk surrounding “LSI keywords”. LSI keywords = sprinkling semantically related kwds (synonyms & related words) throughout content. Semantic search is not this.

Semantic search understands the intent of a query *and* the meaning of terms on the page. Rather than matching keywords in queries with keywords on pages, a semantic search engine can understand the meaning of terms in the query and in content.

So, is Google a semantic search engine? I think the answer to this is, they’re getting there.

In 2013 Hummingbird incorporated semantic search capabilities into the algorithm.
web.archive.org/web/2015050322…

The algorithm could now be less reliant on PageRank and more reliant on “other factors like whether Google believes a page is of good quality”.

PageRank didn’t go away. G’s guide to fighting disinformation (2019) says PR is still closely related to Trust and Authority.

But links pointing to a page don’t specifically tell us whether content is of good quality.

blog.google/documents/37/H…

What does Google consider “good quality content”?
They’ve got a blog post on it:

developers.google.com/search/blog/20…

With PageRank, Google could say, “This query contains these keywords” & “These keywords are present on these pages” & “These pages have links from authoritative sites, so we can trust they’re likely relevant and trustworthy.”

This doesn’t answer the quality questions above tho.

If Google could extract entity information from pages (facts, topics, relationships between them) I believe they can answer many of the questions above if not all of them.

However, this is expensive and difficult. But I think we're getting there.

Algorithms can determine whether facts (entities) in content are consistent with the real world.
youtube.com/clip/Ugkxy93C3…

When we use structured data, we’re pointing out entity information for Google to use in / verify across the knowledge graph. This is likely a huge component of E-A-T - Google determining you and your authors are authentic.

But can Google extract “unstructured data” from pages?

In 2015, we saw RankBrain (AI). RankBrain helped Google rewrite queries to better understand the concepts within them.

@bill_slawski

We saw SERP turbulence with RankBrain. @glenngabe initially called these Phantom updates. Google eventually said we should call these “Quality updates”

searchengineland.com/the-quality-up…

The recommendations were to focus on the Panda questions from Amit Singhal.

These are very very similar to the core update questions.

developers.google.com/search/blog/20…

How could G better understanding a query help determine whether a page provides original content or has substantial value as compared to competitors?

Danny Sullivan (before working for G) theorized that RankBrain was able to classify pages based on the content they contain.

I think that with each advancement in NLP/AI Google gets a little closer to being a semantic search engine.

BERT, released in October of 2019 was a huge advancement. blog.google/products/searc…

It’s interesting to note that a couple of weeks later there was an unnamed big update that affected many sites doing “whitehat” link building.

mariehaynes.com/november-8-201…

I think now that this was G relying a little more on semantic search over PageRank.

Advances in NLP have made it now that instead of being able to extract entities from sentences, they can extract from paragraphs. This is big. Google docs AI summaries use NLP to extract concepts from entire documents.

Neural matching extracts entities from both queries and pages to better understand the concepts they represent.

MUM extracts entity information from text, images, video and across different languages.

In a recent article, @olafkoff speculated MUM could be “Google's next step toward becoming a purely semantic search engine”

searchengineland.com/google-mum-upd…
searchenginejournal.com/google-search-…

This is all nice information, but what does it mean to SEOs? Is there actual value in understanding more about semantic search?

https://twitter.com/I___DEREKflint/status/1516396881798123527?s=20&t=3gruV_-eGTxZ-sBKhxn3Pw

Understanding semantic SEO gives us two advantages:
1) If we know what topics are closely associated with our entities, we can create substantial content (and content hubs.)
2) SEOs who understand NLP know how to write sentences that are easier for AI to extract info from.

Can we get hints as to what entities are closely related to our topics by looking at the SERPS? I’m betting that content that answers these questions thoroughly and factually would do well.

You can use Google’s NLP tool to test your sentences/paragraphs and see if AI can extract the entities you are trying to rank for. If not, keep tweaking.

cloud.google.com/natural-langua…

.@KorayGubur is doing some great stuff with semantic search. Koray, could you perhaps link a couple of your articles/videos that you think would be helpful for people who want to know more about how to use knowledge of semantic search?

@KorayGubur .@thetafferboy is doing cool work using People Also Ask Results. alsoasked.com allows you to map relationships between topics. It’s really helpful.

There are many others I could tag as well.

An entity or semantic first approach to content creation can simply make your content better... Less focus on including keywords, and just writing for humans. If G can pick out these relational properties in text, then keyword inclusion shouldn't matter as much.

What frightens me a little is this thought: If AI can parse entities from text and AI can also write text, AI is likely going to be able to create relevant high quality content. The world of content is going to change.

https://twitter.com/Marie_Haynes/status/1516132805587378176?s=20&t=60Hj36NCqC9v_g2hld0aXQ

I’m incredibly excited about optimizing better based on better understanding G’s use of entities in search.

I’m also open to discussion as there are many others who know much more on this topic than I.

What do you think about semantic search? Are you using it?

/#learnseo

apologies, @Olaf_Kopp! I am so sorry I got your name/handle wrong🫢