Next Google Search Evolution - BERT Models

[rustybrick]

---

System: The following 3 messages were cut out of thread at: https://www.webmasterworld.com/google/4966830.htm [webmasterworld.com] by goodroi - 6:56 am on Oct 25, 2019 (utc -5)

---

Google has confirmed a massive change impacting 1 in 10 of all queries named BERT [blog.google...]

[indigojo]

1st post in 6 years ! One take of mine on this is that news content will have an even shorter shelf life and evergreen content that answers the Q will stand longer. That's not to say news is a waste of time as the spikes might be bigger but shorter lived. Thoughts ?

[JesterMagic]

Not sure why they couldn't tell us when the update was rolling out instead of letting everyone know at the end of it.

Bert is not going to help with a thing with organic search if Google continues to throw up a wall of ads and display their own widgets based on other people's information.

Bert doesn't fix the underlying problems of Google choice for what sites ranks first. Sure it might understand your query language a bit better (and I hope the language in the content) but it still is choosing to pull content from fluff pieces (at least in our niche) of large media companies who are desperately trying to write articles for everything to gain traffic. I think this is why this update wasn't really noticed by the community at large.

Google also doesn't understand the content itself or if it is just someone spouting generalities. For example, an old service in our niche closed down over a year ago. They notified their visitors months in advanced of this and posted this on their social media. A few sites like ours who knows the industry reported on this. The site is since gone. You would expect in the search results now for a query of this service name that their social media pages and other niche news sources like ours would be reported first. Instead you have larger media organizations or competitors who play around with dates still have old content that talks about the service like it still exists, ranking about this service first, followed by other untrusted new players who have creating websites using domain names with the service name in it.

[vivalasvegas]

“2019 brazil traveler to usa need a visa."

I for one find it curious that only in 2019 Google manages to understand the query above correctly. I guess this proves how slowly AI is evolving.

[glakes]

I for one find it curious that only in 2019 Google manages to understand the query above correctly. I guess this proves how slowly AI is evolving.

Many believe AI's primary purpose is to manage profitability by organizing the search results in such a way that more clicks are directed towards paid ads. The fact that only now will AI understand such basic expressions gives weight to that theory.

[samwest]

Well, bounce rate suddenly jumped 22% FNFR, BERT must now be online in my niche.

[EditorialGuy]

Many believe AI's primary purpose is to manage profitability by organizing the search results in such a way that more clicks are directed towards paid ads.

Search evolves (mostly in increments, rather than in giant leaps). Some site owners benefit from that evolution, and some don't, but site owners aren't the target audience for search engines.

Google's stated mission is "to organize the world's information and make it universally accessible and useful," not to help us make money by piggybacking on search.

[Solipsism]

I already see a lot of articles talking about the non-effect of Bert, but how do we know if this update is rolling out ? It was only announced Friday to inform the community, it doesn't mean the algo is on ?!

[rustybrick]

BERT starting rolling out a week ago Monday and finished rolling out the end of last week.

[glakes]

Google's stated mission is "to organize the world's information and make it universally accessible and useful," not to help us make money by piggybacking on search.

There's no need to piggyback off of search as few shoppers use Google anymore. Fortunately Amazon does a much better job and as part of Google's evolution they have become nearly irrelevant in ecommerce.

[EditorialGuy]

There's no need to piggyback off of search as few shoppers use Google anymore. Fortunately Amazon does a much better job and as part of Google's evolution they have become nearly irrelevant in ecommerce.

You'd never know that from reading this forum. :-)

BERT starting rolling out a week ago Monday and finished rolling out the end of last week.

Isn't it a long-term work in progress, though? More of an AI-based process than a "update?"

[rustybrick]

EditorialGuy,

I asked Google, they said it will finish rolling out "EOW" last week. Again, 1 in 10 queries end up using BERT. Will that grow? It depends how user queries change over time?

[EditorialGuy]

Rustybrick: My point was simply that, with artificial intelligence, machine learning is (or can be) an ongoing process. As more data is crunched and new data is pulled into the mix, the results of that process are likely to change even if the algorithm behind that process has been fully rolled out.

Then again, Google Search (like most search engines) has always been a work in progress.

[Selen]

Just like ads have the most relevant titles and descriptions for commercial searches, I expect BERT could do the same to informational queries. Ie. the answer boxes would have the most relevant and comprehensive information that would not require a click to the site and the "organic" results would be general and semi-relevant to minimize their click-through rates. At the same time, there may be a push towards more ads on informational search queries.

[EditorialGuy]

Just like ads have the most relevant titles and descriptions for commercial searches, I expect BERT could do the same to informational queries. Ie. the answer boxes would have the most relevant and comprehensive information that would not require a click to the site and the "organic" results would be general and semi-relevant to minimize their click-through rates.

First of all, answer boxes are like Cliff's Notes: They're useful for people who want quick answers to simple questions, but there' s only so much information that can be crammed into an answer box. What's more, an answer box's packaging and presentation aren't going to be as good as a well-designed and visually appealing page.

As for the notion that Google is going to intentionally dilute or destroy its core product (search) with "general and semi-relevant results" to minimize clickthrough rates, that's what I'd call far-fetched. It's the kind of thinking that owners of churn-and-burn sites may find productive, but there's no evidence that Google is prone to short-term thinking.

At the same time, there may be a push towards more ads on informational search queries.

There already are ads for many informational search queries. Whether there are or aren't on any given SERP is almost certainly dictated by supply and demand, not whether there's a "push" by Google. Remember, most people who are searching for information on, say, how to fry an egg aren't shopping for eggs or frying pans. And most information sites that have editorial pages about how to fry an egg can't justify spending money on ads to promote those pages. (Mind you, there may be some people who still have dreams of getting rich quickly with "click arbitrage," just as there still are people who wear spats and sport "I like Ike!" bumper stickers.)

[brotherhood of LAN]

First of all, answer boxes are like Cliff's Notes: They're useful for people who want quick answers to simple questions, but there' s only so much information that can be crammed into an answer box. What's more, an answer box's packaging and presentation aren't going to be as good as a well-designed and visually appealing page.

IMHO it's a poor analogy to use something like cliff notes (not a fan of analogies anyways TBF). You can take the view that the search engine delivers your 'answer' back (i.e. the cliff note you need out of billions of them) or you take the view a search engine is merely a portal to the rest of the web.

For the former, if the search engine chooses to offer answers on aggregated data, they'll choose the right cliff note and what data to put in it, it's not really a question of how much data can be on one. Depends on the query of course, because if I'm searching for a book title, then they're simply only able to provide metadata rather than the information I'm potentially wanting to absorb.

Google is clearly getting better at the edge cases of NLP, and better at aggregating it into a more generalised knowledge ontology. Good points and bad points to it, personally I think as long as there's alternatives ways to search, then by all means let this be another way. Ultimately it's a trade off between intentions of the user and practicality (dozens of synonyms, choice of phrasing) normalised into results, all the way to the other end of the scale where a searcher wants purely boolean search results.

[JesterMagic]

Also answer boxes don't attempt to just answer the question from the query (if their is even one asked), they list a bunch of similar questions and answers, while each answer is only a sentence or 2 when you take them all together this starts to add up.

This is the same with structured data for things like faq questions and forum posts. These rich snippets display a lot more information.

[glakes]

You'd never know that from reading this forum. :-)

I think this is a matter of perception. While some think webmasters are trying to piggyback Google's search results, those on the other side of the fence believe Google has monetized their hard work and are giving them very little in return. In my industry, our goal is to manufacture the best/highest quality products on the market. I'm of the belief that if Google wants to push cheap Chinese junk to the USA consumer, then that is their problem. Eventually some of those consumers will become our customers when their Chinese products fail. Some of them have mentioned that they could not find us in Google, which our official response is for them to try a different search engine.

[EditorialGuy]

You can take the view that the search engine delivers your 'answer' back (i.e. the cliff note you need out of billions of them) or you take the view a search engine is merely a portal to the rest of the web.

Or, as in the case of today's search engines (not just Google), a SERP can give an "answer" when appropriate and be a portal to the Web, to images, to videos, etc. There's no requirement that a SERP should look the same in 2019 as it did in 1998.

[aristotle]

One or two very recent posters have veered far away from the original topic of this thread.

[martinibuster]

Well if BERT comes up with a better interpretation of a search term, then it should mean that a different group of sites will be competing for the top spots in the organic results for that particular term.

Presumably the previously ranked sites were less relevant. So it means that sites that are already about a topic can now be recognized and ranked.

So if you can anticipate some of the new search terms that your site can now compete for, then you can enhance your site's treatment of its information related to those terms.

I'm having trouble thinking of how one would go about that.

The problem wasn't on the publisher side. The problem was on Google's side. So what BERT does is improve Google's side, thereby bridging the gap between understanding the query and being able to spot the relevant sites.

There's nothing to optimize for because it's not about your site or your content.

[iamlost]

As has been noted by several above, and by G spokes-folk elsewhere, this is an internal coping mechanism change to better comprehend queries and, possibly, but not necessarily, better match answers. All that publishers can do, should they choose to accept the challenge, is explicitly help train the algo without acknowledgement or recompense.

Given what has been trialled/published with BERT to date one or more of the following usage is probable:
* more reliable prediction labelling in named entity model training.

* an added classification level to transformer output for more reliable sentiment analysis of query/sentence.

* more reliable Q&A model training.

RankBrain has been increasingly failing since broad rollout, at the very least BERT is being used to prop up, clean up that mess; given the three additional strengths mentioned I believe this is the biggest Google search change since Hummingbird. If it isn't a RankBrain rabbit hole.

[aristotle]

You might look in GSC and notice new search terms that your site is starting to get impressions for. Of course BERT might or might not be the cause, but whatever the reason, you might want to enhance your site's treatment of the relevant information if there appears to be potential for bringing additional traffic.

[shadowlight]

Search evolves (mostly in increments, rather than in giant leaps). Some site owners benefit from that evolution, and some don't, but site owners aren't the target audience for search engines.

Google's stated mission is "to organize the world's information and make it universally accessible and useful," not to help us make money by piggybacking on search.

Yet google "piggyback" on all the content, information, services and products that site owners provide which all help google to make money. Site owners who publish content are the reason that google search even exists.

[Kendo]

but I'm not seeing how BERT significantly changes SEO best practices

Just another layer of smoke to disguise the fact that they manipulate rankings to suit their own ends, favouring their own products and services that have been plgiarised/reverse engineered from those of their users/clients, favouring local industry and favouring those who pay big bucks for Adwords.