By Cory Collins
24 Jun 2016

Google Search AMA: Interview with Gary Illyes at SMX Advanced 2016

Gary Illyes, Webmaster Trends Analyst at Google, spoke candidly in an interview "AMA" with Danny Sullivan at SMX Advanced on Wednesday, June 22nd.

Gary Illyes has largely taken over the role of Google representative to the SEO community, since Matt Cutts' leave.

I was lucky enough to attend the event. Thanks to Page One Power for sending me, and to everyone I had the pleasure of speaking with—it was an incredible event.


The following is a loose transcription of the interview between Danny Sullivan and Gary Illyes, based upon the notes I was able to take, and the Facebook Live video SMX captured during the event.

I also found this YouTube video from the Growth Crew:

Note: we've covered this session in previous years as well, and it's interesting to reread the notes to see trends. Here's the notes from 2015, 2014, and 2013.

Full Interview of Gary Illyes AMA SMX Advanced 2016

I recommend watching the video of the event to capture the full detail. Seriously, it's worth an hour of your time, and there's nuance in the back and forth I won't be able to capture in the notes below.

The following are my notes, summarizing the questions and answers. I'd recommend using this to quickly understand the conversation, and then viewing the video in full.

Question One: RankBrain

Refinement versus ranking factor

Danny: Explain everything about RankBrain to us. It's the 3rd most important ranking factor, behind links and content.

Is it really a ranking factor? Isn't it more of a query refinement?

Gary: It's a ranking factor. It's a supervised machine learning algorithm.

Note: Gary explains what machine learning is, using the example of his water bottle and how we understand that is water, using patterns (water bottle shape, brand, says water, clear, etc.).

Gary: RankBrain looks at data from past searches, and based on worked it will try to predict what results will be good for a certain query.

RankBrain works much better for longtail searches and queries we've never seen before.

My favorite example is the Mario brothers without using a walk through. Before RankBrain we wouldn't identify the without as being important. We'd return the wrong results.

Danny: But isn't that query refinement? It sounds like RankBrain is better at understanding what the query is about, not returning awesome Mario sites.

Gary: It understand what results will work better for queries. It will also understand certain stop words (without in this example) shouldn't be dropped.

Danny: But isn't that influencing the query itself, not influencing the rankings?

Note: Danny goes on to give an example using apples and bananas.

Gary: The whole thing (RankBrain) is more about scoring the results than understanding the query. It's about what should be involved in scoring the results.

What you are saying (query refinement) is already handled by Hummingbird. We see the entities behind the words. With RankBrain we hopefully rank better results for queries we typically don't rank well.

(No) RankBrain Score, No RankBrain Optimization.

Danny: are there components associated with RankBrain? Is there a RankBrain score?

Gary: No, there is no RankBrain score.

The root of your question is whether you can optimize for RankBrain.

Danny: Yes. I'm trying to wrap my head around RankBrain, and how it fits into the algorithm.

Does RankBrain leverage your existing signals, or is it a new and unique signal in itself?

Gary: RankBrain is a new and unique signal.

You can't optimize for RankBrain.

RankBrain enhances our relevancy in the search results, based upon what you have in your pages. RankBrain makes sure the searches sees your page, if it fits for that query.

RankBrain will help us better rank natural language.

If you keyword stuff your content, it will almost certainly not be good for you.

Author Note: it sounds like RankBrain is Google's signal to ensure they better meet searcher intent.

RankBrain Growth & Stats

Danny: Last year Google told Bloomberg that RankBrain handled a large fraction of queries, and that it helps with 15% of queries you've never seen before.

What is the data on RankBrain now?

Gary: Nothing new to share in terms of RankBrain data or growth.

It will try to affect every query it gets, but in many cases we will see no changes in the results.

Danny: Should we fear that it will take over the world and destroy every life on earth.

Gary: Not every life.

Links, Content, RankBrain: Order of Importance

Danny: So RankBrain is the third largest signal. You've confirmed links and content are the two other leading ranking signals.

What's more important? Links or content?

Gary: There is no set order. Even the first and second depend on many things. There are perspectives where it will look as if links and content are the most important signals, but if you look from another perspective and you will see something completely different.

It could be RankBrain is the second most important - it just depends on the query and results.

I can't give you a concrete answer because it depends on way too many factors.

Question Two: Google assistant

Danny: What's the deal with Google assistant? You revealed it at Google I/O.

Gary: Frankly I have no idea. We're still trying to wrap our heads around how it will apply across our products.

A huge fraction of what powers Google assistant is machine learning. It's going to take trial and error.

Question Three: Machine-learning in the Algorithm

Danny: How much of the algorithm is to become AI-based? Is human-written code going to be replaced by machine learning? Where are we at now?

Gary: Machine learning is extremely important for us, and we are focusing a lot on machine learning, and experimenting with what can be accomplished with machine learning.

I can't tell you a percentage—I only have a vague idea and that might not be correct.

It's still very important that we can easily debug queries. Machine learning can prevent that. We don't want to get to a stage where someone sends us a bad query and we can't debug it.

We're cautious with machine learning, but we are experimenting with it.

Question Four: Keywords as the TLD

Danny: There was a back and forth recently about terms in the URL and ranking. Say I have a domain like

The word "news' in .news would have no impact in Google's understanding of the domain?

Gary: TLDs do not play a role in how we calculate relevancy in a specific piece of content or URL.

Danny: Do you look at a domain name at all?

Gary: There are certain cases where we would look at the domain name. But in most cases, no. I would not buy domain names that are incredibly keyword rich.

Danny: So don't expect it to be powerful?

Gary: Yes. It can help you if you have your brand name, or business, or product in your domain name. It helps your audience understand who you are.

But going for keyword rich TLDs isn't going to help.

If you bought, we wouldn't want to rank you higher for news, just because you have the TLD.

Question Five: Search Console (Google Webmaster Tools) Data

Danny: In September 2013 Google said you'd extend the amount of data in Search Console from 90 days to a full year.

Can we get a year's worth of data now, as promised?

Gary: We are still working on this. We've figured out how to make it happen.

The feedback we received recently at the Google dance definitely made it clear this is important, and it's something we're actively working on. 

Question Six: Search Console Data in Google Analytics

Danny: In May you started bringing Search Console into Google Analytics. It's not clear if that will archive, or if that will be kept in that rolling 90 days.

Gary: I think it's probably going to be rolling 90 days, not store.

Danny: We'd like that to be longer.

Question Seven: Penguin

Danny: The last Penguin update was in December 2014. When's the next update?

Gary: I won't say a date, because I've been wrong too many times. I've been told that's bad for business.

Danny: This year?

Gary: I will not say any time frame anymore.

I checked with the team. By now leadership is checking with the team.

Danny: So everyone impacted back in 2014 - they're still stuck with that.

Question Eight: Panda

Danny: Is it part of the core algorithm? Does it run in real time?

Gary: It is not real time. It's continuously running, but we have to collect the data and refresh it.

Danny: What's the refresh period?

Gary: Months. 

Question Nine: Secure Search

Danny: What's the stats? Are we going to get more of a ranking boost for secure?

Gary: Around 30% of search results are secure.

Danny: Is it going to be a stronger signal?

Gary: We're looking to see whether we can strengthen the signal. At this point I don't think it will happen anytime soon.

We want to make sure we're not sacrificing relevancy.

Question Ten: Google Mobile Updates

Danny: The last mobile update was in May. How'd that go? What's the stats? Future plans?

Gary: I don't think there will be a next Mobile update. There will be more iterations and we'll probably add more signals.

Pagespeed, for example. It's way way way more important on mobile than on desktop. So in the future we may add that into the Mobile update.

Question Eleven: Social Signals in Search

Danny: So you still don't use social signals in search right?

Gary: Correct - we do not use social signals.

We have a problem with social signals. 3rd parties can pull the plug. We need reliable signals.

Danny: Speaking of which, Google+.

Google+ used to greatly impact personalized search. That doesn't seem to be the case anymore. Is that true?

Gary: We often run many experiments in search, and sometimes have to roll things back.

Things in search change.

By the way, we're not using authorship at all anymore.

Danny: At all?! Not even in in-depth articles?

Gary: No.

Danny: So you have no idea who authors a page anymore?

Gary: Of course we do. We're smarter than that - we don't need the authorship tag.

Question Twelve: Voice Search

Danny: Can we have data on how many people came to our site with voice search?

Gary: Possibly. We're either already working on it, or looking into how we could provide the data.

Question Thirteen: Tabbed Content

Danny: Tabbed content. Do you love it or hate it? Should we stop using it?

Gary: No - it's good. 

Question Fourteen: Multivariate Test

Danny: When we serve different versions of a page to test copy - how can we make sure that's not perceived as cloaking?

Gary: Make sure you don't do that for many many months. Do it for a reasonable amount of time.

Question Fifteen: Engagement Signals

Danny: What's a healthy rate of return visitors to a site?

Gary: What? I have no idea - it depends on the site.

Danny: Well people freak out about bounce rate. Should they?

Gary: Probably not. 

Danny: Do you use click through rate in rankings?

Gary: No.

Danny: No? Not even a little bit?

Gary: Well if you mean like bounce rate from analytics then no.

Danny: Do you look at when people click on a website and then they bounce back to the search results?

Gary: We use clicks and CTR for very specific things at Google.

For example, personalization. If you search for "apple" then at first we won't know what you mean. Do you mean the fruit or the company?

If you frequently click on results for the company, then we're reasonably certain you mean the company. 

For experiments, CTR is incredibly good because we have a controlled set of users we can look at.

Question Sixteen: Suggested Search and Politics

Danny: Does Google manipulate their auto suggest results based upon their own political disposition?

Gary: No.

Question Seventeen: RankBrain and Longtail

Danny: Why does RankBrain work better for longtail than head terms? What's the difference?

Gary: Typically the longtail queries are the ones we've never seen before.

We learn from our own failures, and if we don't have historical signals that we can look at, then we may not rank those as well.

With RankBrain, for example, we don't necessarily need to wait for links to accumulate.

We can just say "okay, that kind of results works very well for users."

Question Eighteen: HTTPS & HTTP/2

Danny: How strongly do you recommend migrating to HTTPS 2?

Gary: HTT2 - I would look at what it would take for you to implement it, but I wouldn't rush it. Make sure you have the ability to roll back to the old version as well.

There are browsers that definitely will not handle it well.

How strongly do I recommend moving to HTPPS?

I leave it you to decide how much you want to care about your users and the integrity of your content. 

Question Nineteen: PayDay Loans in Organic

Danny: With the payday loans ad bans, organic is still free for all. What are you doing to make sure that stuff is better?

Gary: I don't know. I have absolutely no idea. 

Question Twenty: UX (user experience)

Danny: Google always talks about great content. What more can we do to create a good user experience?

Gary: Make sure people know about you.

Figure out what your audience wants and provide that for them.

Do make sure your pages load freaking fast. Really, really fast. Median load time for AMP is less than one second. That's what users expect.

Question Twenty One: Footer Links

Danny: We own a number of sites and we link to them in the footer. How do you deal with that?

Gary: It's fine - I don't personally care.

Danny: you're not going to penalize them?

Gary: Just make sure you're doing things within reason. Don't overdo it. 

Question Twenty Two: XML sitemap file size

Danny: Does XML sitemap file size impact indexation?

Gary: No. There's a hard limit of something like 10MBs, but no. It shouldn't effect indexation.  

Question Twenty Three: RankBrain

Danny: You talked about RankBrain. Is there anything more to add?

Gary: Probably not. I don't work on RankBrain — I have a high level understanding of what it does. I've shared most of what I know. 

Question Twenty Four: Different Ranking Signals in Different Industries

Danny: Do you have different ranking factors in different industries?

Gary: No. 

Question Twenty Five: Separate Mobile Index

Danny: Is there going to be a separate mobile index?

Gary: Yes — we're working on it. 

Question Twenty Six: Signals in Determining User Satisfaction

Danny: What signals do you look at to try and determine if a user likes a site?

Gary: Over 200 (referring to ranking signals). 

Question Twenty Seven: Matt Cutts

Danny: Where is Matt Cutts?

Well, he just took his second leave from Google to go protect the US, right?

Gary: Yup. 

Question Twenty Seven: Important Future Developments

Danny: I'd like you to end by telling me what overlooked feature of development SEOs should know about or be thinking about.

Gary: Two things:

  1. Pay attention to AMP, because it's going to be big.
  2. Assistance and chatbot development.

These developments are going to be big. Many people have invested heavily into making these things important. 

You'll want to be among the first to adopt these developments. 

Cory Collins

Cory Collins is the Business Development Manager at Page One Power and has been with the agency since 2012. Cory is an SEO strategist, writer, runner, and outdoor enthusiast residing in Boise, Idaho, with his wife, daughter, and (too) many pets.