Web technologies and their adoption are advancing at a frenetic pace. Content is a game that every type of team and agency plays, so we’re all competing for a piece of that pie. Meanwhile, technical SEO is more complicated and more important than ever before and much of the SEO discussion has shied away from its growing technical components in favor of content marketing.
As a result, SEO is going through a renaissance wherein the technical components are coming back to the forefront and we need to be prepared. At the same time, a number of thought leaders have made statements that modern SEO is not technical. These statements misrepresent the opportunities and problems that have sprouted on the backs of newer technologies. They also contribute to an ever-growing technical knowledge gap within SEO as a marketing field and make it difficult for many SEOs to solve our new problems.
That resulting knowledge gap that’s been growing for the past couple of years influenced me to, for the first time, “tour” a presentation. I’d been giving my Technical SEO Renaissance talk in one form or another since January because I thought it was important to stoke a conversation around the fact that things have shifted and many organizations and websites may be behind the curve if they don’t account for these shifts. A number of things have happened that prove I’ve been on the right track since I began giving this presentation, so I figured it’s worth bringing the discussion to continue the discussion. Shall we?
An abridged history of SEO (according to me)
It’s interesting to think that the technical SEO has become a dying breed in recent years. There was a time when it was a prerequisite.
Personally, I started working on the web in 1995 as a high school intern at Microsoft. My title, like everyone else who worked on the web then, was “webmaster.” This was well before the web profession splintered into myriad disciplines. There was no Front End vs. Backend. There was no DevOps or UX person. You were just a Webmaster.
Back then, before Yahoo, AltaVista, Lycos, Excite, and WebCrawler entered their heyday, we discovered the web by clicking linkrolls, using Gopher, Usenet, IRC, from magazines, and via email. Around the same time, IE and Netscape were engaged in the Browser Wars and you had more than one client-side scripting language to choose from. Frames were the rage.
Then the search engines showed up. Truthfully, at this time, I didn’t really think about how search engines worked. I just knew Lycos gave me what I believed to be the most trustworthy results to my queries. At that point, I had no idea that there was this underworld of people manipulating these portals into doing their bidding.
SEO was born of a cross-section of these webmasters, the subset of computer scientists that understood the otherwise esoteric field of information retrieval and those “Get Rich Quick on the Internet” folks. These Internet puppeteers were essentially magicians who traded tips and tricks in the almost dark corners of the web. They were basically nerds wringing dollars out of search engines through keyword stuffing, content spinning, and cloaking.
Then Google showed up to the party.
Early Google updates started the cat-and-mouse game that would shorten some perpetual vacations. To condense the last 15 years of search engine history into a short paragraph, Google changed the game from being about content pollution and link manipulation through a series of updates starting with Florida and more recently Panda and Penguin. After subsequent refinements of Panda and Penguin, the face of the SEO industry changed pretty dramatically. Many of the most arrogant “I can rank anything” SEOs turned white hat, started software companies, or cut their losses and did something else. That’s not to say that hacks and spam links don’t still work, because they certainly often do. Rather, Google’s sophistication finally discouraged a lot of people who no longer have the stomach for the roller coaster.
Simultaneously, people started to come into SEO from different disciplines. Well, people always came into SEO from very different professional histories, but it started to attract a lot more more actual “marketing” people. This makes a lot of sense because SEO as an industry has shifted heavily into a content marketing focus. After all, we’ve got to get those links somehow, right?
Naturally, this begat a lot of marketers marketing to marketers about marketing who made statements like “Modern SEO Requires Almost No Technical Expertise.”
Or one of my favorites, that may have attracted even more ire: “SEO is Makeup.”
While I, naturally, disagree with these statements, I understand why these folks would contribute these ideas in their thought leadership. Irrespective of the fact that I’ve worked with both gentlemen in the past in some capacity and know their predispositions towards content, the core point they’re making is that many modern Content Management Systems do account for many of our time-honored SEO best practices. Google is pretty good at understanding what you’re talking about in your content. Ultimately, your organization’s focus needs to be on making something meaningful for your user base so you can deliver competitive marketing.
If you remember the last time I tried to make the case for a paradigm shift in the SEO space, you’d be right in thinking that I agree with that idea fundamentally. However, not at the cost of ignoring the fact that the technical landscape has changed. Technical SEO is the price of admission. Or, to quote Adam Audette, “SEO should be invisible,” not makeup.
Changes in web technology are causing a technical renaissance
In SEO, we often criticize developers for always wanting to deploy the new shiny thing. Moving forward, it’s important that we understand the new shiny things so we can be more effective in optimizing them.
Over time more and more folks would examine this idea; ultimately, a comment from this ex-Googler on Hacker News would indicate that this has long been something Google understood needed conquering:
This was actually my primary role at Google from 2006 to 2010.
The really interesting parts were (1) trying to make sure that rendering was deterministic (so that identical pages always looked identical to Google for duplicate elimination purposes) (2) detecting when we deviated significantly from real browser behavior (so we didn’t generate too many nonsense URLs for the crawler or too many bogus redirects), and (3) making the emulated browser look a bit like IE and Firefox (and later Chrome) at the some time, so we didn’t get tons of pages that said “come back using IE” er “please download Firefox”.
I ended up modifying SpiderMonkey’s bytecode dispatch to help detect when the simulated browser had gone off into the weeds and was likely generating nonsense.
However, regarding SEO, the key difference between Angular and React is that, from the beginning, React had a renderToString function built in which allows the content to render properly from the server side. This makes the question of indexation of React pages rather trivial.
AngularJS 1.x, on the other hand, has birthed an SEO best practice wherein you pre-render pages using headless browser-driven snapshot appliance such as Prerender.io, Brombone, etc. This is somewhat ironic, as it’s Google’s own product. More on that later.
View Source is dead
To illustrate the point further, consider this View Source view of Seamless.com. If you look for the meta description or the rel-canonical on this page, you’ll find variables in the place of the actual copy:
If instead you look at the code in the Elements section of Chrome DevTools or Inspect Element in other browsers, you’ll find the fully executed DOM. You’ll see the variables are now filled in with copy. The URL for the rel-canonical is on the page, as is the meta description:
Since search engines are crawling this way, you may be missing out on the complete story of what’s going on if you default to just using View Source to examine the code of the site.
HTTP/2 is on the way
One of Google’s largest points of emphasis is page speed. An understanding of how networking impacts page speed is definitely a must-have to be an effective SEO.
Before HTTP/2 was announced, the HyperText Transfer Protocol specification had not been updated in a very long time. In fact, we’ve been using HTTP/1.1 since 1999. HTTP/2 is a large departure from HTTP/1.1, and I encourage you to read up on it, as it will make a dramatic contribution to the speed of the web.
Quickly though, one of the biggest differences is that HTTP/2 will make use of one TCP (Transmission Control Protocol) connection per origin and “multiplex” the stream. If you’ve ever taken a look at the issues that Google PageSpeed Insights highlights, you’ll notice that one of the primary things that always comes up is limiting the number of HTTP requests/ This is what multiplexing helps eliminate; HTTP/2 opens up one connection to each server, pushing assets across it at the same time, often making determinations of required resources based on the initial resource. With browsers requiring Transport Layer Security (TLS) to leverage HTTP/2, it’s very likely that Google will make some sort of push in the near future to get websites to adopt it. After all, speed and security have been common threads throughout everything in the past five years.
As of late, more hosting providers have been highlighting the fact that they are making HTTP/2 available, which is probably why there’s been a significant jump in its usage this year. The beauty of HTTP/2 is that most browsers already support it and you don’t have to do much to enable it unless your site is not secure.
Definitely keep HTTP/2 on your radar, as it may be the culmination of what Google has been pushing for.
SEO tools are lagging behind search engines
When I think critically about this, SEO tools have always lagged behind the capabilities of search engines. That’s to be expected, though, because SEO tools are built by smaller teams and the most important things must be prioritized. A lack of technical understanding may lead to you believe the information from the tools you use when they are inaccurate.
When you review some of Google’s own documentation, you’ll find that some of my favorite tools are not in line with Google’s specifications. For instance, Google allows you to specify hreflang, rel-canonical, and x-robots in HTTP headers. There’s a huge lack of consistency in SEO tools’ ability to check for those directives.
It’s possible that you’ve performed an audit of a site and found it difficult to determine why a page has fallen out of the index. It very well could be because a developer was following Google’s documentation and specifying a directive in an HTTP header, but your SEO tool did not surface it. In fact, it’s generally better to set these at the HTTP header level than to add bytes to your download time by filling up every page’s <head> with them.
To my knowledge, none of the other crawling tools are doing this yet. I do recognize the fact that it would be considerably more expensive for all SEO tools to make this shift because cloud server usage is time-based and it takes significantly more time to render a page in a browser than to just download the main HTML file. How much time?
A ton more time, actually. I just wrote a simple script that just loads the HTML using both cURL and HorsemanJS. cURL took an average of 5.25 milliseconds to download the HTML of the Yahoo homepage. HorsemanJS, on the other hand, took an average of 25,839.25 milliseconds or roughly 26 seconds to render the page. It’s the difference between crawling 686,000 URLs an hour and 138.
Ideally, SEO tools would extract the technologies in use on the site or perform some sort of DIFF operation on a few pages and then offer the option to crawl headless if it’s deemed worthwhile.
Luckily, until SEO tools catch up, Chrome DevTools does handle a lot of these things. For instance, the HTTP Request and Response headers section will show you x-robots, hreflang, and rel-canonical HTTP headers.
You can also use DevTools’ GeoLocation Emulator to get view the web as though you are in a different location. For those of you who have fond memories of the nearEquals query parameter, this is another way you can get a sense of where you rank in precise locations.
Chrome DevTools also allows you to plug in your Android device and control it from your browser. There’s any number of use cases for this from an SEO perspective, but Simo Ahava wrote a great instructional post on how you can use it to debug your mobile analytics setup. You can do the same on iOS devices in Safari if you have a Mac.
What truly are rankings in 2016?
Rankings are a funny thing and, truthfully, have been for some time now. I, myself, was resistant to the idea of averaged rankings when Google rolled them out in Webmaster Tools/Search Console, but average rankings actually make a lot more sense than what we look at in standard ranking tools. Let me explain.
SEO tools pull rankings based on a situation that doesn’t actually exist in the real world. The machines that scrape Google are meant to be clean and otherwise agnostic unless you explicitly specify a location. Effectively, these tools look to understand how rankings would look to users searching for the first time with no context or history with Google. Ranking software emulates a user who is logging onto the web for the first time ever and the first thing they think to do is search for “4ft fishing rod.” Then they continually search for a series of other related and/or unrelated queries without ever actually clicking on a result. Granted. some software may do other things to try and emulate that user, but either way they collect data that is not necessarily reflective of what real users see. And finally, with so many people tracking many of the same keywords so frequently, you have to wonder how much these tools inflate search volume.
The bottom line is that we are ignoring true user context, especially in the mobile arena.
Rankings tools that allow you to track mobile rankings usually let you define one context or they will simply specify “mobile phone” as an option. Cindy Krum’s research indicates that SERP features and rankings will be different based on the combination of user agent, phone make and model, browser, and even the content on their phone.
Rankings tools also ignore the user’s reality of choice. We’re in an era where there are simply so many elements that comprise the SERP, that #1 is simply NOT #1. In some cases, #1 is the 8th choice on the page and far below the fold.
With AdWords having a 4th ad slot, organic being pushed far below the fold, and users not being sure of the difference between organic and paid, being #1 in organic doesn’t mean what it used to. So when we look at rankings reports that tell us we’re number one, we’re often deluding ourselves as to what outcome that will drive. When we report that to clients, we’re not focusing on actionability or user context. Rather, we are focusing entirely on vanity.
Of course, rankings are not a business goal; they’re a measure of potential or opportunity. No matter how much we talk about how they shouldn’t be the main KPI, rankings are still something that SEOs point at to show they’re moving the needle. Therefore we should consider thinking of organic rankings as being relative to the SERP features that surround them.
In other words, I’d like to see rankings include both the standard organic 1–10 ranking as well as the absolute position with regard to Paid, local packs, and featured snippets. Anything else is ignoring the impact of the choices that are overwhelmingly available to the user.
Recently, we’ve seen some upgrades to this effect with Moz making a big change to how they are surfacing features of rankings and I know a number of other tools have highlighted the organic features as well. Who will be the first to highlight the Integrated Search context? After all, many users don’t know the difference.
What is cloaking in 2016?
Cloaking is officially defined as showing search engines something different from the user. What does that mean when Google allows adaptive and responsive sites and crawls both headless and text-based? What does that mean when Googlebot respects 304 response codes?
Under adaptive and responsive models, it’s often the case that more or less content is shown for different contexts. This is rare for responsive, as it’s meant to reposition and size content by definition, but some implementations may instead reduce content components to make the viewing context work.
In the case when a site responds to screen resolution by changing what content is shown and more content is shown beyond the resolution that Googlebot renders, how do they distinguish that from cloaking?
Similarly, the 304 response code is way to indicate to the client that the content has not been modified since the last time it visited; therefore, there’s no reason to download it again.
Googlebot adheres to this response code to keep from being a bandwidth hog. So what’s to stop a webmaster from getting one version of the page indexed, changing it, and then returning a 304?
I don’t know that there are definitive answers to those questions at this point. However, based on what I’m seeing in the wild, these have proven to be opportunities for technical SEOs that are still dedicated to testing and learning.
Accessibility of content as a fundamental component that SEOs must examine has not changed. What has changed is the type of analytical effort that needs to go into it. It’s been established that Google’s crawling capabilities have improved dramatically and people like Eric Wu have done a great job of surfacing the granular detail of those capabilities with experiments like JSCrawlability.com
Similarly, I wanted to try an experiment to see how Googlebot behaves once it loads a page. Using LuckyOrange, I attempted to capture a video of Googlebot once it gets to the page:
I installed the LuckyOrange script on a page that hadn’t been indexed yet and set it up so that it only only fires if the user agent contains “googlebot.” Once I was set up, I then invoked Fetch and Render from Search Console. I’d hoped to see mouse scrolling or an attempt at a form fill. Instead, the cursor never moved and Googlebot was only on the page for a few seconds. Later on, I saw another hit from Googlebot to that URL and then the page appeared in the index shortly thereafter. There was no record of the second visit in LuckyOrange.
While I’d like to do more extensive testing on a bigger site to validate this finding, my hypothesis from this anecdotal experience is that Googlebot will come to the site and make a determination of whether a page/site needs to be crawled using the headless crawler. Based on that, they’ll come back to the site using the right crawler for the job.
I encourage you to give it a try as well. You don’t have to use LuckyOrange — you could use HotJar or anything else like it — but here’s my code for LuckyOrange:
The moral of the story, however, is that what Google sees, how often they see it, and so on are still primary questions that we need to answer as SEOs. While it’s not sexy, log file analysis is an absolutely necessary exercise, especially for large-site SEO projects — perhaps now more than ever, due to the complexities of sites. I’d encourage you to listen to everything Marshall Simmonds says in general, but especially on this subject.
To that end, Google’s Crawl Stats in Search Console are utterly useless. These charts tell me what, exactly? Great, thanks Google, you crawled a bunch of pages at some point in February. Cool!
There are any number of log file analysis tools out there, from Kibana in the ELK stack to other tools such as Logz.io. However, the Screaming Frog team has made leaps and bounds in this arena with the recent release of their Log File Analyzer.
Of note with this tool is how easily it handles millions of records, which I hope is an indication of things to come with their Spider tool as well. Irrespective of who makes the tool, the insights that it helps you unlock are incredibly valuable in terms of what’s actually happening.
We had a client last year that was adamant that their losses in organic were not the result of the Penguin update. They believed that it might be due to turning off other traditional and digital campaigns that may have contributed to search volume, or perhaps seasonality or some other factor. Pulling the log files, I was able to layer all of the data from when all of their campaigns were running and show that it was none of those things; rather, Googlebot activity dropped tremendously right after the Penguin update and at the same time as their organic search traffic. The log files made it definitively obvious.
It follows conventionally held SEO wisdom that Googlebot crawls based on the pages that have the highest quality and/or quantity of links pointing to them. In layering the the number of social shares, links, and Googlebot visits for our latest clients, we’re finding that there’s more correlation between social shares and crawl activity than links. In the data below, the section of the site with the most links actually gets crawled the least!
These are important insights that you may just be guessing at without taking the time to dig into your log files.
How log files help you understand AngularJS
Like any other web page or application, every request results in a record in the logs. But depending on how the server is setup, there are a ton of lessons that can come out of it with regard to AngularJS setups, especially if you’re pre-rendering using one of the snapshot technologies.
For one of our clients, we found that oftentimes when the snapshot system needed to refresh its cache, it took too long and timed out. Googlebot understands these as 5XX errors.
This behavior leads to those pages falling out of the index, and over time we saw pages jump back and forth between ranking very highly and disappearing altogether, or another page on the site taking its place.
Additionally, we found that there were many instances wherein Googlebot was being misidentified as a human user. In turn, Googlebot was served the AngularJS live page rather than the HTML snapshot. However, despite the fact that Googlebot was not seeing the HTML snapshots for these pages, these pages were still making it into the index and ranking just fine. So we ended up working with the client on a test to remove the snapshot system on sections of the site, and organic search traffic actually improved.
That’s not to say that HTML snapshot systems are not worth using. The Googlebot behavior for pre-rendered pages is that they tend to be crawled more quickly and more frequently. My best guess is that this is due to the crawl being less computationally expensive for them to execute. All in all, I’d say using HTML snapshots is still the best practice, but definitely not the only way for Google see these types of sites.
According to Google, you shouldn’t serve snapshots just for them, but for the speed enhancements that the user gets as well.
In general, websites shouldn’t pre-render pages only for Google — we expect that you might pre-render pages for performance benefits for users and that you would follow progressive enhancement guidelines. If you pre-render pages, make sure that the content served to Googlebot matches the user’s experience, both how it looks and how it interacts. Serving Googlebot different content than a normal user would see is considered cloaking, and would be against our Webmaster Guidelines.
These are highly technical decisions that have a direct influence on organic search visibility. From my experience in interviewing SEOs to join our team at iPullRank over the last year, very few of them understand these concepts or are capable of diagnosing issues with HTML snapshots. These issues are now commonplace and will only continue to grow as these technologies continue to be adopted.
However, if we’re to serve snapshots to the user too, it begs the question: Why would we use the framework in the first place? Naturally, tech stack decisions are ones that are beyond the scope of just SEO, but you might consider a framework that doesn’t require such an appliance, like MeteorJS.
Angular 2 has a whole host of improvements over Angular 1.x, but I’ll let these Googlers tell you about them.
Before all of the crazy frameworks reared their confusing heads, Google has had one line of thought about emerging technologies — and that is “progressive enhancement.” With many new IoT devices on the horizon, we should be building websites to serve content for the lowest common denominator of functionality and save the bells and whistles for the devices that can render them.
If you’re starting from scratch, a good approach is to build your site’s structure and navigation using only HTML. Then, once you have the site’s pages, links, and content in place, you can spice up the appearance and interface with AJAX. Googlebot will be happy looking at the HTML, while users with modern browsers can enjoy your AJAX bonuses.
In other words, make sure your content is accessible to everyone. Shoutout to Fili Weise for reminding me of that.
Scraping is the fundamental flawed core of SEO analysis
Scraping is fundamental to everything that our SEO tools do. cURL is a library for making and handling HTTP requests. Most popular programming languages have bindings for the library and, as such, most SEO tools leverage the library or something similar to download web pages.
Think of cURL as working similar to downloading a single file from an FTP; in terms of web pages, it doesn’t mean that the page can be viewed in its entirety, because you’re not downloading all of the required files.
This is where headless browsing comes into play.
One of the more popular headless browsing libraries is PhantomJS. Many tools outside of the SEO world are written using this library for browser automation. Netflix even has one for scraping and taking screenshots called Sketchy. PhantomJS is built from a rendering engine called QtWebkit, which is to say it’s forked from the same code that Safari (and Chrome before Google forked it into Blink) is based on. While PhantomJS is missing the features of the latest browsers, it has enough features to support most things we need for SEO analysis.
As you can see from the GitHub repository, HTML snapshot software such as Prerender.io is written using this library as well.
PhantomJS has a series of wrapper libraries that make it quite easy to use in a variety of different languages. For those of you interested in using it with NodeJS, check out HorsemanJS.
For those of you that are more familiar with PHP, check out PHP PhantomJS.
A more recent and better qualified addition to the headless browser party is Headless Chromium. As you might have guessed, this is a headless version of the Chrome browser. If I were a betting man, I’d say what we’re looking at here is a some sort of toned-down fork of Googlebot.
To that end, this is probably something that SEO companies should consider when rethinking their own crawling infrastructure in the future, if only for a premium tier of users. If you want to know more about Headless Chrome, check out what Sami Kyostila and Alex Clarke (both Googlers) had to say at BlinkOn 6:
Using in-browser scraping to do what your tools can’t
Scraping this way allows you to skip a lot of the coding that’s required to make sites believe you’re a real user, like authentication and cookie management that has to happen on the server side. Of course, this way of scraping is good for one-offs rather than building software around.
ArtooJS is a bookmarklet made to support in-browser scraping and automating scraping across a series of pages and saving the results to a file as JSON.
A more fully featured solution for this is the Chrome Extension, WebScraper.io. It requires no code and makes the whole process point-and-click.
How to approach content and linking from the technical context
Much of what SEO has been doing for the past few years has devolved into the creation of more content for more links. I don’t know that adding anything to the discussion around how to scale content or build more links is of value at this point, but I suspect there are some opportunities for existing links and content that are not top-of-mind for many people.
Google Looks at Entities First
Googlers announced recently that they look at entities first when reviewing a query. An entity is Google’s representation of proper nouns in their system to distinguish persons, places, and things, and inform their understanding of natural language. At this point in the talk, I ask people to put their hands up if they have an entity strategy. I’ve given the talk a dozen times at this point and there have only been two people to raise their hands.
Bill Slawski is the foremost thought leader on this topic, so I’m going to defer to his wisdom and encourage you to read:
I would also encourage you to use a natural language processing tool like AlchemyAPI or MonkeyLearn. Better still, use Google’s own Natural Language Processing API to extract entities. The difference between your standard keyword research and entity strategies is that your entity strategy needs to be built from your existing content. So in identifying entities, you’ll want to do your keyword research first and then run those landing pages through an entity extraction tool to see how they line up. You’ll also want to run your competitor landing pages through those same entity extraction APIs to identify what entities are being targeted for those keywords.
Similarly, Term Frequency/Inverse Document Frequency or TF*IDF is a natural language processing technique that doesn’t get much discussion on this side of the pond. In fact, topic modeling algorithms have been the subject of much-heated debates in the SEO community in the past. The issue of concern is that topic modeling tools have the tendency to push us back towards the Dark Ages of keyword density, rather than considering the idea of creating content that has utility for users. However, in many European countries they swear by TF*IDF (or WDF*IDF — Within Document Frequency/Inverse Document Frequency) as a key technique that drives up organic visibility even without links.
After hanging out in Germany a bit last year, some folks were able to convince me that taking another look at TF*IDF was worth it. So, we did and then we started working it into our content optimization process.
In Searchmetrics’ 2014 study of ranking factors they found that while TF*IDF specifically actually had a negative correlation with visibility, relevant and proof terms have strong positive correlations.
Based on their examination of these factors, Searchmetrics made the call to drop TF*IDF from their analysis altogether in 2015 in favor of the proof terms and relevant terms. Year over year the positive correlation holds for those types of terms, albeit not as high.
In Moz’s own 2015 ranking factors, we find that LDA and TF*IDF related items remain in the highest on-page content factors.
In effect, no matter what model you look at, the general idea is to use related keywords in your copy in order to rank better for your primary target keyword, because it works.
Now, I can’t say we’ve examined the tactic in isolation, but I can say that the pages that we’ve optimized using TF*IDF have seen bigger jumps in rankings than those without it. While we leverage OnPage.org’s TF*IDF tool, we don’t follow it using hard and fast numerical rules. Instead, we allow the related keywords to influence ideation and then use them as they make sense.
At the very least, this order of technical optimization of content needs to revisited. While you’re at it, you should consider the other tactics that Cyrus Shepard called out as well in order to get more mileage out of your content marketing efforts.
302s vs 301s — seriously?
As of late, a reexamination of the 301 vs. 302 redirect has come back up in the SEO echo chamber. I get the sense that Webmaster Trends Analysts in the public eye either like attention or are just bored, so they’ll issue vague tweets just to see what happens.
For those of you who prefer to do work rather than wait for Gary Illyes to tweet, all I’ve got is some data to share.
Once upon a time, we worked with a large media organization. As is par for the course with these types of organizations, their tech team was resistant to implementing much of our recommendations. Yet they had millions of links both internally and externally pointing to URLs that returned 302 response codes.
After many meetings, and a more compelling business case, the one substantial thing that we were able to convince them to do was switch those 302s into 301s. Nearly overnight there was an increase in rankings in the 1–3 rank zone.
Despite seasonality, there was a jump in organic Search traffic as well.
To reiterate, the only substantial change at this point was the 302 to 301 switch. It resulted in a few million more organic search visits month over month. Granted, this was a year ago, but until someone can show me the same happening or no traffic loss when you switch from 301s to 302s, there’s no discussion for us to have.
Internal linking, the technical approach
Under the PageRank model, it’s an axiom that the flow of link equity through the site is an incredibly important component to examine. Unfortunately, so much of the discussion with clients is only on the external links and not about how to better maximize the link equity that a site already has.
There are a number of tools out there that bring this concept to the forefront. For instance, Searchmetrics calculates and visualizes the flow of link equity throughout the site. This gives you a sense of where you can build internal links to make other pages stronger.
Additionally, Paul Shapiro put together a compelling post on how you can calculate a version of internal PageRank for free using the statistical computing software R.
Either of these approaches is incredibly valuable to offering more visibility to content and very much fall in the bucket of what technical SEO can offer.
Structured data is the future of organic search
The popular one-liner is that Google is looking to become the presentation layer of the web. I say, help them do it!
There has been much discussion about how Google is taking our content and attempting to cut our own websites out of the picture. With the traffic boon that the industry has seen from sites making it into the featured snippet, it’s pretty obvious that, in many cases, there’s more value for you in Google taking your content than in them not.
With Vocal Search appliances on mobile devices and the forthcoming Google Home, there’s only one answer that the user receives. That is to say that the Star Trek computer Google is building is not going to read every result — just one. These answers are fueled by rich cards and featured snippets, which are in turn fueled by structured data.
Google has actually done us a huge favor regarding structured data in updating the specifications that allow JSON-LD. Before this, Schema.org was a matter of making very tedious and specific changes to code with little ROI. Now structured data powers a number of components of the SERP and can simply be placed at the <HEAD> of a document quite easily. Now is the time to revisit implementing the extra markup. Builtvisible’s guide to Structured Data remains the gold standard.
Page speed is still Google’s obsession
Google has very aggressive expectations around page speed, especially for the mobile context. They want the above-the-fold content to load within one second. However, 800 milliseconds of that time is pretty much out of your control.
Based on what you can directly affect, as an SEO, you have 200 milliseconds to make content appear on the screen. A lot of what can be done on-page to influence the speed at which things load is optimizing the page for critical rendering path.
To understand this concept, first we have to take a bit of a step back to get a sense of how browsers construct a web page.
- The browser takes the uniform resource locator (URL) that you specify in your address bar and performs a DNS lookup on the domain name.
- Once a socket is open and a connection is negotiated, it then asks the server for the HTML of the page you’ve requested.
- The browser begins to parse the HTML into the Document Object Model until it encounters CSS, then it starts to parse the CSS into the CSS Object Model.
- Once all of this is complete, the browser constructs the Render Tree, which then builds the layout of the page and finally the elements of the page are painted.
Additionally, external resources contribute significantly to page load time. For instance, I always see Chartbeat’s library taking 3 or more seconds just to resolve the DNS. These are all things that need to be reviewed when considering how to make a page load faster.
If you know much about the Accelerated Mobile Pages (AMP) specification, a lot of what I just highlighted might sound very familiar to you.
Essentially, AMP exists because Google believes the general public is bad at coding. So they made a subset of HTML and threw a global CDN behind it to make your pages hit the 1 second mark. Personally, I have a strong aversion to AMP, but as many of us predicted at the top of the year, Google has rolled AMP out beyond just the media vertical and into all types of pages in the SERP. The roadmap indicates that there is a lot more coming, so it’s definitely something we should dig into and look to capitalize on.
Using pre-browsing directives to speed things up
To support site speed improvements, most browsers have pre-browsing resource hints. These hints allow you to indicate to the browser that a file will be needed later in the page, so while the components of the browser are idle, it can download or connect to those resources now. Chrome specifically looks to do these things automatically when it can, and may ignore your specification altogether. However, these directives operate much like the rel-canonical tag — you’re more likely to get value out of them than not.
- Rel-preconnect – This directive allows you to resolve the DNS, initiate the TCP handshake, and negotiate the TLS tunnel between the client and server before you need to. When you don’t do this, these things happen one after another for each resource rather than simultaneously. As the diagram below indicates, in some cases you can shave nearly half a second off just by doing this. Alternatively, if you just want to resolve the DNS in advance, you could use rel-dns-prefetch.
If you see a lot of idle time in your Timeline in Chrome DevTools, rel-preconnect can help you shave some of that off.
You can specify rel-preconnect with
<link rel=”preconnect” href=”https://domain.com”>
or rel-dns-prefetch with
<link rel=”dns-prefetch” href=”domain.com”>
- Rel-prefetch – This directive allows you to download a resource for a page that will be needed in the future. For instance, if you want to pull the stylesheet of the next page or download the HTML for the next page, you can do so by specifying it as
<link rel=”prefetch” href=”nextpage.html”>
- Rel-prerender – Not to be confused with the aforementioned Prerender.io, rel-prerender is a directive that allows you to load an entire page and all of its resources in an invisible tab. Once the user clicks a link to go to that URL, the page appears instantly. If the user instead clicks on a link that you did not specify as the rel-prerender, the prerendered page is deleted from memory. You specify the rel-prerender as follows:
<link rel=”prerender” href=”nextpage.html”>
I’ve talked about rel-prerender in the past in my post about how I improved our site’s speed 68.35% with one line of code.
There are a number of caveats that come with rel-prerender, but the most important one is that you can only specify one page at a time and only one rel-prerender can be specified across all Chrome threads. In my post I talk about how to leverage the Google Analytics API to make the best guess at the URL the user is likely going to visit next.
Finally, keep in mind that rel-prerender does not work with Firefox, iOS Safari, Opera Mini, or Android’s browser. Not sure why they didn’t get invited to the pre-party, but I wouldn’t recommend using it on a mobile device anyway.
- Rel-preload and rel-subresource – Following the same pattern as above, rel-preload and rel-subresource allow you to load things within the same page before they are needed. Rel-subresource is Chrome-specific, while rel-preload works for Chrome, Android, and Opera.
Finally, keep in mind that Chrome is sophisticated enough to make attempts at all of these things. Your resource hints help them develop the 100% confidence level to act on them. Chrome is making a series of predictions based on everything you type into the address bar and it keeps track of whether or not it’s making the right predictions to determine what to preconnect and prerender for you. Check out chrome://predictors to see what Chrome has been predicting based on your behavior.
Where does SEO go from here?
Being a strong SEO requires a series of skills that’s difficult for a single person to be great at. For instance, an SEO with strong technical skills may find it difficult to perform effective outreach or vice-versa. Naturally, SEO is already stratified between on- and off-page in that way. However, the technical skill requirement has continued to grow dramatically in the past few years.
There are a number of skills that have always given technical SEOs an unfair advantage, such as web and software development skills or even statistical modeling skills. Perhaps it’s time to officially further stratify technical SEO from traditional content-driven on-page optimizations, since much of the skillset required is more that of a web developer and network administrator than that of what is typically thought of as SEO (at least at this stage in the game). As an industry, we should consider a role of an SEO Engineer, as some organizations already have.
At the very least, the SEO Engineer will need to have a grasp of all of the following to truly capitalize on these technical opportunities:
- Document Object Model – An understanding of the building blocks of web browsers is fundamental to the understanding how how we front-end developers manipulate the web as they build it.
- Critical Rendering Path – An understanding of how a browser constructs a page and what goes into the rendering of the page will help with the speed enhancements that Google is more aggressively requiring.
- Structured Data and Markup – An understanding of how metadata can be specified to influence how Google understands the information being presented.
- Page Speed – An understanding of the rest of the coding and networking components that impact page load times is the natural next step to getting page speed up. Of course, this is a much bigger deal than SEO, as it impacts the general user experience.
- Log File Analysis – An understanding of how search engines traverse websites and what they deem as important and accessible is a requirement, especially with the advent of new front-end technologies.
- Chrome DevTools – An understanding of one of the most the powerful tools in the SEO toolkit, the Chrome web browser itself. Chrome DevTools’ features coupled with a few third-party plugins close the gaps for many things that SEO tools cannot currently analyze. The SEO Engineer needs to be able to build something quick to get the answers to questions that were previously unasked by our industry.
- Acclerated Mobile Pages & Facebook Instant Pages – If the AMP Roadmap is any indication, Facebook Instant Pages is a similar specification and I suspect it will be difficult for them to continue to exist exclusively.
- HTTP/2 – An understanding of how this protocol will dramatically change the speed of the web and the SEO implications of migrating from HTTP/1.1.
Let’s Make SEO Great Again
One of the things that always made SEO interesting and its thought leaders so compelling was that we tested, learned, and shared that knowledge so heavily. It seems that that culture of testing and learning was drowned in the content deluge. Perhaps many of those types of folks disappeared as the tactics they knew and loved were swallowed by Google’s zoo animals. Perhaps our continually eroding data makes it more and more difficult to draw strong conclusions.
Whatever the case, right now, there are far fewer people publicly testing and discovering opportunities. We need to demand more from our industry, our tools, our clients, our agencies, and ourselves.
Let’s stop chasing the content train and get back to making experiences that perform.
Go to Source