Rewind to fast forward - WeLoveSEO 2021 Paris

Slide 1

Slide 1 text

1 Rewind to fast forward Play the classics or time for change? Bastian Grimm, Peak Ace AG

Slide 2

Slide 2 text

At least, not really, anymore. Technical SEO doesn’t matter.

Slide 3

Slide 3 text

Fully automated, AI-driven content generation is NOT a thing of the future – it’s already here. Nope, content isn’t, either…

Slide 4

Slide 4 text

pa.ag @peakaceag 4 Back in 2008, I used to explain SEO to C-suites like this: Seriously, who doesn’t love the good old 4:3 slide format?! Even back then, there were three crucial pillars… including links!

Slide 5

Slide 5 text

You didn’t really want me to talk about links, right?

Slide 6

Slide 6 text

I love data as much as the next person – but SEO can do so much more than that. It can help us understand demand, behaviour, and more. Don’t obsess over rankings!

Slide 7

Slide 7 text

pa.ag @peakaceag 7 We’re moving away from “over 200 ranking factors” And as for those “ranking factors studies”… well, ranking signals can't be sorted on a spreadsheet by order of importance – it’s much more complex than that! Source: https://pa.ag/3BbVS6k The MUM algorithm can take images as an input (no keywords!) and provide an answer sorted from web pages around the world, regardless of language. How would a general “ranking factor” like links or keywords in title even work in a scenario like that?

Slide 8

Slide 8 text

pa.ag @peakaceag 8 Let’s look at a few hypotheticals. What if… Technical SEO doesn’t matter anymore? 1 CMSes like WordPress solve major technical issues by themselves Content is no longer a key differentiator? 2 Content can be produced by AI at scale, with almost no human intervention Links don’t move the needle anymore? 3 Declining in relevance and other, more accurate types of ranking signals available

Slide 9

Slide 9 text

Before we go there, let’s take a step back…

Slide 10

Slide 10 text

PAST

Slide 11

Slide 11 text

pa.ag @peakaceag 11 25 years ago, students Brin and Page set up a search engine they called “BackRub”: Source: https://pa.ag/3EAUhJl BackRub in 1996: 75 million indexed URLs Google in 2021: more than 130 trillion URLs

Slide 12

Slide 12 text

pa.ag @peakaceag 12 A bit later, an early version of Google looked like this: This was end of ‘98, and Google! Was! Excited! To! Be! Here!

Slide 13

Slide 13 text

pa.ag @peakaceag 13 And search result pages used to look somewhat different: This was a bit later, around 2006-2007

Slide 14

Slide 14 text

pa.ag @peakaceag 14 Notice anything familiar? Yup, good ol’ left-hand navigation is back – only took Google 15 years or so:

Slide 15

Slide 15 text

pa.ag @peakaceag 15 Ok, in fairness – it’s much smarter than it used to be Google calls this "dynamic organisation“; vertically organised on mobile. It appears in different colours, different positions and is sometimes even sticky: Source: https://pa.ag/3hMvY1C

Slide 16

Slide 16 text

pa.ag @peakaceag 16 Also, continuous search just got introduced in Chrome “Keep searching without needing to hit the back button” – essentially continuous search directly in your Chrome browser, and yes, this can/will also contain competitors: Source: https://pa.ag/3ECb3In […] To make it easier to navigate from one search result to the next in Chrome, we’re experimenting with adding a row beneath the address bar on Chrome for Android that shows the rest of the search results so you can get to the next result without having to go back […]

Slide 17

Slide 17 text

pa.ag @peakaceag 17 As well as continuous scrolling on mobile devices Available in Google Search for most English searches on mobile devices in the US: Source: https://pa.ag/3BLJvhM

Slide 18

Slide 18 text

pa.ag @peakaceag 18 Google is blending Search & Chrome more and more Chrome is adding a “side search” panel which will make it easier to browse previous search results – no need to hit the back button anymore: Source: https://pa.ag/2Yj8N7S

Slide 19

Slide 19 text

pa.ag @peakaceag 19 Google is really becoming creative with more ad space “You can traffic full-page web ads that appear between page views“ – like seriously?! Source: https://pa.ag/3FjJlR4

Slide 20

Slide 20 text

pa.ag @peakaceag 20 Google continues to pull as much data as possible For most queries about this year’s Olympics, there was no need to leave the SERP: Source: Alistair Lattimore via https://pa.ag/3ztZjDM

Slide 21

Slide 21 text

pa.ag @peakaceag 21 Nothing new, right? That’s very true actually; in fact, I used this example in a presentation years ago: Source: Peak Ace presentation from 2018 via https://pa.ag/3hPQfU1 new president usa The searcher instantly found what was expected= happy user!

Slide 22

Slide 22 text

pa.ag @peakaceag 22 Google wants to be the single global source of information Google will change and evolve to meet this goal. If your business doesn't adapt - it's going to lose.

Slide 23

Slide 23 text

pa.ag @peakaceag 23 It’s more than pulling in data: it’s making you “stick” Now, you need one more click to get to what you need (like a phone number, or a route) – essentially, Google is artificially inflating the number of searches, again: Source: https://pa.ag/39kdxwz

Slide 24

Slide 24 text

pa.ag @peakaceag 24 Google dedicates almost half the first page to its own products, which dominate the coveted top of the page: Source: https://pa.ag/2ZgGJTl

Slide 25

Slide 25 text

pa.ag @peakaceag 25 Speak no evil, think no evil! Google makes it obvious that certain words are taboo in both internal and external communication, e.g. don’t use “market share”, or “market” – instead, use “industry”: Source: https://pa.ag/3nQV9Ug & https://pa.ag/3zkWDIN

Slide 26

Slide 26 text

pa.ag @peakaceag 26 Tons of smaller changes, impossible to really keep track Around July ’21, Google started testing indenting search results from the same domain: Source: https://pa.ag/3hPsSda

Slide 27

Slide 27 text

pa.ag @peakaceag 27 Google introduced various types of in-SERP warnings E.g. for fast-changing information and to fight misinformation: Source: https://pa.ag/3tUcVqR

Slide 28

Slide 28 text

pa.ag @peakaceag 28 Back to the big stuff: Much more than visual changes High-authority sites (with health info) started seeing massive increases in June 2020, which were (partially) scaled back during the December 2020 core update: Source: Sistrix Toolbox & Lily Ray via https://pa.ag/39Bkslf

Slide 29

Slide 29 text

pa.ag @peakaceag 29 But speaking of updates… summer of updates, much? Passage ranking (EN only) 10.2. June core update 5.6. Page experience update 15.6. Web spam update (Part #1) 24.6. Web spam update (Part #2) 30.6. July core update 4.7. “About this result” panel update 22.7. Page title update 25.8.

Slide 30

Slide 30 text

pa.ag @peakaceag 30 Can‘t keep up? Sistrix (Google Updates Checker) or Semrush (Sensor) has got you covered, for free! Source: https://pa.ag/3koWG1S & https://pa.ag/3hLQDTi

Slide 31

Slide 31 text

Who'd have thought that Google would actually mention links from time to time ... "Web Spam Updates“ are back…!

Slide 32

Slide 32 text

(Yet) Maybe links aren’t entirely dead?

Slide 33

Slide 33 text

No one really uses fancy “new“ attributes like rel=sponsored, but Google desperately wants the data. My guess?

Slide 34

Slide 34 text

pa.ag @peakaceag 34 I think there’s a reason why this is all happening at once… In May 2021, Google published a major release of “TF-Ranking” that enables full support for natively building LTR models using Keras (a high-level Tensor Flow 2 API): Source: https://pa.ag/3EJYRoG These [Keras] components make building a customised LTR model easier than ever and facilitate rapid exploration of new model structures for production and research. Our most recent release [is] the culmination of 2.5 years of neural LTR research.

Slide 35

Slide 35 text

LTR is a class of techniques applying supervised machine learning (ML) to solve ranking problems. LTR = Learning to Rank

Slide 36

Slide 36 text

pa.ag @peakaceag 36 Interpretable LTR using GAMs (=interpretable rankings) GAMs are compact, intrinsically interpretable models which consider both the ranked items and context features (e.g. query/user profile) Source: https://pa.ag/2ZcvBXs

Slide 37

Slide 37 text

pa.ag @peakaceag 37 So, let’s try to make this a bit more visual: Source: https://pa.ag/3EJYRoG For each input feature (e.g. distance), a sub- model produces a sub- score that can be examined, providing transparency. Context features (e.g. user device) can be used to derive importance weights of sub models.

Slide 38

Slide 38 text

Now, developers (and you) can “understand“ choices, selections and groups of rankings created by those LTR ML models, for much faster improvements This is REALLY huge!

Slide 39

Slide 39 text

pa.ag @peakaceag 39 Clearly, this trend will only continue: Google recently updated the “How Google Search Works” website reported that they made 4,500 “improvements” to search in 2020 alone: Source: https://pa.ag/3BbqN2H

Slide 40

Slide 40 text

Fail to give searchers what they want, and your chances of ranking are slim to none But it‘s not only updates; intent also plays a huge role!

Slide 41

Slide 41 text

pa.ag @peakaceag 41 30-second recap: what’s search intent anyways? Search intent is the why behind a search query: why did the person make this search? Are they looking for information, to make a purchase, or for a specific website? Informational Navigational Commercial Transactional ▪ “Jason Statham movies” ▪ “Berlin Paris distance” ▪ “what are carbs” ▪ “peak ace address” ▪ “gmail” ▪ ”instagram login” ▪ “Dubai winter temperature” ▪ “haircut near me” ▪ “best webinar software” ▪ “Audi rsq8 price” ▪ “champagne next day delivery” ▪ “BER CDG flights”

Slide 42

Slide 42 text

pa.ag @peakaceag 42 Google is obsessed with “Intent” The current version of their Search Quality Evaluator Guidelines mentions “Intent” over 420 times – the “Needs Met” section spans over almost 30 pages: Source: https://pa.ag/2W1qRCS

Slide 43

Slide 43 text

pa.ag @peakaceag 43 Hate to say but it… again, ML plays a role here as well: Back in 2007, Microsoft published a patent that suggests that 87% of ambiguous queries can be identified and understood with supervised machine learning: Source: https://pa.ag/2XHdZTt We propose a machine learning model based on search results to identify ambiguous queries. The best classifier achieves accuracy as high as 87%. By applying the classifier, we estimate that about 16% queries are ambiguous in the sampled logs.

Slide 44

Slide 44 text

pa.ag @peakaceag 44 Thanks to recent advances in ML, Google has made huge leaps ahead with getting search intent right - and they're only going to get better at it. I expect them to reduce the number of results once they’re ~100% certain.

Slide 45

Slide 45 text

pa.ag @peakaceag 45 Can’t get your head round it? Automating at scale? Kevin Indig has got you covered! Go check out his two articles on the topic: Source: https://pa.ag/3u41oFj

Slide 46

Slide 46 text

pa.ag @peakaceag 46 You‘re late to the party if you haven‘t figured this out yet: It’s of utmost importance right now to get intent mapping right; intent means relevance and therefore better rankings. Get this wrong, and you have no chance of ranking long term.

Slide 47

Slide 47 text

So yep, tons of things going on – let’s fast forward to today:

Slide 48

Slide 48 text

PRESENT

Slide 49

Slide 49 text

Wait a second - this isn't new! Isn't this just what we used to call “domain authority”? Expertise, Authoritativeness and Trustworthiness (E-A-T)

Slide 50

Slide 50 text

pa.ag @peakaceag 50 Back in 2019, Google gave us their official “confirmation” E-A-T is an important part of their algorithms. If you have been negatively affected by a core update, you need to get to know the QRG as well as E-A-T specifically: Source: https://pa.ag/3u1kBrm The concept of E-A-T is discussed in detail in Google’s Quality Raters’ Guidelines (QRG). Demonstrating good E-A-T both on and off your website can (potentially) help improve rankings.

Slide 51

Slide 51 text

pa.ag @peakaceag 51 Let‘s try to summarise what E-A-T actually is Surfacing results with good E-A-T is a goal of Google, and what the algorithms are supposed to do – but E-A-T itself is not an explanation of how the algorithms currently work. Because there are soooo…(!) many misconceptions out there:

Slide 52

Slide 52 text

Google's algorithms don't give an E-A-T score. Quality raters analyse E-A-T in their checks, but don't give a score and it doesn't directly affect your rankings. There is no E-A-T score

Slide 53

Slide 53 text

pa.ag @peakaceag 53 E-A-T is not an algorithm (on its own) [Google has] a collection of millions of tiny algorithms that work in unison to spit out a ranking score. Many of those […] look for signals in pages or content. When you put them together […], they can be conceptualised as E-A-T. Gary Illyes at PubCon in October 2019:

Slide 54

Slide 54 text

pa.ag @peakaceag 54 E-A-T is not a “real ranking factor” Source: https://pa.ag/3zAqvAO See what I did there?

Slide 55

Slide 55 text

pa.ag @peakaceag 55 E-A-T approximates what the algorithms should do Source: https://pa.ag/3CCXW7I […] what would Google do algorithmically to impact those [E-A-T] things? When it comes to, say, health – would Google employ BioSentVec embeddings to determine which sites are more relevant to highly valuable medical texts? […] I tend to think they’re experimenting here [… and] this is a far better conversation than say, should I change my byline to include ‘Dr.’ in hopes that it conveys more expertise?” This quote from AJ Kohn contains a fantastic, hands-on description:

Slide 56

Slide 56 text

pa.ag @peakaceag 56 BioSentVec? Sentence embeddings? YES, more ML! Sentence embeddings represent entire sentences and their semantic information as vectors. This helps the machine to understand context, intention, and other nuances: Source: https://pa.ag/3u3EttK BioSentVec is a set of biomedical embeddings pre-trained on 30M+ articles; specifically for the health vertical, this would allow search engines to more accurately judge content for accuracy and trustworthiness.

Slide 57

Slide 57 text

pa.ag @peakaceag 57 Still confused about what EAT is & how to improve it? Check out these articles from Marie Haynes (MHC) and Fajr Muhammad (iPullRank): Source: https://pa.ag/39uPgUo & https://pa.ag/2W3gb6T

Slide 58

Slide 58 text

Introduced in early 2021, it gives more information about the sites that appear in Google Search “About this result“ panel

Slide 59

Slide 59 text

pa.ag @peakaceag 59 Continued investment in information literacy features “About this Result” has been viewed 400M+ times since its launch, and a new version with even more details is on the way: Source: https://pa.ag/2YlOdUQ The panel will now include information about the source itself (Wikipedia description), and what the site says about itself, as well as news, reviews and other contextual information that can help the user to better evaluate unfamiliar or new sources.

Slide 60

Slide 60 text

… or Passage Indexing? Or both? Passage Ranking

Slide 61

Slide 61 text

pa.ag @peakaceag 61 Passage Indexing >> Passage Ranking Google’s approach to better understand and rank “less well-structured” long-form content: Source: https://pa.ag/2W0Sqw3 Focus on very long pages and/or pages that target multiple topics Improved understanding of certain sections (“passages”) of a page better which previously might have seemed irrelevant Passages won’t be indexed alone; the passage identified will be given additional weight in ranking, thus “passage ranking”.

Slide 62

Slide 62 text

pa.ag @peakaceag 62 Passage Ranking went live on February 10, 2021 But: “only in the US in English” (read: for English-language search queries) Source: https://pa.ag/2W0Sqw3 Sooo… maybe they’ll tell us, maybe not?!

Slide 63

Slide 63 text

pa.ag @peakaceag 63 Passage Ranking: what you really need to know Google’s Martin Splitt says: “It’s just us getting better at more granularly understanding the content of a page, and being able to score different parts of a page independently” Source: https://pa.ag/3kx6o2h

Slide 64

Slide 64 text

“Continue to focus on great content” – that’s what Google tells us. So why even bother? There’s nothing special creators need to do!

Slide 65

Slide 65 text

pa.ag @peakaceag 65 Could there be another “ML connection” going on here? Check out Dawn Anderson’s fantastic coverage of BERT, its capabilities as a re-ranker including current limitations and why BERT is (most likely) used in passage ranking: Source: https://pa.ag/3u0KuaN […] it is highly likely BERT has a strong connection to the change [passage indexing], given the overwhelming use of BERT (and friends) as a passage re-ranker in the research of the past 12 months or so.

Slide 66

Slide 66 text

pa.ag @peakaceag 66 Passage re-ranking using BERT BERT has probably been (completely) repurposed to add contextual meaning to a training set of passages in two stages: Source: https://pa.ag/3oCy0Wh Super super(!) simply put, a “re-ranker” takes classic rankings signals and then re- ranks the initial results based on additional or more refined input and/or data. Essentially, a re-ranker is a layer on top.

Slide 67

Slide 67 text

So, where does all this lead?

Slide 68

Slide 68 text

FUTURE

Slide 69

Slide 69 text

pa.ag @peakaceag 69 Here’s what I think is going to happen… Google is moving towards becoming a fully automated recommender system, operating in an (almost entirely) query-less world, which anticipates your every question based on your individual search journey/context.

Slide 70

Slide 70 text

pa.ag @peakaceag 70 What’s a recommender system? Recommenders produce items based on user history/similarities. Results are computed by predicting their rating or by recommending similar items: Source: https://pa.ag/3CCSIJa

Slide 71

Slide 71 text

Google has been figuring out what people might ask based on search history, user data and other data points for a long time. Just see “people also ask”! Anticipate questions before asking?

Slide 72

Slide 72 text

pa.ag @peakaceag 72 In fact, at Search On 2021 Google confirmed exactly this: Prabhakar Raghavan (SVP , Google) said: Source: https://pa.ag/3amVV3L My team and I spent a great deal of time providing high-quality answers to questions that haven’t even been asked yet.

Slide 73

Slide 73 text

pa.ag @peakaceag 73 A query-less world – but how? Obviously, it’s already possible to search on lots of devices without a keyboard; but AI- driven solutions allow for surfacing info/content without actively searching for it: Source: https://pa.ag/2W1E3ro Google Discover is a content recommendation engine that suggests content across the web based on a user’s search history and behaviour. Google Assistant allows users to engage in two-way conversations and get answers from the system without ever even looking at a “classic” search result. Google Lens lets you search what you see - from your camera or photo. Over 3 billion searches monthly already, and especially popular in learning.

Slide 74

Slide 74 text

Moving beyond standalone, individual search queries which were meant to provide “the best answer” towards understanding context and language in search. Search journeys?

Slide 75

Slide 75 text

pa.ag @peakaceag 75 The Google Multitask Unified Model (MUM) Google’s most recent push into AI, seeking to deliver search results that overcome language and format barriers to deliver an improved search experience: Source: https://pa.ag/3kvuUAQ & https://pa.ag/3CFMMiP ▪ Like BERT, it’s built on a transformer architecture ▪ 1,000x more powerful than BERT ▪ Can acquire deep knowledge of the world ▪ Understand and generate language ▪ Trained across 75 languages ▪ Understand multiple forms of information

Slide 76

Slide 76 text

pa.ag @peakaceag 76 A lot of innovation in NLP comes with larger datasets MUM uses the T5 model which is pre-trained on C4 and achieves state-of-the-art results on many NLP benchmarks: Source: https://pa.ag/3EFfRfT To accurately measure the effect of scaling up […], one needs a dataset that is not only high-quality and diverse, but also massive. […] To satisfy these requirements, we developed the Colossal Clean Crawled Corpus (C4), a cleaned version of Common Crawl that is two orders of magnitude larger than Wikipedia.

Slide 77

Slide 77 text

pa.ag @peakaceag 77 One thing that often gets overlooked… Source: https://pa.ag/3CFMMiP MUM is multimodal, so it understands information across text and images and, in the future, can expand to more modalities like video and audio.

Slide 78

Slide 78 text

pa.ag @peakaceag 78 Google Cloud > Vision AI (for images) Vision API offers access to powerful pre-trained ML models. Detect objects and faces, read printed and handwritten text, etc. Source: https://pa.ag/3u3nOGR

Slide 79

Slide 79 text

pa.ag @peakaceag 79 “Search part of the page with Google Lens“, anyone? Want a test-drive? Go to > chrome://flags > Enable Lens Region Search (restart Chrome) Source: https://pa.ag/3DfKc3o

Slide 80

Slide 80 text

pa.ag @peakaceag 80 Also, Google is getting into brands in a big way They will soon be measuring brand penetration using image recognition: Source: https://pa.ag/3AK3Lz0 Image analysis by e.g. utilizing Google Streetview can tell them a lot about brand saturation and capacity in different geographic areas – Google might already know how much more than what we actually think they do.

Slide 81

Slide 81 text

pa.ag @peakaceag 81 Google Cloud > Video AI Current core features around understanding “things” in a video (e.g. objects, location and actions), various new stuff in beta (celebrity, face and person detection): Source: https://pa.ag/3CAE2KD Google’s Video AI API services have some really powerful features: ▪ Streaming video analysis ▪ Object detection and tracking ▪ Text detection and extraction ▪ Explicit content detection ▪ Automated closed captioning & subtitles ▪ Celebrity recognition ▪ Face detection ▪ Person detection with pose estimation ▪ … and more!

Slide 82

Slide 82 text

pa.ag @peakaceag 82 Google Cloud > Speech-to-Text (for audio, e.g. podcast) Audio input processing at-scale including complex features such as multi-speaker recognition, etc. Source: https://pa.ag/3lLG3wO

Slide 83

Slide 83 text

pa.ag @peakaceag 83 MUM & Lens were the hottest topics at Search On 21 According to Google, MUM technology is going to revolutionise the way we engage with information; if you haven’t watched the video yet – make sure you do: Source: https://pa.ag/3l8DAgK MUM can simultaneously understand information across a wild range of formats and draw implicit connections between concepts, topics, and ideas of the world around us.

Slide 84

Slide 84 text

Google is already very capable of understanding formats way beyond simple text!

Slide 85

Slide 85 text

What does that mean for our SEO work?

Slide 86

Slide 86 text

Maybe “doesn’t matter” is a little strong. Going forward, I see tech SEO as more of an “enabler” Technical SEO

Slide 87

Slide 87 text

Or as some people call it: “edge SEO“ or “SEO on the edge”. Ever heard of it? Serverless SEO

Slide 88

Slide 88 text

pa.ag @peakaceag 88 Why “SEO on the edge” and how does it work? Using a CDN, all requests will pass through “edge servers“. When we ignore DNS, databases etc for a minute, this is what it would look like: First request, ever. peakace.js is not cached on edge server yet Origin server Request: peakace.js Request: peakace.js peakace.js delivered from origin server Response: peakace.js peakace.js gets cached on edge server

Slide 89

Slide 89 text

pa.ag @peakaceag 89 Why “SEO on the edge” and how does it work? The 2nd request of “peakace.js” is sent to the edge server, however this time “the edge” knows about it and can deliver it straight away; the request won’t reach the origin: Second request (independent of user) Origin server Request: peakace.js peakace.js delivered from edge server peakace.js is cached on edge server

Slide 90

Slide 90 text

pa.ag @peakaceag 90 Back in Sept 2017, Cloudflare introduced their “Workers“ Workers use the V8 JavaScript engine built by Google and run globally on Cloudflare's edge servers. A typical Worker script executes in <1ms – that’s fast! Source: https://pa.ag/3otrFMK

Slide 91

Slide 91 text

Using Workers to overcome challenges and limitations with popular CMS and e-commerce platforms.

Slide 92

Slide 92 text

Workers are fairly straightforward and easy to implement, requiring only minimal dev efforts. Easily build a proof-of-concept rollout & business case

Slide 93

Slide 93 text

pa.ag @peakaceag 93 Does this only work with Cloudflare? Similar implementations are also available with some of the most popular CDN providers out there: Compute@Edge Edge Workers Cloudflare Workers Lambda@Edge

Slide 94

Slide 94 text

pa.ag @peakaceag 94 You can easily test new page titles / descriptions An HTMLRewriter allows you to build comprehensive and expressive HTML parsers inside of a Cloudflare Workers application: Element selectors are super powerful yet easy to use:

Slide 95

Slide 95 text

pa.ag @peakaceag 95 With full control over the HTML response, it’s easy to test new content!

Slide 96

Slide 96 text

Signed exchanges (SXG)?

Slide 97

Slide 97 text

pa.ag @peakaceag 97 SXG allow Google Search to prefetch your content Similar to AMP , SXG allows resources like HTML, JS, CSS, images and fonts to be pre- fetched directly from the SERP – allowing for an “instant experience“ post click: Source: https://pa.ag/3AbSlUg

Slide 98

Slide 98 text

pa.ag @peakaceag 98 Again, Cloudflare has got you covered: The technical implementation process is not simple, so I expect this to be huge! Source: https://pa.ag/3uFs3IW

Slide 99

Slide 99 text

pa.ag @peakaceag 99 According to Sistrix’s research, CWV seem to have impact: Source: https://pa.ag/2WHBn2t Page experience in the form of the Core Web Vitals has a measurable influence on the Google rankings. […] for most commercial websites, it is worth it. In addition, fast websites not only help the Google ranking, but also improve UX.

Slide 100

Slide 100 text

pa.ag @peakaceag 100 ▪ A deep-dive into all of the Core Web Vitals metrics ▪ Introduction to Google Lighthouse ▪ Critical rendering path optimisation ▪ Image optimisation strategies (formats, file types, compression, loading strategies) and tools ▪ Font loading strategies ▪ Performance budgeting & monitoring ▪ TTFB, preloading and pre-fetching, CDNS and more. A deep-dive into Google’s Core Web Vitals Use my checklist on SpeakerDeck.com to double check: All slides on SpeakerDeck: https://pa.ag/3874aQL

Slide 101

Slide 101 text

User-Agent client hints?

Slide 102

Slide 102 text

pa.ag @peakaceag 102 The User-Agent string is messy, like, very messy: Over the decades, this string has accrued a variety of details about the client making the request as well as cruft, due to backwards compatibility: Mozilla/5.0 (Linux; Android 6.0.1; Nexus 5X Build/MMB29P) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/93.0.4577.82 Mobile Safari/537.36 (compatible; Googlebot/2.1; +http://www.google.com/bot.html)

Slide 103

Slide 103 text

pa.ag @peakaceag 103 The UA string will be frozen, client hints to take over User-Agent Client Hints are a new expansion to the Client Hints API, that enables developers to access information about a user's browser – or a crawler’s features: Source: https://pa.ag/3AiiUaI

Slide 104

Slide 104 text

pa.ag @peakaceag 104 It‘s never too early to start testing these things: Googlebot (running Chrome >89) already populates those CH-headers:

Slide 105

Slide 105 text

pa.ag @peakaceag 105 There will always be new things in search: Technical SEO will almost exclusively focus on testing for humans and crawlers alike - providing crucial recommendations enabling sites to rank in search.

Slide 106

Slide 106 text

pa.ag @peakaceag 106 Technical SEO testing – Peak Ace runs its very own test lab We are trying to understand how Googlebot handles “things“… Set up new HTML documents/tests with the click of a button Add an unlimited number of server-side headers, such as X-Robots, canonicals, hreflang, redirects, caching, etc. Add elements to the document , for example meta robots, canonical or tags to run JS Add unique content to the page, depending on the language you want to test for (sometimes, content generation has a valid use-case) Add any type of HTML to the <body> / DOM Integrated bot tracking (JS for evergreen Googlebot + non-JS) by default Automatically generate output by using standard tags (e.g. <iframe>) as well as JavaScript (to ensure rendering is in play) And lots more…

Slide 107

Slide 107 text

pa.ag @peakaceag 107 Testing beyond technical SEO stuff – real SEO AB testing SearchPilot and Ryte offer robust solutions to get you going (as in testing!) asap: Go check them out: https://www.searchpilot.com/ & https://en.ryte.com/

Slide 108

Slide 108 text

How to win in the era of “infinite content”? Content

Slide 109

Slide 109 text

pa.ag @peakaceag 109 The danger of heading towards search singularity: AI makes it easy to churn out vast quantities of mediocre content […] there’s a real risk of medium to long-tail targeted search results becoming a battle between human- and AI-generated content - a search singularity.

Slide 110

Slide 110 text

pa.ag @peakaceag 110 GPT3 is only just the beginning… In Sept 2020, The Guardian had GPT-3, OpenAI’s powerful language generator write an essay for them from scratch based on a short instruction and some prompts: Source: https://pa.ag/3moPX83 GPT-3 produced eight different outputs, or essays. Each one was unique, interesting and advanced a different argument. The Guardian could have just run one of the essays in its entirety. […] Overall, it took less time to edit than many human op-eds.

Slide 111

Slide 111 text

pa.ag @peakaceag 111 Check out Jarvis – its quality is already really good! AI trained to generate original, creative content such as headlines, blog posts, sales emails, video transcripts, and more: Source: https://www.jarvis.ai ▪ Relies on GPT-3 API, meaning its best results are in EN ▪ 50+ templates for super-specific briefings (e.g. FB ads, blog posts, headlines, etc.) ▪ Specific modules for Amazon, online shops, or functionality such as a “summarizer”. ▪ German language available ;) ▪ Don’t just take my word for it! And try Copy.ai, Writesonic or Copysmith

Slide 112

Slide 112 text

OK, let’s talk about the elephant in the room: quality. But it won’t be good enough to rank!

Slide 113

Slide 113 text

pa.ag @peakaceag 113 But it won’t be good enough to rank! Or will it? Source: https://pa.ag/3BoVRMo It’s easy to argue that AI-written content is… not good enough to rank; that it simply dumps connected ideas together [and connects them with] passable sounding phrases. [A] simulacrum of good writing, [it] looks good at first blush but falls apart on closer inspection.

Slide 114

Slide 114 text

Especially on mid- and longtail, it’s fairly common: No narrative. Repetitive information. Unoriginal formats. Have you looked at the SERPs lately!?

Slide 115

Slide 115 text

pa.ag @peakaceag 115 Truth is, machine-generated content already ranks well! Granted, this doesn’t always last long term – but still, its totally possible. And has been for years already, long before AI – with good ol’ “spun” texts: Source: https://pa.ag/3Bg4xok

Slide 116

Slide 116 text

(maybe add some… links?) And if your content doesn’t rank?

Slide 117

Slide 117 text

pa.ag @peakaceag 117 In all seriousness though: We’re going to get to a point where language models – not GPT-3, but one of the successors in the near future – will be able to generate perfectly optimised content.

Slide 118

Slide 118 text

Even if it‘s just to generate some headlines, titles and meta data for you; I‘m sure you‘ll be surprised! Give GPT-3/Jarvis a try!

Slide 119

Slide 119 text

pa.ag @peakaceag 119 Quality is a powerful differentiator today, but it’s about to become even more important: Source: https://pa.ag/3BoVRMo ▪ Focus on information gain in every article you create ▪ Diversify beyond search and invest in thought leadership (counter-narrative opinions, personal narratives, network connections, industry analysis & data storytelling) ▪ Share the same information but create a new experience

Slide 120

Slide 120 text

Cool! Now, who’s excited to hear about some links?? Links

Slide 121

Slide 121 text

Nope, still not going there…

Slide 122

Slide 122 text

122 Now what?

Slide 123

Slide 123 text

pa.ag @peakaceag 123 I hope by now we can all agree on this? AI is fundamentally going to change the next generation of search experience.

Slide 124

Slide 124 text

pa.ag @peakaceag 124 In my view, we’re going to see a fundamental shift: Technical SEO, content and links - machines will take care of them all as a basic requirement.

Slide 125

Slide 125 text

Experience & Satisfaction

Slide 126

Slide 126 text

pa.ag @peakaceag 126 If a page's elements and content don't affect Google's understanding of it, user experience becomes the differentiating factor. Experience and satisfaction will be most important to users, and therefore search engines. Let’s fast forward a bit then, shall we? If Google were omniscient and could understand content/context perfectly, how would you rank one page above another if both are equal in quality and relevance?

Slide 127

Slide 127 text

pa.ag @peakaceag 127 The three cornerstones of SEO – 2022 edition Ensure crawl- & renderability, optimise architecture, internal targeting and linking. Provide unique, holistic and qualitative coverage of relevant topics for your readership. Off-page On-page “Get people to talk about us.” External linking, citations, brand mentions & PR Trust Technical Content Experience & Satisfaction

Slide 128

Slide 128 text

pa.ag @peakaceag 128 The war for data is already raging! Google is delaying cookie blocking, Amazon is blocking Google’s FLoC, IOs 14 tracking prevention, etc. Source: https://pa.ag/3rBMynk

Slide 129

Slide 129 text

pa.ag @peakaceag 129 Google is going to double-down on ecom/payment data Until AI works perfectly, Google is going all-in on payment – because ecom data (like shopping baskets) for attribution and measurement (of satisfaction) are gold! Image Source: https://pa.ag/3ovhF5D Experimental features are already part of Chrome - try for yourself: chrome://flags (#ntp-chrome-cart-module)

Slide 130

Slide 130 text

I don’t think so – but here’s some takes on “near” future changes I predict we’ll be seeing: Too far in the future?

Slide 131

Slide 131 text

pa.ag @peakaceag 131 Here’s what I think we’ll be seeing soon: 01 Continued push for entities & structured data With a major focus on solving the challenge of inconsistent data sources and to train ML algos to perfection. 02 Establishing Chrome as the OS for the web Google needs this layer of data and will push hard (e.g. Apple/Safari deal) for continued market domination 03 Increased competition due to MUM While AI will make Google even better at interpreting complex intents, at the same time you’ll need to compete against more websites. 04 Emphasis on task-driven (classic) search To remain relevant in “classic search”, Google needs help answering any user question at any time. Re-finding things will become a major task; the “new” SERPs will reflect that. 05 “Buy now” button in search results With ecom and CMS’s moving headless and APIs everywhere, we should see this within 12 months…!

Slide 132

Slide 132 text

www.pa.ag twitter.com/peakaceag facebook.com/peakaceag Take your career to the next level: jobs.pa.ag Looking for a new challenge? Peak Ace is hiring for a variety of international SEO roles, from trainees to team leads. Get in touch today! Bastian Grimm bg@pa.ag