Mastering_SEO_in_a_Generative_AI_World__Turning_Google_s_Secrets_into_ROI.pdf

Transcript

1. 1 1
2. 2 2 Download this deck: https://speakerdeck.com/ipullrank
3. 3 Salutations! I’m Mike King (@iPullRank)
5. 5 There is a Generational Shifting In Search Behavior
6. 6 6 “Search It Up” is the new “Google it”
7. 7 7
8. 8 8
9. 9 There’s A Lot of Discussion of ChatGPT Replacing Google
10. 10 10 40% of People Leaving ChatGPT Go to Google My assumption is that many of these people are fact-checking. That is a bad behavior to establish for nearly half of your users.
11. 11 11 Yes, TikTok is a (Nascent) Search Engine 41% of Tiktok users perform searches, but the search volume around a series of broad and meaningful queries is not there to make it more than a small supplement to Google Search.
12. 12 12 20% of People Going to Tiktok Come from Google 22.5% of People Leaving Tiktok Go to Google
13. 13 13 The Last Time the Major Search Engine Died it Looked Like This
14. 14 14 Google Still Dwarfs Everything, but More People are Using More Channels
15. 15 15 It means there is fragmentation in how information needs are being met.
16. 16 16 Users Have Very High Expectations of a Search Box
17. 17 17 …but search outside of major search engines is still really bad
18. 18 18 Google is still the main event, but we are going back into a world where we need to optimize for multiple search engines across a series of channels.
19. 19 The Mechanics of Modern Search
20. 20 20 Real quick. Let’s talk about how search engines work.
21. 21 21 Search Engines Work based on the Vector Space Model Documents and queries are plotted in multidimensional vector space. The closer a document vector is to a query vector, the more relevant it is.
22. 22 22 TF-IDF Vectors The vectors in the vector space model were built from TF-IDF. These were simplistic based on the Bag-of-Words model and they did not do much to encapsulate meaning.
23. 23 23 Relevance is a Function of Cosine Similarity When we talk about relevance, it’s the question of similar is determined by how similar the vectors are between documents and queries. This is a quantitative measure, not the qualitative idea of how we typically think of relevance.
24. 24 24 https://ipullrank.com/cosine-similarity-knn-in-google-sheets
25. 25 25 In SEO we still think Google is Here
26. 26 26 The lexical model counts the presence and distribution of words. Whereas the semantic model captures meaning. This was the huge quantum leap behind Google’s Hummingbird update and most SEO software has been behind for over a decade. Google Shifted from Lexical to Semantic a Decade Ago
27. 27 Word2Vec Gave Us Embeddings Word2Vec was an innovation led by Tomas Mikolov and Jeff Dean that yielded an improvement in natural language understanding by using neural networks to compute word vectors. These were better at capturing meaning. Many follow-on innovations like Sentence2Vec and Doc2Vec would follow.
28. 28 28 Words are Converted to Multi-dimensional Coordinates in Vector Space
29. 29 29 This Allows for Mathematical Operations Comparisons of content and keywords become linear algebraic operations.
30. 30 30 Word2Vec Gave Us Hummingbird
31. 31 31 Google Has Been More Like This Since Hummingbird
32. 32 32 We Went from Sparse Representations to Dense Representations
33. 33 33 This is a huge problem because most SEO software still operates on the lexical model.
34. 34 34 What’s interesting is that Generative AI comes from Search technology
35. 35 35 8 Google Employees Are Responsible for Generative AI https://www.wired.com/story/eight-google-employees-invented-modern-ai-transformers-paper/
36. 36 36 The Transformer The transformer is a deep learning model used in natural language processing (NLP) that relies on self-attention mechanisms to process sequences of data simultaneously, improving efficiency and understanding in tasks like translation and text generation. Its architecture enables it to capture complex relationships within the text, making it a foundational model for many state-of-the-art NLP applications.
37. 37 37 The Transformer Gave us BERT
38. 38 38 Word2Vec Captured Relationship, but Not Context – BERT Captures Context
39. 39 39 BERT Yields Embeddings with Higher Dimensionality and Information Capture
40. 40 40 Since the Introduction of BERT, Google Has Looked More Like This
41. 41 41 Under the SGE/AIO Model, Google is Structured Liked This
42. 42 42 Embeddings are one of the most fascinating things we can leverage as SEOs to catch up to what Google is doing.
43. 43 MixedBread’s Open Source Embeddings are Highly Performant Last week @dejanseo shard his research on how MixedBread’s embedding models perform better than anything else for his SEO use cases. He also talked about lowering the dimensionality and converting them to binary representations to save space.
44. 44 Site Embeddings Are Used to Measure How On Topic a Page is Google is specifically vectorizing pages and sites and comparing the page embeddings to the site embeddings to see how off-topic the page is. Learn more about embeddings: https://ipullrank.com/content-relevance
45. 45 45 You Can Compute This Too I’m using the embeddings with cosine similarity and clustering to examine two ways that how pages relate or don’t relate to the site average of the embeddings. Notice how my recent posts on AI related topics for SEO all have high PageSiteSimilarity whereas my post about MozCon from 2011 does not.
46. 46 Check out the Colab This uses Requests, Trafilatura, Transformers, PyTorch, Scikit Learn, and Sentence Transfomers to compute SiteScore and a dataframe of cosine similarities and cluster based scores for all URLs crawled. https://colab.research.google.com/drive /19PJiFmv8oyjhB-jwzEK9TPlbfK-xB573 You can remove the outliers to improve your site focus score. Add this to your content audits.
47. 47 47 When I Run it on my whole Sitemap, My Site is Not Very Focused
48. 48 48 Add this data to your content audits to make data-driven decisions of what to cut.
49. 49 Content needs to be more focused We’ve learned definitively that Google uses vector embeddings to determine how far off given a page is from the rest of what you talk about. This indicates that it will be challenging to go far into upper funnel content successfully without a structured expansion or without authors who have demonstrated expertise in that subject area. Encourage your authors to cultivate expertise in what they publish across the web and treat their bylines like the gold standard that it is.
50. 50 Build Topic Clusters Well defined topic clusters can position your website and brand into an authority in your space and strengthen your entities in the eyes of Google. A site that focuses on a series of topics that are relevant to each other are going to benefit in rankings. Here are a few tools that can help you design and build your topic clusters systematically. Thruuu https://thruuu.com/keyword-clustering-tool Keyword Insights https://www.keywordinsights.ai/features/keyword-clustering/
51. 51 Let Screaming Frog Do the Heavy Lifting Generate embeddings while you crawl using Screaming Frog SEO Spider. Take the file to Colab and do the following things: Keyword - Landing Page Relevance Scoring Keyword Mapping Link Building Target Identification Redirect Mapping Internal Link Mapping https://ipullrank.com/vector-embedding s-is-all-you-need You can also work with your language model to combine crawl data with SERP data and do things like information gain calculations.
52. 52 52 Leveraging generative AI is a combination of content strategy, your unique creative angles, and deep understanding of the technical nuances of a channel.
53. 53 53 This is our opportunity to get up from the kids table. This is our opportunity to get up from the kids table
54. 54 So, It Turns Out Google is All About the Clicks
55. 55 55 Google’s Algorithms Inner Workings Have Been Put on Full Display Lately Through a combination of what’s come out of Google’s DOJ antitrust trial and the Google API documentation leak, we have a much clearer picture of how Google actually functions.
56. 56 56 I was the First to Publish on the Google Leak, but it was a Team Effort
57. 57 57 We Now Have a Much Stronger Understanding of the Architecture https://searchengineland.com/how-google-search-ranking-works-445141
58. 58 58 These specifics help us zero in on what really matters.
59. 59 The Primary Takeaway is the Value of User Behavior in Organic Search Google’s Navboost system keeps track of user click behavior and uses that to inform what should rank in various contexts.
60. 60 60 Google Has Denied this Many Times
61. 61 61 Many times…
62. 62 62 But it’s Been Confirmed in Pandu Nayak’s Testimony
63. 63 63 And in Google’s documentation for their Cloud Search services
64. 64 It Looks Like Google is Using Clicks After All…
65. 65 65 I remain adamant that both Google and the SEO community owes @randfish an apology.
66. 66 66
67. 67 67 That’s Also Why Google Was So Mad when @DejanSEO Did This…
68. 68 68 User Click Data is What Makes Google More Powerful Than Any Other Search Engine The court opinion in the DoJ Antitrust trial, Google’s leaked documents, and Google’s own internal documentation all support the fact that click behavior is what makes Google perform the way that it does.
69. 69 69 Re-Ranking
70. 70 70
71. 71 71 Re-Ranking via Learning to Rank
72. 72 72 User Based Modeling Looks at the Attractiveness of Results In Addition to the Clicks
73. 73 73 13 Months of Google Data = 17 Years of Bing Data
74. 74 74 You Need to Keep People on your Website to Win at Organic Search
75. 75 Modern SEO Needs UX Baked-in Google has expectations of performance for every position in the SERP. The user behavior signals collected reinforce what should rank and demote what doesn’t perform just like a social media channel. The best way to scale this is by generating highly-relevant content with a strong user experience.
76. 76 UX That Yields Better NavBoost Performance
77. 77 NavBoost Performance Starts at the SERP Itself Continually testing your metadata is a must. Check out the SearchPilot team’s case studies for ideas: https://www.searchpilot.com/resources /case-studies/tag/meta-data
78. 78 78 Design Content for the Human Condition Design your content so it is easier to consume and it will yield better performance metrics. https://moz.com/blog/10-super-easy-seo-copywriting-tips-for-link-building
79. 79 79 Google is Using Passage Indexing to Try to Drop the User Into the Right Spot
80. 80 80 Use Logical Chunking To Get Users to the Information Faster https://www.nngroup.com/articles/in-page-links-content-navigation/
81. 81 81 Build Your Jump Links
82. 82 82 Scannable Headlines
83. 83 83 Jump Links are a Good Sign
84. 84 84 Build Pages that Are Easy to Parse Create semantically relevant content Build a table of contents Drop anchor links throughout the page to help Google understand where the user is meant to go.
85. 85 Stop Leading With This I came for a recipe. Not your Grandma’s life story!
86. 86 Less is More, More or Less It’s time to cut out the content madness
87. 87 87 You Don’t Need Link Volume, You Need Link Quality Indexing Tier Impacts Link Value A metric called sourceType that shows a loose relationship between the where a page is indexed and how valuable it is. For quick background, Google’s index is stratified into tiers where the most important, regularly updated, and accessed content is stored in flash memory. Less important content is stored on solid state drives, and irregularly updated content is stored on standard hard drives. The higher the tier, the more valuable the link. Pages that are considered “fresh” are also considered high quality. Suffice it to say, you want your links to come from pages that either fresh or are otherwise featured in the top tier. Get links from pages that live in the higher tier by modeling a composite score based on data that is available.
88. 88 88 No one is actually improving content when they do the Skyscraper Technique
89. 89 89 Google Stores Your Content Like the Wayback Machine and Uses the Change History Google’s file system is capable of storing versions of pages over time similar to the Wayback Machine. My understanding of this is that Google keeps what it has indexed forever. This is one of the reasons you can’t simply redirect a page to an irrelevant target and expect the link equity to flow. The docs reinforce this idea implying that they keep all the changes they’ve ever seen for the page. You’re not going to get away with things by simply changing your pages once.
90. 90 Indexing is Also Harder It’s not being talked about as much, but indexing has gotten a lot harder since the Helpful Content update. You’ll see a lot more pages in the “Discovered - currently not indexed” and “Crawled - currently not indexed” than you did previously because the bar is higher for what Google deems worth capturing from the web.
91. 91 91 Google Wants to Crawl Even Less Gary Illyes has indicated that he wants to have Google crawl less. Search quality certainly cannot suffer, so crawlin has to get increasingly intelligent.
92. 92 92 I Believe This is a Function of Information Gain Conceptually, as it relates to search engines, Information Gain is the measure of how much unique information a given document adds to the ranking set of documents. In other words, what are you talking about that your competitors are not?
93. 93 93 That’s not all though…
94. 94 94 There are Gold Standard Documents There is no indication of what this means, but the description mentions “human-labeled documents” versus “automatically labeled annotations.” I wonder if this is a function of quality ratings, but Google says quality ratings don’t impact rankings. So, we may never know. 🤔
95. 95 Measure Your Content Against the Quality Rater Guidelines Elias Dabbas created a python script and tool that uses the Helpful Content Recommendations to show a proof of concept way to analyze your articles. We’d use the Search Quality Rater Guidelines which serve as the Golden Document standard. Code: https://blog.adver.tools/posts/llm-content-evaluation/ Tools: https://adver.tools/llm-content-evaluation/
96. 96 96 In conclusion: “More content” is no longer inherently the most effective approach because there’s no guarantee of traffic from Google. Google’s sophistication won’t allow it.
97. 97 97 The only content you should be making
98. 98 98 I’m Leaving Y’all with Three Actions Today 1. How to Prune Your Content 2. How to Appear in LLMs 3. How to Use LLMs to Generate Valuable Content
99. 99 The Content Pruning Process
100. 10 0 10 0 Pruning and Optimization Work Quite Well Together
101. 10 1 Aleyda Has a Process Aleyda’s workflow is a great place to work through whether your content should be pruned or not. https://www.aleydasolis.com/en/crawli ng-mondays/how-to-prune-your-website- content-in-an-seo-process-crawlingmon days-16th-episode/
102. 10 2 10 2 We like automate to get to a Keep. Revise. Kill. (Review.)
103. 10 3 10 3 Content Decay The web is a rapidly changing organism. Google always wants the most relevant content, with the best user experience, and most authority. Unless you stay on top of these measures, you will see traffic fall off over time. Measuring this content decay is as simple comparing page performance period over period in analytics or GSC. Just knowing content has decayed is not enough to be strategic.
104. 10 4 10 4 It’s not enough to know that the page has lost traffic.
105. 10 5 10 5
106. 10 6 10 6 The Content Potential Rating (CPR).
107. 10 7 10 7 Content Potential Score
108. 10 8 10 8 Interpreting the Content Potential Rating 80 - 100: High Priority for Optimization 60 - 79: Moderate Priority for Optimization 40 - 59: Selective Optimization 20 - 39: Low Priority for Optimization 0 - 19: Minimal Benefit from Optimization If you want quick and dirty, you can prune everything below a 40 that is not driving significant traffic.
109. 10 9 10 9 Combining CPR with pages that lost traffic helps you understand if it’s worth it to optimize.
110. 11 0 11 0 Step 1. Pull the Rankings Data from Semrush Organic Research > Positions > Export
111. 11 1 11 1 Step 2: Pull the Decaying Content from GSC Google Search Console is a great source to spot Content Decay by comparing the last three months year over year. Filter for those pages where the Click Difference is negative (smaller than 0) then export.
112. 11 2 11 2 Step 3: Drop them in the Spreadsheet and Press the Magic Button
113. 11 3 The Output is List of URLs Prioritized by Action Each URL is marked as Keep, Revise, Kill or Review based on the keyword opportunities available and the effort required to capitalize on them. Sorting the URLs marked as “Revise” by Aggregated SV and CPR will give you the best opportunities first.
114. 11 4 11 4 Get your copy of the Content Pruning Workbook : https://ipullrank.com/cpr-sheet
115. 11 5 How to Kill Content Content may be valuable for channels outside of Organic Search. So, killing it is about changing Google’s experience of your website to improve its relevance and reinforce its topical clusters. The best approach is to noindex the pages themselves, nofollow the links pointing to them, and submit an XML sitemap of all the pages that have changed. This will yield the quickest recrawling and reconsideration of the content.
116. 11 6 11 6 How to Revise Content Review content across the topic cluster Use co-occurring keywords and entities in your content Add unique perspectives that can’t be found on other ranking pages Answer common questions Answer the People Also Ask Questions Restructure your content using headings relevant to the above Add relevant Structured markup Expand on previous explanations Add authorship Update the dates Make sure the needs of your audiences are accounted for Add to an XML sitemap of only updated pages
117. 11 7 How to Review Content The sheet marks content that has a low content potential rating and a minimum of 500 in monthly search volume as “Review” because they may be long tail opportunities that are valuable to the business. You should take a look at the content you have for that landing page and determine if you think the effort is worthwhile.
118. 11 8 What is Retrieval Augmented Generation (RAG)?
119. 11 9 11 9 Combining a Search Engine with a Language Model is called “Retrieval Augmented Generation” Neeva (RIP), Bing, and now Google’s Search Generative Experience all use pull documents based on search queries and feed them to a language model to generate a response. This concept was developed by the Facebook AI Research (FAIR) team.
120. 12 0 12 0 Google’s Initial Version of this is called Retrieval-Augmented Language Model Pre-Training (REALM) from 2021 REALM identifies full documents, finds the most relevant passages in each, and returns the single most relevant one for information extraction.
121. 12 1 12 1 DeepMind followed up with Retrieval-Enhanced Transformer (RETRO) DeepMind's RETRO (Retrieval-Enhanced Transformer) is a language model that combines a large text database with a transformer architecture to improve performance and reduce the number of parameters required. RETRO is able to achieve comparable performance to state-of-the-art language models such as GPT-3 and Jurassic-1, while using 25x fewer parameters.
122. 12 2 Google’s Later Innovation Retrofit Attribution using Research and Revision (RARR) RARR does not generate text from scratch. Instead, it retrieves a set of candidate passages from a corpus and then reranks them to select the best passage for the given task.
123. 12 3 12 3 AIO is built from REALM/RETRO/RARR + PaLM 2 and MUM MUM is the Multitask Unified Model that Google announced in 2021 as way to do retrieval augmented generation. PaLM 2 is their latest (released) state of the art large language model. The functionality from REALM, RETRO, and RARR is also rolled into this.
124. 12 4 12 4 Sounds cool, but how does it work?
125. 12 5 12 5 Documents are Broken into Chunks and the Most Relevant Chunks are Fed to the Language Model to Generate a Response
126. 12 6 How to Use LLMs for Content and SEO
127. 12 7 12 7 The Three Laws of Generative AI content 1. Generative AI is not the end-all-be-all solution. It is not the replacement for a content strategy or your content team. 2. Generative AI for content creation should be a force multiplier to be utilized to improve workflow and augment strategy. 3. You should consider generative AI content for awareness efforts, but continue to leverage subject matter experts for lower funnel content.
128. 12 8
129. 12 9 12 9
130. 13 0 13 0 Think back to 7 Minutes Ago - Retrieval Augmented Generation
131. 13 1 13 1 It’s Not Difficult to Build with Llama Index sitemap_url = "[SITEMAP URL]" sitemap = adv.sitemap_to_df(sitemap_url) urls_to_crawl = sitemap['loc'].tolist() ... # Make an index from your documents index = VectorStoreIndex.from_documents(documents) # Setup your index for citations query_engine = CitationQueryEngine.from_args( index, # indicate how many document chunks it should return similarity_top_k=5, # here we can control how granular citation sources are, the default is 512 citation_chunk_size=155, ) response = query_engine.query("YOUR PROMPT HERE")
132. 13 2 132 Everyone can code now.
133. 133 PAGE GENERATIVE AI PRODUCTIVITY USE CASES RAG opens up a series of generative AI use cases that work well for your situation. Briefing & Business Cases Content Analysis First-pass Brand Review First-pass Legal Review Content First Draft Keyword Insertion Structured Data Generation Link Identification & Insertion Generating Voiceovers Generating Images Generating Videos Writing Code
134. 13 4 @BritneyMuller’s Guide to Using Colab Britney talked about how easy it is to use Colab with Python. Now it’s even easier to using LLMs. https://github.com/BritneyMuller/colab- notebooks?tab=readme-ov-file
135. 13 5 Just describe what you want You can tell your language model what you want the code to do and it will handle the rest. If it doesn’t work, just describe what went wrong or paste the error and it will fix it for you. In this example my prompt is: {write python code for colab that takes a csv file of keywords and using bertopic with the chatgpt to compute the natural language topics for each row.}
136. 13 6 13 6 Colab + Gemini
137. 13 7 13 7 Llama Index - RAG - https://www.llamaindex.ai/
138. 13 8 13 8 LangChain - Build Agents - https://www.langchain.com/
139. 13 9 13 9 LangFuse - Prompt Management & Observability tool - https://langfuse.com/
140. 14 0 14 0 Literal AI
141. 14 1 141 You don’t need to code for any of this.
142. 14 2 14 2
143. 14 3 14 3
144. 14 4 14 4
145. 14 5 145
146. 14 6 146
147. 14 7 147
148. 14 8 148
149. 14 9 14 9 Bubble - No Code Apps - https://bubble.io/
150. 15 0 Integrate Promptitude with Zapier or Make
151. 15 1 How to Appear in LLMs
152. 15 2 15 2 Blocking LLMs is a Mistake. Appearing in these places will be recognized as brand awareness opportunities very soon.
153. Embrace Structured Data There are three models gaining popularity: 1. KG-enhanced LLMs - Language Model uses KG during pre-training and inference 2. LLM-augmented KGs - LLMs do reasoning and completion on KG data 3. Synergized LLMs + KGs - Multilayer system using both at the same time https://arxiv.org/pdf/2306.08302.pdf Source: Unifying Large Language Models and Knowledge Graphs: A Roadmap
154. 15 4 15 4 What is Mitigation for AIO? 1. Manage expectations on the impact 2. Understand the keywords under threat 3. Re-prioritize your focus to keywords that are not under threat 4. Optimize the passages for the keywords you want to save
155. 15 5 15 5
156. 15 6 15 6 We Can Also Show You Per Keyword How You Show Up
157. 15 7 15 7 It’s all about the Fraggles. (Fragment + Handle)
158. 15 8 15 8
159. 15 9 15 9 The Fraggles Show What AIO Used for the AI Snapshot
160. 16 0 16 0 Scroll to Text
161. 16 1 16 1 Fraggles Relevance Relevance against the chunks to keyword: Relevance against AI Snapshot:
162. Check out MarketBrew’s Free Tool to Help
166. 16 6 The GEO team shared their ChatGPT prompts The GEO team also shared the ChatGPT prompts that help them improve their visibility. You can augment them and put them to work right away. https://github.com/GEO-optim/GEO/blo b/main/src/geo_functions.py
167. Check out @GarrettSussman’s post on how to optimize for AI Overviews: https://ipullrank.com/optimize-content-for-sge
168. 16 8 Roll the Credits
169. 16 9 16 9 What you should know and do to win Google is still the primary show in town Relevance is a quantitative measure GenAI works on the same math as search engines Focus on making your chunks for relevant to rank in GenAI Search Improve UX to drive more long clicks Focus on content your audience wants, prune what they don’t Use RAG to generate content with AI
170. 17 0
171. Contact me if you want to get better results from your SEO: mike@ipullrank.com Thank You | Q&A Award Winning, #GirlDad Featured by Download the Slides: https://speakerdeck.com/ipullrank Mike King Chief Executive Officer @iPullRank