All You Can EEAT: Navigating SEO in a Generative AI World

Transcript

1. 1 1
2. 2 Salutations! I’m Mike King (@iPullRank)
5. 5 5
6. 6 6 I don’t think you came here for me to tell you to make great content and use real authors.
7. 7 You Can Google EEAT Best Practices (or just look at TheZebra.com) Write high quality content on subjects you actually know things about. Have an author bio and page with links out to other places you write about similar subjects and links to your social media. Make sure those sources highlight your expertise. Who you write for, where you studied, books you’ve written, etc. Get links from and give links to authoritative sources and people on similar subjects. Make sure the user experience on the sites you write on is great too.
8. 8 8 I think you came here because you want to know the nature of the threats ahead.
9. 9 9 After All, Organic Search is Likely Most of Your Traffic
10. 10 A Threat to Google is a Threat to You
11. 11 11 TikTok Supplanted Google
12. 12 12 ChatGPT was a code red for Google
13. 13 13 ChatGPT is now the Star Trek computer Google wants to be.
14. 14 14 Users believe Google Search quality is on a steep decline
15. 15 15 The media is amplifying this idea
16. 16 16 Google missed earnings earlier this year because ad sales were down
17. 17 17 The DOJ is coming for Google and secrets are coming out
18. 18 This is from a post from the Google Cloud team discussing how their Search product works. Although Google has been telling on itself for years
19. 19 19 These threats will impact you as you look to do content marketing and SEO
20. 20 The TikTok Threat will Mean More Visual Content Ranking To compete with the visual content channels, Google is surfacing more visual content in the SERPs and adding more features that allow users to get exactly where they want to go. This will threaten standard Organic positions for web content.
21. 21 21 Short Form Video is About to Get More Competitive The video and image real estate in Google is going to become even more competitive since marketers recognize short form video as high ROI and the primary way to reach Gen Z.
22. 22 22 Publish Your Short Form Video on your site A primary mistake that content marketers make is only publishing their short form videos on a channel like TikTok, Instagram, or YouTube. You should also publish them on your site using tools like Wistia and marking them up so they can appear in the SERPs.
23. 23 23 Ad Sales Being Down Means more Ads What’s up with all this whitespace? What’s up with this featured snippet?
24. 24 24 The real estate will get smaller, so your content must be that much more effective when it shows up in the SERPs.
25. 25 25 The Last Time People Said Search Quality Was Bad we Got Panda and Penguin Panda fundamentally changed Organic Search. You could no longer create “SEO content” and rank. The SEO community then embraced content marketing knowing that it’s the only way forward with creating content that yields utility. Penguin did the same for links. Google’s Helpful Content update could be the new sheriff in town.
26. 26 The Threat of Generative AI
27. 27 27
28. 28 28 47% Are Increasing their Blog Content
29. 29 29 Marketers Have the Highest Adoption of Generative AI
30. 30 30 There’s a Growing List of Generative AI Tools
31. 31 31 If you’re using ChatGPT, you need AIPRM for prompt management.
32. 32 32 If your prompt is just one sentence, don’t be surprised when you get garbage back.
33. 33 33 Every tool on earth is integrating ChatGPT
34. 34 34 Now we have AutoGPT that can do a series of tasks without prompts.
35. 35 35 Doug Kessler Warned us Back in 2013 Marketers are about to ramp up the content marketing deluge. https://blog.hubspot.com/blog/tabid/ 6307/bid/34080/Why-Marketers-Nee d-to-Rise-Above-the-Deluge-of-Crappy- Content.aspx
36. 36 36 Google’s Loosened their Stance on Generated Content
37. 37 37
38. 38 I don’t believe Google can reliably detect LLM content.
39. 39 39 OpenAI Can’t Even Reliably Detect It Sure, there are a variety of tools out there that “detect” generative AI content. However, they are all unreliable in that they can yield both false negatives and false positives. Even the people who built the best generative AI tools can only correctly detect it at 26% accuracy.
40. 40 40 But Google can use it as a signal among other signals.
41. 41 41 There are Reports that Some Sites Using Generative AI Have Been Crushed These are sites that don’t edit the content prior to publishing, so they deserve it.
42. 42 42 The Helpful Content Update is Finally Showing its Teeth Google has been working on getting the Helpful Content classifier right. The early iterations had limited impact, but now sites are getting smacked left and right. We’re also seeing the threshold for crawling and indexing pages has been raised.
43. 43 43 A Lot of “Niche” Sites are Getting Smacked
44. 44 44 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?
45. 45 Google’s Information Gain Patent Google’s patent indicates that they are specifically scoring for documents that feature net new information over other documents on the same topic.
46. 46 46 So Many People Are Just Creating Copycat Content WHAT GENERATIVE AI MEANS FOR GOOGLE SEARCH
47. 47 47
48. 48 48 If you want to survive what’s coming, you’ll need to deliver stronger content than everyone else.
49. 49 49 The only content you should be making
50. 50 50 Confession.
51. 51 51 I think EEAT is silly.
52. 52 52 I call it E-TEA instead. I call it E-TEA instead.
53. 53 53 How When Authorship Markup Tops out at 3%?
54. 54 54 How When Authorship Markup Tops out at 3%? And this markup does not always specify the author!
55. 55 How I Actually Started to Believe in E-TEA (or How our Understanding of Search is out of date)
56. 56 56 At a Base Level, This is What all Search Engines Do Fundamentally, this is the basis of how search engines function. Google has developed many layers on top of this, but this is the core of what they all do.
57. 57 57 Google’s High-Level Pipeline Abstraction
58. 58 58 We know this, but there is a single set of innovations that sped Google past the SEO community.
59. 59 59 Lexical Search vs Semantic Search are the Two Primary Search Models What we as the SEO community do not have a strong enough handle on is that most of what Google’s doing is on the semantic side and that has all improved dramatically over the last 10 years based on machine learning.
60. 60 60 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.
61. 61 61 Words are Converted to Multi-dimensional Coordinates in Vector Space
62. 62 62 This Allows for Mathematical Operations Comparisons of content and keywords become linear algebraic operations.
63. 63 63 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.
64. 64 64 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.
65. 65 Word2Vec Gave Us Embeddings Word2Vec was an innovation led by Tomas Milosevic 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.
66. 66 66 We Went from Sparse Embeddings to Dense Embeddings
67. 67 67 Word2Vec Captured Relationship, but Not Context – BERT Captures Context
68. 68 68 BERT Yields Embeddings with Higher Dimensionality and Information Capture
69. 69 69 Source: https://cloud.google.com/blog/topics/developers-practitioners/find-anything-blazingly-fast-googl es-vector-search-technology
70. 70 70
71. 71 Dense Retrieval You remember “passage ranking?” This is built on the concept of dense retrieval wherein there are more embeddings representing more of the query and the document to uncover deeper meaning.
72. 72 72 Dense Retrieval is Scoring down to the Sentence Level
73. 73 73 Introducing Google’s Version of Dense Retrieval Google introduces the idea of “aspect embeddings” which is series of embeddings that represent the full elements of both the query and the document and give stronger access to deeper information.
74. 74 74 Dense Representations for Entities Google has improved its entity resolution using embeddings giving them stronger access to information in documents.
75. 75 75 Embeddings = Google really understands content relevance now.
76. 76 Website Representation Vectors Just as there are representations of pages as embeddings, there are vectors representing websites and Google has recently made improvements in understanding when content is not relevant to a given site.
77. 77 Author Vectors Similarly, Google has Author Vectors wherein they are able to identify an author and the subject matter that they discuss. This allows them to fingerprint an author and their expertise.
78. 78 78 So, really E-TEA is a function of information associated with vector representations of websites and authors.
79. 79 79 As a content marketer, you need to treat your byline like the asset that it is.
80. 80 80 Also, relevance isn’t qualitative to Google.
81. 81 81 Embeddings keep getting better at capturing meaning while SEO tools still operate on the Lexical Search model
82. 82 82
83. 83 83 I Feel Like My Page is More Relevant to [Enterprise SEO]
84. 84 84 Relevance isn’t Qualitative to Google.
85. 85 85
86. 86 86
87. 87 87 See! My page is more relevant, but it’s not ranking as well.
88. 88 88 https://ipullrank.com/tools/orbitwise
89. 89 The threat of Google’s Search Generative Experience (SGE)
90. 90 At I/O Google Announced a Dramatic Change to Search The experimental “Search Generative Experience” brings generative AI to the SERPs and significantly changes Google’s UX.
91. 91 91 Queries are Longer and the Featured Snippet is Bigger 1. The query is more natural language and no longer Orwellian Newspeak. It can be much longer than the 32 words that is has been historically in order 2. The Featured Snippet has become the “AI snapshot” which takes 3 results and builds a summary. 3. Users can also ask follow up questions in conversational mode. 3 2 1
92. 92 92 Sundar is All In. In Sundar’s recent press run he keeps saying how Google will be doubling down on SGE. So it’s going to be a thing moving forward.
93. 93 The Search Demand Curve will Shift With the change in the level of natural language query that Google can support, we’re going to see a lot less head terms and a lot more long tail term. Going down Going up
94. 94 94 The CTR Model Will Change With the search results being pushed down by the AI snapshot experience, what is considered #1 will change. We should also expect that any organic result will be clicked less and the standard organic will drop dramatically. However, this will likely yield query displacement.
95. 95 Rank Tracking Will Be More Complex As an industry, we’ll need to decide what is considered the #1 result. Based on this screenshot positions 1-3 are now the citations for the AI snapshot and #4 is below it. However, the AI snapshot loads on the client side, so rank tracking tools will need to change their approach.
96. 96 96 Context Windows Will Yield More Personalized Results SGE maintains the context window of the previous search in the journey as the user goes through predefined follow questions. This will need to drive the composition of pages to ensure they remain in the consideration set for subsequent results.
97. 97 97 Ask to Trigger.
98. 98 98 Auto-Trigger.
99. 99 99 We’ve seen this take up to 30 seconds to generate. Although, it’s a lot faster now.
100. 10 0 10 0 HELLO EMPTY REAL ESTATE! (aka more ad clicks)
101. 10 1 10 1
102. 10 2 10 2
103. 10 3 10 3
104. 10 4 10 4 It’s an “experiment” so we don’t know much, but here’s what we can infer.
105. 10 5 10 5
106. 10 6 10 6 Most autoloading AI Snapshots take 3-30 seconds to load.
107. 10 7 10 7 We Know that Featured Snippets Take 35.1% of Clicks
108. 10 8 10 8 Using All that Information We’re Modeling the Threat of SGE
109. 10 9 10 9 Get your threat report: https://ipullrank.com/sge-report
110. 11 0 What is Retrieval Augmented Generation (RAG)?
111. 11 1 11 1 This 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.
112. 11 2 11 2 Google’s Version of this is called Retrieval-Augmented Language Model Pre-Training (REALM) from 2021
113. 11 3 11 3 SGE is built from REALM + 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 state of the art large language model.
114. 11 4 11 4 If You Want More Technical Detail Check Out This Paper https://arxiv.org/pdf/2002.08909.pdf
115. 11 5 11 5 Search Engines Are Now OK with Not Being Right They evaluate Bing Chat, NeevaAI, http://perplexity.ai & YouChat—only 52% of statements are supported by citations and 75% of citations actually support their statements. https://arxiv.org/abs/2304.09848
116. 11 6 11 6 Sounds cool, but how does it work?
117. 11 7 11 7
118. 11 8 11 8 It’s so easy that I built a proof of concept
119. 11 9 11 9
120. 12 0 12 0
121. 12 1 12 1 AvesAPI + Llama Index + ChatGPT = Raggle Rankings data Vector index & operations Clearly you know what this does.
122. 12 2 12 2 It’s pretty simple # 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("Answer the following query in 150 words: " + query)
123. 12 3 12 3 Limitations of my POC It doesn’t do follow up questions It’s not responsive It only does the informational snippet
124. 12 4 12 4 You can play with it at raggle.net
125. 12 5 Optimizing for SGE?
126. 12 6 Dense Retrieval You remember “passage ranking?” This is built on the concept of dense retrieval wherein there are more embeddings representing more of the query and the document to uncover deeper meaning.
127. 12 7 12 7 Dense Retrieval is Scoring down to the Sentence Level
128. 12 8 12 8 It’s all about the chunks. So use Llama Index to determine the your chunks and improve the similarity to the query.
129. 12 9 12 9 I’ve Added A Chunk Explorer so You Can See Which Text was Used
130. 13 0 The Content Opportunity of RAG
131. There’s a Lot of Synergy Between KGs and LLMs 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
132. Organizations are doing RAG with Knowledge Graphs • Anyone can feed their data into an LLM as a fine-tuning measure to improve the output. • People are currently using their knowledge graphs to support this.
133. 13 3 13 3 The code is not much different 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")
134. Fact Verification • Although Google has historically said they do not verification of facts. • LLM + KG integrations make this a possibility and Google needs to combat the wealth of content being produced with LLMs. So, it’s likely they will use this functionality. Source: Fact Checking in Knowledge Graphs by Logical Consistency Source: FactKG: Fact Verification via Reasoning on Knowledge Graphs
135. 13 5 Brands are Using Generative AI as a Force Multiplier • 52% of business leaders are currently using AI content generation tools to assist their content marketing efforts. • 64.7% of business leaders plan to use AI content generation tools to assist their content marketing efforts in 2023. Major brands are using tools like ChatGPT and Midjourney to scale their content marketing efforts. The brands that don’t leverage these tools are quickly falling behind. Source: Siege Media + Clearscope
136. 13 6 13 6 Sadly, Everyone is Using it and No One Has a Strategy
137. 13 7 But… Brands Still Need Content Strategy to Capitalize On It Individuals are using tools like ChatGPT in isolation, but for an organization to capitalize on it there needs to be a generative AI content strategy that encourages governance and consistency of the content created.
138. 138 13 8 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. GENERATIVE AI OPPORTUNITIES & THREATS
139. 13 9 13 9 How We’re Helping Brands Capitalize on Generative AI Leveraging our extensive enterprise Content Strategy experience, we take an 8-step approach to make generative AI tools learn to speak in your brand voice and we build out solutions to bake the functionality into your toolkit. We take a deep dive into how your Content Strategy currently operates to replicate and expand on it through AI. We look for places in your existing processes and tools to integrate AI functionality. We build out the content models, workflows, governance models, and toolkit for generative AI. We develop a library of prompts to be used across your organization for various content use cases. We run the prompts through a series of QA tests to ensure that content is always generated as expected. We improve prompts that do not pass our QA tests. We deliver the prompts and training on how to use the new content systems. We update and optimize prompts as generative AI tools update and emerge. Strategic Planning We tailor our approach to your goals and existing content strategy. Generative AI Delivery We deliver vetted prompts and train your team on generative AI systems. Review Client Goals and Content Strategy Identify AI integration points Prepare Generative AI Content Plan Output QA Build Prompt Library Optimize Outputs Knowledge Transfer Maintenance OUR GENERATIVE AI PROCESS
140. 14 0 14 0 Don’t forget that ChatGPT is very much an unfinished product.
141. Custom Index Functionality Coming Very Soon
142. 14 2 Roll the Credits
143. 14 3 14 3 Mike, that was a lot. What should I be doing? Write with Information Gain in mind Keep an eye on threats in the SERPs Use structured data wherever possible Use tools to understand how relevant Google thinks your content is Build an actual content strategy around generative AI Build a prompt library Build custom indexes for stronger generative AI content creation Treat your byline as the asset that it is By ready for search behavior to change Optimize the chunks
144. 14 4
145. 145 14 5 We’ve Been Using GPT Tech Since 2020 GENERATIVE AI OPPORTUNITIES & THREATS
146. 14 6 14 6 Get your threat report: https://ipullrank.com/sge-report
147. Custom Index Functionality Coming Very Soon
149. Mike King Founder / CEO @iPullRank Thank You | Q&A mike@ipullrank.com Award Winning, #GirlDad Featured by Get Your SGE Threat Report: https://ipullrank.com/sge-report Play with Raggle: https://www.raggle.net