How AI Chatbots Are Changing the Rules of Content Optimization

How AI Chatbots Are Changing the Rules of Content Optimization 🔗 **Source:** [How AI Chatbots Are Changing the Rules of Content Optimization](https://www.youtube.com/watch?v=tR2qt3BTwig) > [!note]+ Document Notes > tags: [AI-SEO, content-optimization, ChatGPT, google-ai-overviews, brand-building] > author: [MISSING] > source: YouTube > reference: https://www.youtube.com/watch?v=tR2qt3BTwig > date created: 2024-12-19 20:30 > > # 🏷️ Title: How AI Chatbots Are Changing the Rules of Content Optimization >  ## 🧰 Action Inbox - [H1] Implement chunking strategy — break content into 1-3 sentence paragraphs with multiple subheadings #content-optimization - [H2] Add FAQ sections to service pages — use "People Also Ask" from Google for questions #ai-seo - [H3] Reverse engineer AI responses — use Gemini's "double check response" feature to find citation gaps #competitive-analysis ## 🧭 Table of Contents - [📝 Summary](#-summary) - [📖 Main Content](#-main-content) - [🗒️ Notes](#️-notes) - [📚 References](#-references) - [🗣️ Paraphrase](#️-paraphrase) - [📰 Newsletter](#-newsletter) - [✅ Action Items](#-action-items) - [🧪 Hypotheses](#-hypotheses) - [📊 Methodology](#-methodology) - [❓ Questions](#-questions) - [🔮 Future Research](#-future-research) - [💡 Implications](#-implications) - [🧠 Learning Styles Assessment](#-learning-styles-assessment) - [📊 Accessibility Assessment](#-accessibility-assessment) - [🌍 DEIB Assessment](#-deib-assessment) - [✅ Logical Validity](#-logical-validity) - [🛡️ DARVO Analysis](#-darvo-analysis) - [🎯 Conclusion](#-conclusion) ### 🔗 References - [ChatGPT](https://openai.com/chatgpt) - [Google Gemini](https://gemini.google.com) - [Easy Table of Contents Plugin](https://wordpress.org/plugins/easy-table-of-contents/) ## 📝 Summary - AI chatbots are fundamentally changing SEO by using a "answer first, then find sources" approach rather than summarizing search results - Traditional SEO remains the foundation, but content must now be optimized for AI chunking with granular sections, bullet points, and FAQ formats - Brand building becomes critical as LLMs evaluate credibility, popularity, and recency (CPR) when selecting sources - Content should follow BLUF (Bottom Line Up Front) structure and avoid AI-generated text to maintain discoverability - The shift moves from optimizing whole pages to optimizing individual content "building blocks" that can stand alone ## 📖 Main Content The presentation explains how AI chatbots like ChatGPT, Gemini, and others fundamentally operate differently from traditional search engines. Instead of crawling and summarizing web results, they first generate answers using their trained knowledge, then search the web for sources to back up what they've already stated. This creates a confirmation bias where they select sources that support their predetermined responses. The speaker emphasizes that traditional SEO remains essential since AI chatbots currently rely on search engines (ChatGPT uses Bing, Gemini uses Google) to find supporting sources. However, content optimization must now account for how LLMs process and chunk information into small, discrete pieces rather than treating web pages as single units. Key technical concepts include vector databases that store information as mixed tidbits rather than complete pages, and the importance of E-A-T-E (Experience, Expertise, Authority, Trustworthiness) where Google added "Experience" specifically to combat bland AI-generated content flooding the web. ## 🗒️ Notes #### ⏱️ [00:00](https://youtu.be/tR2qt3BTwig?t=0) — [H1] Introduction and SEO Death Claims - Key idea: People claim SEO is dead but then ask how to get found in ChatGPT - Evidence: "Everyone was like, 'Oh, Chat GPT has web search now or Google has AI overviews. SEO is dead. It's really dead this time.' But then, how do I get my website found in chat GPT?" #### ⏱️ [03:45](https://youtu.be/tR2qt3BTwig?t=225) — [H2] AI Chatbots Use Traditional Search Engines - Key idea: AI chatbots currently rely on existing search engines to find web sources - Evidence: "ChatGPT uses Bing as its default search engine, as does Copilot. Gemini obviously uses Google. Claude uses Brave, and perplexity uses blended results from Bing, Google, and other sources" #### ⏱️ [06:20](https://youtu.be/tR2qt3BTwig?t=380) — [H1] How LLMs Actually Work - Key idea: LLMs predict text patterns like filling in blanks, not thinking and responding - Evidence: "It's predicting the next word that is correct given the context. The sky is... [audience says blue] without even having to think about that" #### ⏱️ [09:15](https://youtu.be/tR2qt3BTwig?t=555) — [H1] Answer First, Then Find Sources - Key idea: LLMs answer questions first, then search for sources to back up their responses - Evidence: "They answer first with that general knowledge, doing that fill-in-the-blank exercise first and then going and searching the web for sources to back up what it already said" #### ⏱️ [12:30](https://youtu.be/tR2qt3BTwig?t=750) — [H2] CPR Framework for Source Selection - Key idea: AI selects sources based on Credible, Popular, and Recent criteria - Evidence: "We select sources that are credible, popular, and recent... you're going to pick one with 14 million views over one with fewer views" #### ⏱️ [16:45](https://youtu.be/tR2qt3BTwig?t=1005) — [H1] Brand Building Becomes Critical - Key idea: 2026 will be about brand building more than ever for AI discoverability - Evidence: "The LLMs have this broader thinking capacity to look at the brand and assess it from that broader perspective of how many times have you been mentioned in the news" #### ⏱️ [22:10](https://youtu.be/tR2qt3BTwig?t=1330) — [H1] Content Chunking Revolution - Key idea: Web pages are no longer single optimization units but collections of chunks - Evidence: "Gone are the days of thinking of a single web page as a single thing to optimize. It's these individual tidbits and chunks of information that are the opportunities to get found" #### ⏱️ [27:20](https://youtu.be/tR2qt3BTwig?t=1640) — [H2] Reverse Engineering AI Responses - Key idea: Use Gemini's "double check response" to see which sources were cited and find gaps - Evidence: "Go to gemini.google.com, ask it about your topic, then click the three dots and 'double check response' to see where it got which points" #### ⏱️ [32:50](https://youtu.be/tR2qt3BTwig?t=1970) — [H2] New Content Structure Requirements - Key idea: Content must be heavily chunked with 1-3 sentences per paragraph and many subheadings - Evidence: "Gone are the days of writing blog articles like college essays with 5-10 sentences per paragraph and little to no subheadings" #### ⏱️ [36:15](https://youtu.be/tR2qt3BTwig?t=2175) — [H3] FAQ Integration Strategy - Key idea: Add FAQs to individual service pages, not just a general FAQ page - Evidence: "Go to each individual core page about the product or service and add specific FAQs about that page to that page" #### ⏱️ [39:30](https://youtu.be/tR2qt3BTwig?t=2370) — [H3] BLUF Writing Style - Key idea: Write with Bottom Line Up Front - main point first, then supporting details - Evidence: "Make your main key point immediately at the top of the article and for each section... get right to the point, make the point right away, then provide supporting detail" #### ⏱️ [43:20](https://youtu.be/tR2qt3BTwig?t=2600) — [H1] Avoid AI-Generated Content - Key idea: Don't use AI to write content because AI is excellent at detecting AI-generated text - Evidence: "If there's one thing AI is really good at, it's recognizing other AI generated content" #### ⏱️ [47:40](https://youtu.be/tR2qt3BTwig?t=2860) — [H2] Google's E-A-T-E Update - Key idea: Google added "Experience" to E-A-T specifically to combat bland AI content - Evidence: "Google formally revised it to E-A-T-E, adding Experience defined as your personal human firsthand real world experience" #### ⏱️ [52:30](https://youtu.be/tR2qt3BTwig?t=3150) — [H3] Blocking AI Crawlers Discussion - Key idea: Technologies like LLMs.txt being developed but not formally adopted - Evidence: "There's a proposed protocol LLMs.txt for basically the equivalent of robots.txt but for AI crawlers... it's not been formally adopted by any major one" ## 📚 References - [Google Gemini](https://gemini.google.com) - [Easy Table of Contents WordPress Plugin](https://wordpress.org/plugins/easy-table-of-contents/) - [Google Quality Rater Guidelines](https://static.googleusercontent.com/media/guidelines.raterhub.com/en//searchqualityevaluatorguidelines.pdf) - [Cloudflare CDN](https://cloudflare.com) ## 🗣️ Paraphrase - AI chatbots generate answers first using their training data, then search the web for sources to support those predetermined answers, creating a confirmation bias in source selection - Traditional SEO fundamentals remain essential because AI systems currently use existing search engines to find web sources, but content must be restructured for AI processing - The shift from page-level to chunk-level optimization means breaking content into small, standalone pieces with extensive use of bullets, FAQs, and granular subheadings ## 📰 Newsletter **AI is Reshaping How WordPress Content Gets Discovered** The rules of content optimization are changing fast with AI-powered search. While your SEO fundamentals still matter, you now need to optimize for how AI chatbots chunk and process information. Key changes for your WordPress content strategy: - **Structure matters more**: Break content into 1-3 sentence paragraphs with lots of subheadings - **Add FAQs everywhere**: Use Google's "People Also Ask" to add relevant Q&As to service pages - **Build your brand**: AI rewards credible, popular, and recent sources - so focus on real authority building - **Avoid AI-written content**: Stick to human experience and expertise in your writing Quick win: Install the Easy Table of Contents plugin and add anchor-linked navigation to your long-form content. AI chatbots love this structure. Resources: [Learn WordPress](https://learn.wordpress.org/), [Developer Docs](https://developer.wordpress.org/), [WP-CLI](https://wp-cli.org/) ## ✅ Action Items #### WordPress - Core: Implement content chunking standards in block editor templates - Documentation: Create AI-SEO guidelines for content structure and FAQ integration - Themes: Develop theme patterns optimized for AI discoverability with proper heading hierarchy - Plugins: Evaluate table of contents and FAQ plugins for AI optimization features #### Community - Contributors: Share AI-SEO best practices through meetups and WordCamps - Maintainers: Consider AI crawler considerations in plugin/theme development - Sponsors: Support development of AI-friendly content tools and plugins #### Marketing - Campaigns: Focus on brand building and authority development over traditional link building - Content strategy: Implement BLUF writing methodology and granular content structure - Social calendars: Integrate CPR (Credible, Popular, Recent) framework into content planning #### Design - UX audits: Review content hierarchy and chunking for AI readability - Design tokens: Create systems that support extensive subheading and list structures - Research: Study how AI-optimized layouts affect human user experience ## 🧪 Hypotheses - Content structured with extensive chunking and FAQ formats will outperform traditional long-form content in AI search results - Brand authority signals (mentions, social following, press coverage) will become stronger ranking factors than traditional backlinks - Human-written content emphasizing first-hand experience will maintain competitive advantage over AI-generated content - The gap between AI-friendly and human-readable content formatting will decrease as AI processing improves ### 📊 Methodology - A/B testing chunked vs. traditional content formats across similar topic clusters - Tracking citation rates in AI overviews before and after implementing BLUF structure - Measuring brand mention correlation with AI search visibility using tools like Brand24 ### ❓ Questions - How will the conflict between AI-optimized chunking and human readability resolve as AI browsers emerge? - What happens when AI chatbots develop their own indexing systems independent of traditional search engines? - Can metadata solutions bridge the gap between machine-readable and human-readable content formats? ### 🔮 Future Research - Development of LLMs.txt or similar protocols for granular AI crawler control - Impact of multimodal AI capabilities on podcast and video content optimization - Evolution of brand authority measurement beyond traditional backlink analysis ### 💡 Implications - Practice: Content creators must balance AI optimization with human readability, requiring new editorial workflows - Policy: Need for clearer guidelines on AI crawler permissions and copyright protection for training data - Security: Risk of over-optimization leading to bland, homogeneous content that lacks authentic human perspective - Education: SEO professionals require training on AI-specific optimization techniques beyond traditional methods - Funding: Increased investment needed in brand building and authority development rather than just technical SEO ### 🧠 Learning Styles Assessment Reading level estimate: 12th grade (technical presentation with specialized terminology) #### 👀 Visual learners - Present: Screenshots of AI responses, citation examples, content structure comparisons - Improve: Flowcharts showing AI decision process, before/after content formatting examples #### 🔊 Aural learners - Present: Live presentation with Q&A discussion, verbal explanations of concepts - Improve: Audio summary of key takeaways, podcast-style breakdown of technical concepts #### 🗣️ Verbal learners - Present: Analogies (sky is blue, building blocks), defined jargon (LLM, chunking, BLUF) - Improve: More storytelling examples, case study narratives #### 🧩 Kinesthetic learners - Present: Hands-on demonstration of Gemini's double-check feature, practical exercises - Improve: Step-by-step content audit checklists, interactive content optimization workshops #### 🧠 Logical learners - Present: CPR framework, structured methodology for reverse engineering AI responses - Improve: Decision trees for content optimization, measurable KPIs for AI discoverability #### 🧑‍🤝‍🧑 Social learners - Present: Interactive Q&A session, discussion of industry challenges and solutions - Improve: Peer review groups for content optimization, collaborative case studies #### 🧘‍♀️ Solitary learners - Present: Individual analysis exercises, self-directed research prompts - Improve: Personal content audit templates, reflection questions on current strategy #### 🌿 Naturalistic learners - Present: Evolution of search ecosystem, organic content development processes - Improve: Ecosystem thinking about content relationships, natural language processing concepts ### 📊 Accessibility Assessment #### Screen reader semantics - Present: Strong heading hierarchy, descriptive link text for citations - Improve: More detailed alt text for screenshots, explicit labeling of visual examples #### Keyboard navigation - Present: Standard presentation format with logical flow - Improve: Ensure all interactive elements in examples are keyboard accessible #### Contrast and readability - Present: High contrast slides, readable font sizes for live presentation - Improve: Ensure color coding in examples meets contrast requirements #### Text resizing and reflow - Present: Presentation scales appropriately for different screen sizes - Improve: Responsive design considerations for online viewing #### Media and captions - Present: Live presentation with clear audio - Improve: Full transcript provided, chapter markers for key sections #### Forms and interactions - Present: Simple Q&A format with clear questioning protocol - Improve: Structured feedback mechanisms, accessible response collection --- ## Full Transcript [00:00] Thank you. [applause] [00:01] >> Yes, you have a mic, too. And uh yeah, [00:03] use that mic. And uh at the end when we [00:05] have questions, I'm going to pass around [00:06] the mic so it can [00:08] >> uh capture. [00:10] >> How do I turn it on? [00:11] >> Oh, you want to just use mine? [00:13] >> Sure, that's easier. [00:16] >> Hello everyone. Great to see those of [00:18] you again who I haven't seen in a while. [00:20] And uh awesome to meet or about to meet [00:24] the rest of you. Thank you for coming [00:26] out. Not true myself here. Um, oh, is [00:30] there a way to turn the like these [00:33] lights off to make this a little more [00:35] >> maybe [00:38] >> just imagine on the [00:40] >> that's probably better. [00:43] >> All right, so welcome to this session [00:47] where you're going to nerd out with me [00:48] on how AI chatbots are changing the [00:51] rules of content optimization. Um the [00:56] this is going to go beyond the [00:58] traditional SEO uh tactics. Although I [01:01] will start by saying that those are the [01:03] foundation of getting found in LLMs or [01:07] chat bots. I it's been just absolutely [01:10] hysterical over the last year because [01:11] everyone was like, "Oh, Chat GPT has web [01:14] search now or uh Google has AI [01:16] overviews. SEO is dead. It's really dead [01:18] this time." So somebody says that every [01:20] year for some reason, right? But no, [01:22] they're like, "No, Merilla is dead this [01:23] time." Oh, and by the way, how do I get [01:25] my website found in chat GPT? [01:28] [laughter] [01:29] >> Well, [01:31] starts with search engine optimization [01:33] because a little fun fact, which I'll [01:35] get to, but just as a precursor to this, [01:37] the AI chatbots as of now are just using [01:40] traditional search engines to search the [01:43] web for you. So getting found in short [01:47] in AI is getting found in search [01:49] engines. But there is more to it and [01:51] there's more to being um prefer having [01:54] your content your website content be [01:56] preferable to AI chat bots LLM and uh [02:00] we're going to get into all of that. So [02:03] that's basically what I just said and [02:06] specifically what we'll cover is um the [02:09] shift to this is another funny thing [02:11] over the last year it's like what the [02:12] heck are we going to call this? Um, I'm [02:15] going with AI SEO, but some people are [02:17] saying GEO for generative engine [02:19] optimization, LLMO for LLM optimization. [02:23] I'm just going with AI SEO, whatever the [02:26] heck you want to call it. Um, it's [02:29] basically, you know, I'm going with AI [02:31] SEO, getting found by AI powered search. [02:34] Um, so we'll touch upon that a touch [02:37] more than I just did. And then uh we'll [02:39] get into how AI chatbots determine which [02:42] web content to suggest to users. I think [02:44] that's important to understand like the [02:45] inner workings of how it all happens in [02:48] the background so that you can take [02:50] advantage of uh how of knowing how it [02:54] works to do what it likes basically. um [02:57] which also will include how to [02:59] conducting AI based topic research and [03:02] formatting your content to be easily [03:04] digestible and preferable for AI [03:06] chatbots to select. So to oh my [03:11] formatting got a little bit bit messed [03:13] up there but if you can't read what [03:15] there those are supposed to have [03:16] backgrounds but if you can't read that [03:17] it's search with AI and AI with search [03:19] and basically it's what I just described [03:21] is traditional search engines are adding [03:24] AI chatbot type functionality like AI [03:26] overviews at the top of Google and if we [03:30] can call them traditional chat bots [03:31] although it's been I guess three over [03:34] three years now since chat GPT launched [03:36] so traditional chat bots are now adding [03:38] uh web search functionality. So it's [03:41] basically all morphing into one thing [03:43] whether it came from one direction or [03:44] the other. It's just search web search [03:47] now has AI in it. [03:51] So LLM is a term that gets tossed around [03:54] a lot and uh if if you're not familiar [03:57] with that it stands for large language [03:58] model which is kind of the technical [04:00] term for an AI chatbot. So if you used [04:03] chat GBT, Gemini, Bing's co-pilot, any [04:07] of those, you've interacted with a [04:09] chatbot which is powered by an LLM. So [04:11] you'll hear us kind of like in the [04:12] industry like talk the use these terms [04:15] interchangeably. [04:16] LLMs are largely trained and constantly [04:19] retrained on the content of the internet [04:23] amongst other sources, books and other [04:26] things, but a lot of it is the internet. [04:29] And what they do is they predict text [04:33] patterns. So it's it's a little bit [04:36] weird to think about because it's it's [04:37] so natural when you talk to these [04:39] things. You feel like you're like it's [04:41] almost like thinking and responding. [04:43] It's actually predicting the next word [04:45] that is correct given the context. So [04:50] basically the same way that your brain [04:52] right now is probably filling in the [04:54] blank with [04:56] >> blue [laughter] [04:58] without even like you didn't even have [04:59] to think about that. It was just like [05:00] the sky is next word obviously is blue. [05:04] That's actually how these large language [05:06] models work. So whether it's answering a [05:08] question or just conversing it's [05:11] predicting the next right word in a [05:14] sentence basically at a large scale. So [05:17] like I said your brain just didn't even [05:19] have to think about that. It just knows [05:21] that that's what goes next. That's how [05:22] these LLMs work. So, after digesting [05:25] basically the entirety of the internet, [05:27] the LLM's perform that fill-in-theblank [05:30] exercise to answer a question or [05:32] converse with you. [05:35] So, why this is important with getting [05:37] your web website content found and [05:39] recommended by these LLM is because you [05:43] you got to keep in mind that they answer [05:45] first. It looks like they're summarizing [05:48] web search results for you. They [05:49] actually are answering first with that [05:51] general knowledge, doing that fill-in- [05:53] thelank exercise first and then going [05:55] and searching the web for sources to [05:58] back up what it already said. That's it [06:01] took me a bit in my head to like get [06:04] that wrap my head around that because [06:06] it's like it just seems counterintuitive [06:09] and it looks like especially on Google [06:10] with the AI overviews at the top. really [06:12] looks like it's summarizing the results [06:15] for you. Especially because a lot of the [06:17] times the stuff in the top 10 results is [06:19] what gets cited in the AI overview at [06:21] the top, but it's actually doing this [06:25] two-step process. It answers the [06:27] question with that fill-in- thelank type [06:28] text prediction, then goes and looks for [06:31] websites to back up the points of what [06:33] it already said. It's actually very [06:35] similar to what we as humans do when we [06:38] want to prove a point, right? We say [06:41] something that we believe to be true. [06:42] What color is the sky? Sky is blue. Now, [06:45] if someone said, "You need to prove that [06:48] to me." What [snorts] might we do? We [06:50] might go to a search engine and search [06:51] for some sources of some scientific [06:54] facts to back up our statement. And as I [06:56] said before, that's actually what the as [06:58] of now, I do believe this is evolving as [07:00] we speak. But as of now, the chatbots [07:04] are really just going and using Google [07:06] or Bing or some mix thereof to find the [07:10] proof to back up what it's saying. So [07:12] Chad GBT uses Bing as its default search [07:14] engine, as does Copilot. Gemini [07:17] obviously uses Google. Claude uses a [07:20] lesser known uh search engine, but that [07:22] has kind of a cult privacy following [07:24] called Brave. And perplexity uses [07:26] blended results from Bing, Google, and a [07:28] wide variety of other sources like [07:30] Wikipedia. As it stands right now, [07:32] they're really just using the search [07:34] engine for you. So, if you want to get [07:36] found in LLMs, get found in search [07:38] engines. [07:40] So, let's go deeper than that, though, [07:42] because otherwise this would just be a [07:44] plain old uh SEO talk where I would go [07:46] over all the basics, which I'm not going [07:48] to do at all. I know you have recently [07:49] had a very good SEO talk from Susan. So, [07:52] bypassing all of that, but know that it [07:54] all matters, okay? All the foundational [07:56] stuff definitely still matters for SEO, [07:59] but we're going to just talk about what [08:02] to layer on top of that to really appeal [08:05] to these AI powered search engines. [08:08] So, back to the sky is blue analogy, [08:11] your your brain fills in blue as the [08:13] answer for blank because of a couple [08:16] reasons. One, you hold that opinion [08:17] yourself. You've also heard others state [08:20] it. and you know that scientific studies [08:23] and trust signals and credible resources [08:25] exist that you could find to back up [08:27] what you're saying. Um so again, if you [08:29] needed to prove that, you would go to a [08:31] search engine, type in something along [08:33] the lines of proof that the sky is blue, [08:35] and then you would go through and select [08:37] what you feel are the most compelling [08:38] sources and present them to your [08:40] audience. But here's what to really kind [08:44] of have stick with you is when you're [08:47] doing research and looking for sources, [08:49] where does your brain go? Like, how does [08:51] your brain assess what's a reputable [08:54] source? Which one should you pick to [08:56] help prove your point? Because that's [08:57] basically what the LLMs are doing when [08:59] they're providing web links. They've [09:01] already made a point and they're looking [09:02] for data to back that up. Um, so I I [09:06] turned this into an acronym to help it [09:08] be a little easier to remember. Um, I'm [09:10] calling it CPR. We select sources that [09:12] are credible, popular, and recent. And [09:16] there's a lot more to it than that, but [09:17] that's kind of like in my mind, that's [09:19] the top three. It's like, is it [09:20] credible? We're more likely to pick [09:22] something that backs up our point from [09:24] like NASA when it comes to talking about [09:26] this guy or by someone who has a PhD [09:29] than some lesser recognized, less [09:32] credible source. And we're also going to [09:35] look at like social proof. So that's [09:37] where the popular comes in. Like if you [09:39] had to pick a video to prove your point, [09:41] you're going to pick one like the [09:43] example here, 14 million views. And also [09:46] there's um you know, I always find [09:47] myself really focusing on recency as [09:51] meaning more, I guess, because it's [09:53] like, yeah, okay, so 15 years ago, some [09:56] PhD or NASA said the sky is blue, but [09:59] what have we learned since then? So [10:00] recency is definitely a factor as well. [10:04] And then of course we as humans, but [10:07] also LLMs do this too, select sources [10:09] that support our beliefs. It's [10:11] confirmation bias. So if I'm presented [10:14] with these two options when doing my [10:16] research on the point that I want to [10:17] prove, uh, understanding why the sky is [10:19] not blue or understanding why the sky is [10:21] blue, guess which one I'm going to pick. [10:24] And you will see that happen. Um I I [10:27] didn't want to include too many [10:28] screenshots, but I did when I was [10:29] pulling screenshots for this, I did a [10:31] little bit of exercise where it was like [10:32] proof that the sky is blue and then it [10:35] gave me certain results and then proof [10:37] that the sky is not blue and it actually [10:40] gave me like the first one was like [10:41] questioning is the sky really blue like [10:43] digging into is this really real or is [10:45] it a facade or whatever. So it will kind [10:47] of twist uh its answer or its selections [10:51] of sources to back up uh the point that [10:54] it thinks you want to make. Um but [10:56] certainly more about like what it is [10:58] already telling you. So like I said they [11:01] first answer the question based on their [11:02] own knowledge. They meaning like [11:04] personifying them already. [laughter] [11:07] they as uh the things that they are the [11:10] technology um answers the question based [11:12] on their knowledge and then looks for [11:14] the credible popular and recent sources [11:17] to back up the conclusion they already [11:18] made that's the where the confirmation [11:20] bias comes comes in for them. So, if you [11:24] want AI chatbots to let's start high [11:27] level, we'll get into some nitty-gritty, [11:29] too. But high level, if you want an AI [11:31] chatbot to know a fact about your brand [11:34] and tell that to other people, first of [11:36] all, I'm going to pause there and say [11:38] the one thing you should not do and does [11:39] not work that a lot of people think [11:41] works is training the chatbot by talking [11:44] to it. People think that like I've heard [11:47] this so many times like oh I'm training [11:49] chatbt about my company and my services [11:52] and my product by telling it all about [11:54] it. I'm like, "Oh, God." Okay. And it [11:57] but it told me back. It told me back [11:59] what I said. Yes. Yes. They do have [12:01] memory limited private memory about your [12:05] own chats. So, I always tell people [12:07] like, "Don't take my word for it. Do [12:08] this exercise. Make up a fake word. [12:10] Train chat GBT or some chatbot what your [12:13] fake word means. Ask it. It will tell [12:15] you the definition of your fake word [12:17] back to you, but then ask your friend to [12:19] ask their chatbt what the fake word [12:21] means. It will not know because that [12:23] memory is limited." and not training [12:26] data for the LLMs. What is training data [12:29] for the LLMs is the internet. So if you [12:32] want the LLM to know a fact about your [12:35] brand, put it on your website. First of [12:38] all, emphasize not just put it there, [12:40] but really emphasize that fact on your [12:43] own website. But then reinforce that [12:46] because remember sky's blue example, you [12:47] know, we we believe that ourselves, but [12:49] we know that others believe that too. is [12:51] the second part that really makes us [12:52] just do that fill-in- thelank type [12:54] answer, that automatic answer. So, it's [12:57] important to get other reputable sources [12:59] to talk about your brand that way also. [13:02] And then third, build those CPR signals, [13:05] credibility, popularity, and recency to [13:09] back that up. So real world example uh [13:13] just as your brains say blue for the [13:16] color of the sky I have and I hope it I [13:19] hope it sticks for a while but as of now [13:21] don't want to jinx myself knock on wood [13:23] I have trained uh LLMs to associate my [13:27] company name Hayman marketing with [13:29] womanowned SEO agency and it's is a [13:32] niche search but it is something that a [13:34] lot of women business owners search for [13:36] and uh so I've made it into the [13:39] automatic fill in in the blank response [13:41] with the major LLMs for that by doing [13:44] basically what I just said. I optimized [13:47] my website with that term, really [13:49] emphasized that on my own website, [13:52] described my company that way on other [13:54] reputable websites as well. And then [13:57] then did the super simple but definitely [14:00] not easy part of building the [14:02] credibility, the popularity and and the [14:05] recency of like keeping up with keeping [14:07] my website um fresh with regularly [14:10] published articles and publishing things [14:12] elsewhere in uh having a really strong [14:15] professional bio and credentials, press [14:18] coverage, winning awards, building a [14:20] large social following over many years. [14:22] Like I said, simple but not easy. But in [14:25] short, like if one if there's one key [14:28] thing, well, I'd hope you take away more [14:30] than one key thing from this, but one of [14:32] the key things I'd like you to take away [14:33] from this is that 2026 is going to be [14:35] about brand building. It's definitely a [14:38] bigger factor than ever. I've always [14:40] preached that helps helps to have a [14:42] strong social following. There's always [14:43] been a correlation between strong [14:45] between content that performs strongly [14:47] on social media and content that [14:48] performs well in search engines and you [14:51] know PR getting inbound links and stuff [14:54] has always helped with SEO. man. Like [14:57] now more than ever, the brand building, [14:59] truly building a brand for real in the [15:02] real world, it's going to matter more [15:04] than ever because the search engines [15:06] took a very simplistic look at that. It [15:08] would look at your social following or [15:10] your social signals and content and [15:12] they'd look at what other reputable [15:14] relevant websites uh pointed to your [15:16] website by linking to it. But it kind of [15:18] stopped there. the LLMs do have this [15:21] broader like thinking capacity to look [15:22] at the brand and assess it from that [15:25] broader perspective of you know just how [15:27] many times have you been mentioned in [15:29] the news like forget lakes like even if [15:31] not linking to your website at all um [15:33] and just really digging into social [15:35] content and really understanding like [15:38] the the strength of a brand so branding [15:41] is a bigger part of all of this than [15:44] ever before whether we decide to call it [15:45] SEO l whatever the heck It's all getting [15:49] found. Online visibility, like real good [15:52] oldfashioned branding is is definitely [15:55] got to be center stage in the coming [15:57] year. [15:59] Pause there for a quick breath. And [16:02] since we have a small informal group, [16:03] any burning questions? We'll of course [16:05] do formal Q&A at the end, but if anyone [16:07] has any burning questions right away, [16:08] >> I have a mic that I can pass around to [16:10] anybody that does have a question, so [16:12] it's captured on the video. [16:14] >> Oh, you have? Of course you do. Yeah, [16:17] I'm the one. Okay. [16:19] So, how are we where are things in terms [16:23] of uh LLM's un I have a couple [16:27] questions, but we'll start with this [16:28] one. Understanding let's say podcasts [16:30] and audio content and putting that in [16:33] the mix um and and using that as a way [16:38] of building brand. And when you talk [16:40] about brand building, I'm assuming [16:41] you're also talking about personal [16:42] branding, not just corporate branding. [16:45] Yes, definitely. All right. Yep. [16:47] Personal and corporate. Um, whichever [16:49] applies or both if both are applicable [16:51] in any given situation. Um, and then [16:54] yeah, multimodal. That's something I [16:55] didn't dig in too much in this [16:56] particular presentation, but good point [16:58] to bring up. Uh, search is more [17:01] multimodal or multimedia than ever [17:03] before. As it stands right now, like I [17:05] said, the LMS are still for the most [17:07] part relying on the traditional search [17:08] engines, but the AI that's been baked [17:11] into the search engines for some years [17:13] now is getting much better. multime [17:14] media. So that is a something that the [17:18] content itself will is starting to be [17:20] better understood than ever. The content [17:22] I mean like audio content, video content [17:24] and imagery um beyond like alt text you [17:28] know it's like actually starting to be [17:30] understood by the search engines and uh [17:33] certainly by you know the LLMs. Um but [17:36] simply just having those things for [17:38] brand building that's kind of like a [17:40] separate thing right like the technology [17:41] of it yes is evolving a lot but also [17:44] just simply having those things is is a [17:47] brand building [17:48] >> okay one other question that you'll kill [17:49] me for but how good or how big probably [17:53] in your opinion uh is uh AI in terms of [17:57] uh creating SEO titles descriptions tags [18:02] how can we rely on that what do you [18:04] think about that [18:07] Very very very good question and um I'll [18:09] definitely touch upon this more in the [18:11] coming bits here but the short answer [18:14] would be used sparingly and responsibly [18:18] for small bits and pieces like those. [18:21] >> I would give that a thumbs up. [18:22] >> Okay, good. [18:24] >> With a big caveat that's coming at the [18:26] end. [laughter] [18:28] >> So perfect perfect teeing up. Thank you. [18:30] >> Any more questions before him? Okay. [18:35] Can I ask a question? [18:36] >> Thank you. [18:38] >> Uh my question is first of all is gio or [18:40] like generative engine optimization is [18:42] kind of worth it if your SEO isn't uh [18:45] great to begin with and kind of follow [18:47] up to that. So what is kind of the [18:48] upside potential of GEO compared to SEO? [18:53] >> Great question. So um yeah, like I said, [18:55] I'm going to dig into mostly the AI [18:57] based stuff in this presentation, but [19:00] absolutely the SEO stuff is a foundation [19:02] is like the core. You have to have that [19:04] before any of this will work. So yeah, [19:07] to your point is is GEO, whatever we're [19:09] calling it, worth doing, worth tending [19:11] to without tending to the basics. No, [19:14] >> no, the basics need to be tended to [19:17] first. I mean, for in short, because as [19:20] it stands right now, the AIS are simply [19:21] using traditional search engines to [19:24] search for what they're the web links [19:25] that they're giving. So, um yeah, [19:27] definitely got to start with all of the [19:28] all the traditional stuff and then um [19:31] for the second part like [19:34] uh I guess you're kind of like [19:37] leading into what's the like quantifying [19:40] the upside. [19:42] Okay, so I mean that's a tricky one [19:45] because AI overviews at the top of [19:48] Google have dramatically decreased [19:50] click-through rates. They've [19:52] dramatically increased impressions [19:55] though. So, it depends on how you look [19:58] at that. Is that glass half empty or [20:00] glass half full? I like to look at it as [20:03] glass half full because all getting out [20:07] there, you know, people people spend a [20:08] lot of people, companies, brands spend a [20:10] lot of money just to get in front of our [20:12] eyeballs, right? You know, Coca-Cola [20:13] billboards and TV Super Bowl commercials [20:16] and and on a smaller scale display [20:18] advertising and social media [20:20] advertising. So those impressions I feel [20:24] like count for a lot towards brand [20:26] building. Um but you know to be fair [20:29] losing the clicks, having less clicks [20:31] coming to your website because of all [20:33] this AI stuff is obviously a downside. [20:36] Um but what's interesting that we're [20:39] seeing alongside of the trend of [20:40] impressions going up is the alligator [20:42] mouth. I have a video online if you [20:44] follow my socials about the alligator [20:46] mouth. Alex did a nice edit for me with [20:48] that. What I mean by that is in uh in [20:50] Google Search Console where it shows how [20:52] often your website's shown in Google [20:54] versus clicked on, a lot of people are [20:55] seeing an alligator mouth in their [20:57] charts where impressions are going up [20:59] but clicks are going down. But what's [21:01] interesting is alongside of that, a lot [21:02] of cases, not every case, we're seeing [21:04] conversion rates go up along with it. So [21:08] you may be getting less clicks to your [21:10] website, but the ones you are getting [21:11] are more likely to convert. can be more [21:15] likely to be like I said not all cases [21:16] but a lot of cases we're seeing that so [21:19] it's people are still clicking and it's [21:21] the people who wanted to take an action [21:24] if they just were wondering if what [21:26] color is the sky they're going to read [21:27] the AI overview they're going to stop [21:29] they're not going to click but as it [21:31] stands right now you still need to click [21:34] in order to transact with a business to [21:36] purchase something or to reach out to [21:38] them make an appointment it's you know [21:40] so the clicks that you do get are more [21:42] likely to convert [21:44] long long answer to the potential lift [21:47] or upside I think is is huge and I think [21:51] you know if you it's also just Google's [21:53] world and we just live in it. So if you [21:54] don't roll with what they're doing [21:55] you're going to be left behind so you [21:57] might as well optimize to get found in [21:59] those AI overviews and these other AI [22:01] powered um chatbot search engine things. [22:06] All right. So, back to the concept of [22:11] the LLM's answer first and then go [22:15] looking for sources to back up what [22:17] they're saying. This requires us to [22:20] really think a little bit more backwards [22:23] than ever before. Um, so I explained how [22:26] if you want the LLM to know a fact about [22:28] your brand, the example that I gave, but [22:31] when we're now talking about like more [22:33] long form content such as, you know, [22:35] thought leadership articles, blog [22:37] articles, etc., it works the same way, [22:39] but it's much much much more granular. [22:42] So basically, I'm sure you've noticed [22:45] that AI answers are very bulleted, very [22:49] chunked up, broken up into little [22:51] tidbits, right? So each and every tidbit [22:54] of information that it gives is a [22:57] separate fact that could have not always [23:00] does but could have a separate and [23:01] different citation to back it up. [23:04] Citation meaning web link. So like gone [23:07] are the days of thinking of a single web [23:11] page as a single thing to optimize. It's [23:16] these individual tidbits and chunks of [23:18] information that are the opportunities [23:20] to get found. Um, so basically like in [23:25] in the way that AI works, it's actually [23:28] it's called chunking. It's kind of a [23:30] gross word, but it's called chunking. [23:32] When an AI digests information, it [23:34] chunks it up into little tidbits and [23:37] stores it that way, all mixed up. So [23:39] imagine like a big Excel spreadsheet. [23:41] And instead of how u traditional search [23:44] engines store web pages, which is like [23:46] one row, one whole page of content, next [23:48] row, one whole page of content, it the [23:51] LLMs use what's called the vector [23:53] database, and it just takes all those [23:54] different tidbits, mishmashes them all [23:57] into different cells, and then [23:59] reassembles them into answers when you [24:01] ask it something. So a web page being a [24:04] single thing in an index still [24:08] technically is a thing in traditional [24:09] search but not in AI powered search and [24:13] so that it's and it's just going to keep [24:15] going in that direction even you know [24:18] beyond the days where the LLMs are still [24:21] using the traditional search engines to [24:22] find things like it's already kind of [24:24] like even though that's true it's [24:26] already just changing the way that we [24:28] get found. We get found by chunking up [24:32] our content into think of it like a [24:35] brick wall. I like this visual for this. [24:37] Instead of thinking of a web page as a [24:39] single thing, you got to chunk it down [24:42] into building blocks, separate sections [24:45] or all separate opportunities to get [24:48] found. So you're optimizing. It sounds [24:50] very overwhelming, I know, but in a [24:53] sense, I find it a little bit more in [24:54] the way my brain works a little bit [24:56] easier to tackle in a sense than than [24:58] optimizing a whole big long thing with [25:01] big paragraphs. It's like, no, I'm going [25:03] for this here and that there and this [25:05] here. And I I think it can make it a [25:07] little bit easier, but I understand it [25:08] sounds overwhelming, but the the [25:10] thought, the perception is what's [25:12] important that we no longer can think of [25:14] a single web page as a single thing to [25:16] optimize. It's the building blocks of [25:19] the page are all separate things to [25:21] optimize. And an example of that is I [25:25] know this is probably really small um [25:27] but basically it's a an example where I [25:31] ask AI how to do a forensic SEO audit [25:35] and it gives as per usual bulleted [25:39] numbered lists step-by-step advice. And [25:42] in this case, the yellow circles are [25:44] where it picked an article because I [25:46] have an article that happened to be I [25:47] actually wrote it prels, but it happened [25:50] to be in checklist format. So it lined [25:53] up nicely with the way the AI wanted to [25:55] give the answer. And so it cited my site [25:57] as proof of points one, two, and three. [26:01] And then other sites, the orange circles [26:04] are where it linked to other sites to [26:06] prove its other points. And so only [26:10] pieces of my article were recommended [26:12] essentially. And so basically it's on me [26:15] now to reverse engineer what it wants to [26:18] say and go back to my article and revise [26:21] it to make the points that I didn't make [26:23] or I didn't make strong enough in order [26:25] to try to capture more of those other [26:28] remaining points. I know it looks like [26:30] spaghetti on the wall or something, but [26:31] basically it's just, you know, going [26:34] this this is what reverse engineering [26:36] the AI answer looks like in practice is [26:40] I'm looking at my article over here and [26:41] I'm looking at its answer and how it [26:44] picked other websites other than mine to [26:45] back up its answer. I'm like, well, I [26:46] got to dig into this more. I didn't even [26:48] say this. I don't even say this. I got [26:50] to expand on that. And that's what I [26:52] mean by reverse engineering the the AI's [26:55] answer to do this from scratch. If you [26:59] want to write or rewrite an article [27:01] about a topic, basically what I [27:03] recommend is going to Gemini. I picked [27:05] Gemini because Gemini is the AI that [27:08] powers the AI overviews at the top of [27:10] Google, which as of now still does have [27:12] a 90 whatever percent search market [27:15] share. So I do like to go to the source [27:19] of Google itself. So go to [27:21] gemini.google.com, google.com. Ask it [27:24] about your topic to tell you about the [27:25] topic that you want to write about. And [27:27] then there's this little three dot thing [27:30] at the bottom that if you click it, I [27:32] mean, you'll see the little link icons [27:34] in there, but this is this is nice view. [27:37] If you click the three dots and then you [27:38] go to double check response, what you'll [27:40] see is it will h it will go through its [27:43] response again. It won't change it. [27:45] it'll kind of like comb through it again [27:46] and highlight more specifically and [27:49] obviously where it got which points. And [27:52] if you click on the link, like this [27:55] example of um weather.gov, if you were [27:58] to click on that, it would actually take [27:59] you right to the part, it's like an [28:01] anchor link, it would take you right to [28:02] the part of the article that it cited [28:03] and it'll highlight it. So you can just [28:05] kind of see it in action, like see [28:07] behind the scenes what what site it [28:09] picked, what sentence it picked, and you [28:13] can kind of infer why it picked that. [28:17] Equally important, if not more though, [28:19] is when it doesn't find a citation to [28:23] back up something that it said. That I [28:25] call the golden opportunity for you to [28:28] write something. Again, just think of [28:30] not a whole article, but some building [28:33] block of an article that really [28:35] emphasizes that point that it seemingly [28:38] would like to have a citation for that [28:40] it doesn't. Now, that's not to say [28:42] there's no content on the internet that [28:44] talks about how red light has long lazy [28:47] wavelengths. I'm sure there is, but it [28:49] just didn't seem to find the right [28:51] combination that it liked of it being, I [28:53] guess, described that way from a [28:55] reputable source in the right context. [28:57] And so it's an opportunity to get found. [28:59] Obviously, the more niche the topic, the [29:01] easier the opportunity is. [29:04] So on top of that reverse engineering [29:06] exercise where you're looking at the [29:09] topic, finding the answer the LLM is [29:11] going to give, and people like are [29:14] asking me like, really, I just have to [29:15] reverse engineer what it says? Like what [29:17] if I want to make more points? Well, [29:19] yes, of course you can, but it's only [29:21] going to say what it's going to say. At [29:22] least as far as right now. Maybe you can [29:24] convince your content can convince the [29:26] LLM to start making a new point about [29:29] something, but as it stands now, yeah, [29:32] they're going to answer what they're [29:33] going to answer and you're either going [29:35] to be cited in it or you're not going to [29:36] be cited in it. So, you have to follow [29:39] their lead. But on top of that, there's [29:40] other things you can do to make your [29:42] content more easily digestible and say [29:45] those two words back, digestible and [29:47] preferable for AI chat bots to select. [29:51] Um, so beyond this reverse engine that I [29:54] can't say that too fast either. Beyond [29:56] this reverse engineering exercise, um, [30:00] there's also some structural things that [30:02] can make your content much more easily [30:05] discoverable by LLMs. So, as I said [30:09] before, LLMs have no choice but to store [30:12] content. Just by their design, the way [30:14] their databases are designed, they have [30:15] no choice but to store information in [30:18] tiny tidbits. So whether you break it up [30:20] for them or not, it's going to make its [30:23] best attempt to take any content you [30:24] give it and break it down into tiny [30:26] chunks. But if you do it for it, it does [30:29] it a lot more accurately and effectively [30:31] and then therefore it's more confident [30:33] in your information because it was it [30:36] made sense to it because it was chunked [30:38] up. So, gone are the days of writing uh [30:42] blog articles like essay, like college [30:43] essays with like five to 10 sentences [30:46] per paragraph and little to no [30:48] subheadings and just having something [30:50] look like a term paper when you zoom [30:52] out. And this is this is the way to [30:54] write now. And I know it drives some of [30:55] my copyriter friends insane. [30:58] But again, it's Google's world. We just [31:00] live in it. So, you want to get found, [31:03] this is the new way. of super super [31:05] granular tidbits, one to three sentences [31:08] per paragraph, lots and lots and lots of [31:10] bulleted lists, number of lists, lots [31:12] and lots and lots lots of subheadings [31:15] and subheadings to your subheadings and [31:17] subheadings to your subheadings to your [31:18] subheadings. As much as makes sense. [31:20] Obviously, don't do any of this to any [31:22] kind of ridiculous level just for the [31:23] sake of doing it. But as much as a topic [31:26] can be broken down, it should be broken [31:28] down. [31:30] AI chatbots [31:32] also absolutely love FAQs because the [31:35] Q&A format is just kind of the way it's [31:37] like their language with the way they [31:38] roll. So adding unique and different [31:41] FAQs to as many different pages on your [31:43] website as possible is a highly [31:46] recommended kind of lowhanging fruit [31:48] thing to do. Um it's pretty simple. You [31:50] can just look at the people also ask [31:52] section of Google about a topic and [31:54] those are the questions that you can go [31:56] ahead and add to let's say your service [31:58] pages on your website. I do want to [32:00] clarify that um a lot of times I bring [32:03] this up and people say, "Well, I already [32:05] have an FAQ page on my website." I'm not [32:06] talking about an FAQ page. That's great. [32:09] That's fine. People love those. That's [32:11] fine. If you have a high level FAQ by [32:12] your company, leave it there. But also [32:15] go to each individual core page that's [32:18] about the product that you sell or the [32:20] service that you do and add specific [32:23] FAQs about that page to that page. [32:28] And FAQs are great to incorporate in [32:31] long form content as well. So, back to [32:34] kind of like the blog article style [32:35] content, using those since you need to [32:38] have a lot of headings and subheadings [32:40] and subheadings to the subheadings, use [32:42] the FAQs as your subheadings for your um [32:46] [snorts] granular content. So, kind of [32:48] we weave those in that way. LMS also [32:52] really like tables of contents. This is [32:54] something um Alex who works with me we [32:58] uh he's always adding to our clients [33:00] sites is these anchor linked table of [33:03] contents not on all pages obviously but [33:05] for long form content like that thought [33:07] leadership articles blog content type [33:09] stuff it should always have now that you [33:11] have to have so many different [33:13] subheadings to subheadings subheadings [33:15] it should always have an anchor link [33:17] table of contents at the top um AI [33:20] chatbots love that personally I as a [33:22] human love that too because I'm just I [33:24] don't want to read the whole thing. A [33:25] lot of times, you know, we they're all [33:27] moving at a million miles an hour. I [33:28] like when I be able to just be like, [33:30] "Oh, this is the part I want to read [33:31] about." Click on it, jump to it. Um [33:33] which is great for AI discoverability as [33:35] well. WordPress makes this easy with [33:38] plugins. The one that we use is um easy [33:41] table of contents by magazine 3. I'm [33:44] sure there's other great ones out there [33:45] as well, but I like that one. It just [33:48] pop backwards for a second. It lets us [33:50] um customize like how many levels down [33:52] do we want to go? uh we just added this [33:54] plugin to a site that didn't have too [33:56] many levels down and if they did it was [33:59] in some I don't know some structure that [34:00] didn't make sense so we were like yeah [34:02] on this site let's just go two levels [34:03] down but like on my site we'll go [34:06] however five levels down um so I like [34:09] custom custom customization capability [34:12] of that one all right two more key [34:15] points and then we'll get to Q&A another [34:19] um tip on how to make your content AI [34:22] friendly is to write bluff style. Um, [34:26] I've been hearing about this. I had I [34:28] just realized today when I was looking [34:29] for screenshots for this is that it's a [34:32] military term, but I've been hearing a [34:33] lot of about it in terms of AI [34:36] discoverability. Bottom line up front is [34:39] what it stands for. It's basically make [34:42] your main key point immediately at the [34:45] top of the article and for each section [34:48] because remember a one article is not [34:50] one thing anymore. is not a thing. It's [34:52] a bunch of building blocks. So, within [34:54] each building block, also get right to [34:56] the point, make the point right away, [34:58] then provide the supporting detail. [35:00] Apparently, it's like a military [35:02] communication thing, too. I didn't know [35:03] that. But, um, it's really like putting [35:06] your conclusion sentence first or [35:08] another similar thing is like the TLDDR, [35:10] too long, didn't read, you know, the [35:13] Reddit style thing, but right up at the [35:15] top. Um you again think to the AI [35:18] structure of its answers and it's just [35:20] like it loves to just answer it directly [35:22] first and then give all the supporting [35:24] detail. So if you write your content [35:26] like that in small chunks but with that [35:28] answer first and then supporting detail, [35:30] it's going to do a lot better at getting [35:33] recommended by AI chat bots. So again [35:36] the the whole article and each section [35:39] within it. So you want your key point [35:41] for your like conclusion of the whole [35:44] thing first. And but then in each [35:46] section, don't forget, they're each [35:48] individual bricks that can stand on [35:50] their own and get found on their own. [35:51] Now, so in each and every section, you [35:53] want to start with that clear, succinct [35:56] statement of what's the key information [35:58] here that you need to know and then the [36:00] supporting detail after that. [36:03] So here's the part where I'm going to [36:04] preach about using AI to generate [36:07] content. Um, so Bud asked before about [36:10] using it for like titles and meta [36:12] descriptions and things like that that [36:15] that I can get on board with, but at [36:17] large you do not want to. It's so [36:20] tempting. I know. I know how tempting it [36:22] is. I'm tempted every day. It's so [36:24] tempting to have AI write your content [36:26] for you. it comes out really nice and [36:29] just it's eloquent and it's easy and [36:31] it's just but unfortunately [36:35] it's not good for um your website's [36:38] discoverability because it is people say [36:42] to well oh I rewarded a bit or I ran it [36:44] through this other thing that makes it [36:46] undetectable but the if there's one [36:48] thing AI is really really good at is [36:50] recognizing other AI generated content [36:53] it's like oh this is a match for me and [36:55] how I wear things it's just too easily [36:57] detectable. Um, it's also embedded with [37:00] like subtle hidden characters sometimes [37:02] or even just token patterns meaning like [37:05] word p common word patterns. Um, I'm [37:07] sure everybody at this point recognizes [37:09] that long dash which I didn't even know [37:11] the name of it. M dash dash dash [37:14] emphasis dash [37:16] >> that is like a dead giveaway because [37:18] nobody uses it quite that way except for [37:20] AI. So those little clues are just like [37:23] it's just too obvious when you put them [37:25] all together. Even if you do edit, you [37:28] really you have to edit it to the point [37:30] of like editing so much you might as [37:32] well written it from the beginning that [37:34] way yourself. Um, but one of the most [37:37] important things to remember is that [37:39] LLMs are actually trained to try to find [37:42] original sources because they're trained [37:45] to try not to create a game of [37:47] telephone. The last thing an AI wants to [37:50] do is provide as its proof a source that [37:54] some other AI [37:55] gave generated because that's just a [37:58] summary of something else. It's a [37:59] summary of a summary of a summary like [38:01] the game of telephone things get get [38:03] diluted and so they're really looking [38:06] for original content anyway. Um and on [38:11] top of that straight from the words's [38:13] mouth by horse I mean Google not John [38:16] Mueller but Google [38:19] says they don't want you to do it and [38:21] they send some mixed messages. I know [38:24] they say, "Oh, well, you know, in this [38:25] context or in that context, read between [38:28] the lines." And I mean, really like go [38:31] read their Google's quality raider [38:32] guidelines, which isn't technically a [38:34] literal definition of their algorithm, [38:36] but it is what they use. They they give [38:39] their human testers to test the efficacy [38:42] of their algorithm against. So [38:44] basically, it's what their algorithm is [38:45] designed around is these quality raider [38:47] guidelines. They have revised them in [38:49] several places uh but most notably this [38:52] one to state that AI generate if the [38:56] majority of the what they call the MC [38:57] the main content the majority of content [39:00] is auto or AI generated it should be [39:03] rated as lowest quality. So there it is [39:06] John Nuller Google putting it up on the [39:08] screen. I know there's a lot of debate [39:09] about it. So I was like I loved getting [39:11] this photo. I was like putting that on [39:14] [laughter] [39:15] but used sparingly used for ideiation is [39:18] great. Create an outline. Absolutely. [39:20] Yeah. I mean I told you before to [39:21] reverse engineer the answer. So yes use [39:24] AI to come up with your content outlines [39:27] for sure. Um ideas and just any any help [39:31] you need other than straight up writing [39:34] the whole thing. But title tags, meta [39:37] descriptions, I alt text, I think those [39:39] kinds of things are are fine. That's not [39:42] like the the beef, the meat of the [39:44] content where the AI is really going to [39:46] be looking for that that firsthand [39:48] experience. And speaking of that, I'll [39:50] close out with this. Google's been [39:52] telling us for many years to demonstrate [39:54] EAT in our content, which is expertise [39:58] or was expertise, authoritiveness, and [40:00] trustworthiness. They have gone ahead [40:03] and coincidentally or not so [40:04] coincidentally the very year that chatb [40:06] was released they formally revised it to [40:09] EAT they added this new E which is [40:12] experience they defined as your personal [40:15] human firthand real world experience [40:19] because you probably don't know if you [40:21] can see what that says that's a fire [40:22] hose of bland content little bits of [40:24] paper that say bland content [laughter] [40:27] that's what happened the moment chatgpt [40:29] got released was everyone started [40:30] writing website content with it. It's [40:32] all bland, generic. It's all repetitive. [40:34] And immediately Google was like, "We [40:37] need original content out there on the [40:38] web. So, we're going to tell everyone, [40:39] you now have to talk about your personal [40:41] human real world experience because [40:44] that's something that AI just can't [40:47] replicate for you." So, super important, [40:50] the new E. [40:52] So, with that, all of that, I know that [40:54] was a lot, but uh hopefully it made [40:58] sense and some of it some of it sticks [41:00] with you and happy to answer any [41:03] questions you may have. [41:05] >> Again, we have the mic, so just raise [41:07] your hand. I'll come over. [41:12] >> Everybody had their questions midway [41:14] through. [laughter] [41:15] Answer everybody. [41:16] >> It was either completely overwhelming. [41:17] Everyone's like, "Darlight, I understood [41:19] none of that." Or maybe all my sense. [41:23] >> I just [41:23] >> Yeah, [41:25] >> you might as well give it to me. [41:26] >> But you got something to say. [41:27] >> All right. Yeah, I don't. [41:28] >> There you go. So, [41:29] >> you have to Oh, yeah. We're taping. [41:32] Okay. So, [41:34] I'm hearing from web hosts that they're [41:36] adding like new a new feature [41:39] where you can block um uh AI bots from [41:44] indexing, if that's the right word, now [41:46] who knows, uh content. [41:49] Is that have you heard that kind of [41:51] thing? Or can you just reuse robots uh [41:54] the robots file or what? What are you [41:56] what are you hearing about that? [41:59] >> Yeah, awesome questions. [42:00] >> Because I don't like to write something [42:01] and then have it all just eat, you know, [42:03] >> I mean, I don't really care, but you [42:04] know, [42:05] >> well, but so that's my my instant [42:07] reaction is why would you not want it to [42:10] find and recommend your content? I don't [42:13] I know that. Well, but but for the [42:14] purposes of of [42:16] >> Yeah. But but for the purposes of um [42:19] just feeding the machine, I don't want [42:23] you know content to be or there might be [42:25] companies that don't because I write for [42:27] web hosting, they may not want that [42:29] content [42:31] uh used. [42:33] >> Yes. Yes. Yes. From the copyright [42:36] perspective, from the training data [42:38] >> training [42:38] >> perspective, right? So there's different [42:40] perspectives there. You're absolutely [42:41] right. It definitely depends on what [42:43] perspective you're looking at. This [42:44] entire presentation was about like the [42:46] getting found getting I want my articles [42:48] to get found. I want my website service [42:51] pages to get found. [42:52] >> So therefore they have to be consumed by [42:55] the LLM in order to get found. Now the [42:58] copyright angle that is that is [43:00] different. There's bazillion lawsuits [43:02] going on now like New York Times is [43:04] suing OpenAI is one of the most famous [43:06] ones that's still ongoing. um where they [43:08] you're like you trained this thing using [43:11] our copyrighted content and that's not [43:14] okay. So from that perspective yes that [43:17] things are being developed particularly [43:18] for publishers who have been hit the [43:20] hardest by this whole AI overview reduce [43:23] clickthrough rate thing um where there's [43:25] technologies being looked at for how to [43:28] tell the AI bots that they can or cannot [43:31] um crawl your content for training. So [43:34] again, we've we've switched gears. Not [43:35] for the discoverability because if you [43:37] put something on your public website, [43:38] it's not a subscription thing or [43:40] anything like it's if it's indexed in a [43:42] search engine, it's going to be found by [43:43] AI. But if you're putting other kind of [43:45] content out there and it's getting [43:47] despite your best efforts, it's, you [43:49] know, that's supposed to be paywalled or [43:50] it's supposed to be behind some kind of [43:52] a login or whatever. Um, gated is the [43:55] word I was looking for. Uh, and it's [43:57] still getting picked up by them. Yeah, [43:58] there's it's there's no like clear [44:00] answer right now. Um, there's a couple [44:02] different technologies being looked at [44:03] for how you can tell the AIS like yes or [44:06] no you're allowed to or not allowed to, [44:08] right? [44:09] >> Um yes, you could try using robots set [44:11] text. Right now the efficacy is whether [44:15] they're respected. [44:17] >> So you can't you can't in other words [44:19] you can't separate SEO like traditional [44:22] SEO getting found in Google from AI [44:27] um indexing or AI [44:29] >> as of now. No, because as it stands now [44:32] really the AI chatbots are just using [44:34] the search engines to find and recommend [44:37] web links. I do believe they're working [44:39] on their own indexing and a whole [44:41] different method of this. Also, another [44:43] thing coming right around the corner is [44:44] AI browsers. [44:46] >> So, the whole web browsing experience is [44:48] gonna [44:50] >> it's [snorts] going to change [44:51] significantly. [44:52] >> Chrome's going to [44:54] >> Yeah. Yeah. Perplexity's got one out the [44:56] opening's got. So yeah, that's coming [44:58] around the corner too. Um, but the other [45:00] thing to keep an eye on, two other [45:02] technologies with that regard are LM [45:05] LLMs.txt [45:09] LLM. [45:10] >> LLMs.ext. [45:11] >> Oh, what is that? [45:12] >> Um, it's a proposed uh protocol for [45:16] basically the equivalent of robots.ext, [45:18] but for AI crawlers. [45:21] >> Why they need a different one doesn't [45:23] really make too much sense to me. Um, [45:25] it's not been formally adopted by any [45:28] major one that I know of right now. So, [45:31] as far as as far as I'm aware, like some [45:34] websites have putting up these LLMS.ext [45:37] files, but no main company clawed or [45:40] open. [45:41] >> They might have different site maps on [45:43] there, [45:44] >> I guess. Maybe. [45:45] >> Yeah, maybe. But why couldn't you do [45:46] that in robots text? I don't know. My [45:49] prediction would be, this is just a wild [45:51] guess, but my prediction would be it's [45:53] just going to come right back around to [45:54] robots.ext getting its use case [45:56] expanded. [45:56] >> She's answering my question, [45:57] >> but or llms.ext [46:01] becomes a thing. Um, [46:03] >> but it's not not formally supported as [46:06] it stands right now. [46:07] >> Um, so the other technology to keep an [46:09] eye out for is what Cloudflare is doing. [46:13] Um, they I actually got really angry [46:15] when they first did this or they said [46:17] they did it and then they didn't or [46:19] whatever. Anyway, if you use Cloudflare [46:21] as a CDN, which we do because PageB for [46:24] back to traditional SEO is a strong, you [46:26] know, technical SEO factor. Um, and CDNs [46:28] can help with that. So, we have pretty [46:30] much all of our clients set up on [46:31] Cloudflare because it's free and it [46:33] works well. Well, they just decided that [46:35] we're going to just block all AI bots [46:37] from from all of our clients sites, [46:41] right? like I was like, "Wait a minute, [46:42] what?" Like, "I don't want that." I [46:45] panicked. And so, thankfully, they [46:47] either reversed it immediately or they [46:49] didn't go through with it after that [46:50] announcement. But that announcement made [46:52] my heart skip several weeks because I [46:54] was like, "Oh my god." Like, they don't [46:56] have the right to decide. I'm the one as [46:58] the website owner has the right to [46:59] decide if I want AI bots to be allowed [47:02] to crawl on my content and serve it up [47:05] or and or be trained on it. That should [47:07] be up to me. Thankfully, it looks like [47:08] they didn't um force it to be [47:12] implemented for all of their people, but [47:14] they they did put a control in there. [47:17] So, you can go into Cloudflare and say [47:20] block all known AI bots from my site. [47:24] Problem is right now, it's like you [47:25] can't specify for what, right? [47:27] >> Like I would like to be able to say, [47:29] "Yeah, I have some gated content. I have [47:30] some, you know, paid downloads or [47:32] whatever and I want to block those types [47:35] of files from AI discoverability and [47:38] training. And then the rest of it maybe [47:40] I want to block from training but not [47:42] from discoverability. That that level of [47:44] granularity is not there yet. But [47:46] they're working with some of the real [47:47] big publishers like New York Times I [47:49] think um but of that level on uh premium [47:53] uh pricing plans where you can say yes I [47:55] want to allow the AI chat bots to um [47:58] crawl and train on my data but they have [48:01] to pay me a royalty or licensing fee of [48:05] x amount per word token. I don't know [48:08] the technologies not fully developed yet [48:10] but that's something to keep an eye on [48:12] as well. [48:14] [laughter] [48:15] No, it's a lot. [48:16] >> It's a lot. [48:17] >> It's a lot. Yeah. [48:18] >> So, I guess I touched upon your [48:20] question. [48:22] >> Um, yeah. So, first of all, thank you [48:24] for such a comprehensive presentation, [48:27] you know, because when you were talking, [48:29] I thought, oh, also she didn't mention [48:31] table of the contents and you did. So, [48:35] thank you. Um, so I have a question [48:38] about you didn't mention chunking, [48:40] right? that we when we write a blog, a [48:44] post, we should focus on chunking and [48:48] presenting information and bullet points [48:50] and etc. Right? But also from [48:53] traditional point of view and co and [48:56] experience that you mention right we [49:00] know that to make the post more readable [49:03] it should read as a text right it should [49:06] read as a post it should read as a [49:08] because it's I think for experience you [49:12] don't want to read bullet points right [49:15] that's a good point [49:16] >> so this is a conflict like okay so when [49:19] you write a pause and I know the answer [49:22] in the way that like okay so you make it [49:24] readable and maybe you make it a middle [49:28] ground right doing some bullet points [49:31] and all of that but I just wanted to ask [49:34] you what what do you think about that [49:36] when you work with the clients or [49:38] businesses when they ask you okay how do [49:41] we write a post like do we make it more [49:43] readable and more like okay more [49:44] thousand word more than thousand words [49:47] right that was the old rule so is it the [49:50] same old rule now because we still [49:53] notice as a business that our longer [49:55] posts let's say they still get some [49:58] traction because again that increases [50:01] the session per website right [50:04] that's reading that's experience so [50:07] what's your point on that and do you [50:10] have this like a little bit of a [50:11] conflicting [50:13] >> Absolutely definitely I realized like [50:16] the um the almost hypocrisy in the in [50:19] the presentation that I just gave and [50:21] that I was like, you can't put out bland [50:23] AI generated content that looks like [50:24] every other bland AI generated article, [50:26] but make sure it looks like a bland AI [50:28] generated article. You you have to write [50:30] it, but make it look like that. [50:32] [laughter] [50:34] I see the the irony there for for sure. [50:38] Um, yeah, you want to publish stuff that [50:39] doesn't look like an AI generated answer [50:43] for thought leadership purposes and [50:45] really telling in-depth real world [50:48] experience, word count, etc. [50:50] But at the same time, we have to work [50:53] with the way that they work, which is in [50:56] chunks. So, you hit upon it already. The [50:58] happy medium, if there's a happy medium [51:00] in there somewhere, is what I would [51:01] recommend. Obviously, I chose examples [51:03] for the screenshots here where I really [51:05] wrote really granularly in in some of [51:07] those articles. I don't always write [51:09] that way completely that granularly, you [51:12] know, I try to mix it up and make it [51:15] look, you know, not written by not [51:18] written by or for AI, but written for [51:20] AI. [laughter] [51:22] So, yeah, it's it's definitely a [51:24] balancing act. I hope that it gets [51:26] better. Um, you know, as it stands right [51:28] now, the LLM chunking is very kind of [51:30] rudimentary and uh we do have to chunk [51:34] it down like really granularly. I'm [51:35] hoping that that'll evolve and we have [51:37] to do it quite so literally anymore. Um, [51:40] and can can bring back some along. I [51:42] know I've had the debates too with like [51:44] a copyrighter friend of mine. She's like [51:45] going off her rocker right now. She's [51:47] like a camera like this. [laughter] [51:49] I know. I get it. But you got to like [51:51] find some happy medium. [51:54] >> Is there a way to get the best of both [51:55] worlds where like on the screen you see [51:57] something that's for human readable but [51:59] some sort of metadata that AI can pick [52:01] up on? [52:02] >> I wish [laughter] [52:03] that would be awesome. [52:05] >> I really Well, maybe, maybe. I mean, [52:09] >> try it out. See if it works. [52:10] >> I with with AI browsers around the [52:13] corner, I can envision something like [52:15] that. Like the machine version of the [52:17] content versus the human version or [52:21] >> Hey AI, this part's for you. [52:23] >> Exactly. I would love to I would love [52:25] for that to become a thing. Maybe that [52:27] becomes part of the use case for [52:29] >> Yeah. Do like the job twice or But you [52:31] know, whatever. [52:31] >> I know, right? Yeah. As if we don't have [52:33] to [52:33] >> we don't have enough to do. [laughter] [52:36] I did pick another question. [52:37] >> Oh yeah, [52:38] >> thank you. [52:38] >> Yeah. [52:39] >> Um, how about measurability? How do you [52:42] know how well your site is doing on AI [52:45] search? [52:46] >> Ah, excellent question. Um, [52:48] >> that's a good question. [52:49] >> Yes. [laughter] [52:51] Well, so it's really difficult within [52:55] the I guess I'll call them like kind of [52:56] the walled off chat bots like the [52:58] separate chat bots like um chattyp.com, [53:02] gemini.google.com, [53:04] um, copilot microsoft.com. Really hard [53:07] to have any visibility within there. [53:09] There's all sorts of tools popping up [53:10] that say that they monitor your brand [53:12] mentions, but some of them are like old [53:14] school keyword ranker style, like you [53:16] have to put the phrase in that you want [53:18] to track. And who can imagine all the [53:21] different ways that someone might ask a [53:23] chatbot something. So, um, there's [53:26] there's some brand tracking tools coming [53:28] out though where you can at least put [53:29] your brand name in and say monitor how [53:31] many times you see my brand name come up [53:34] in these chat bots, which I still doubt [53:36] the efficacy of because they don't let [53:40] you crawl the entirety of the private [53:42] any of private chats. So, it's really [53:45] just got to be these tools are just [53:47] doing test chats and reporting to you [53:50] the results of of their test chats, what [53:52] they're testing on exactly. Some of them [53:54] you can control, some of them you can't. [53:56] >> Unfortunately, you know that there's [53:58] very little visibility into those. What [54:00] we can see is in um Google and Bing, the [54:03] AI overviews at the top, the traditional [54:06] rank track, most of the traditional SEO [54:09] rank tracking, keyword ranking tools [54:11] have evolved to be able to report to you [54:13] on what key phrases you are in the AI [54:16] overviews at the top for. And chances [54:18] are if you're in that AI overview at the [54:20] top of Google or Bing, you're probably [54:22] going to make it into chat GPT's [54:24] response, call it response, etc. So [54:27] that's what we're using as our main KPI [54:29] is the appearance of um keywords that [54:33] they're get our clients are getting [54:35] found for in AI overviews. Particularly [54:37] the ones though that we're looking for [54:39] are the lowhanging fruit opportunities [54:40] where they have a keyword ranking on the [54:42] first page or two, but not yet in the AI [54:44] overview. We track that as well. So the [54:47] AI overview exists. Well, first of all, [54:49] it exists. So not all keywords, a lot of [54:52] them do, but not all keywords at this [54:53] moment trigger an AI overview. So A, [54:56] does it have one? B, are we in it? And [54:59] so we developed a looker studio report [55:01] for that to highlight or literally like [55:03] flag them with the word opportunity. Um, [55:06] when there is an AI overview, we are on [55:08] the first page or two for of regular [55:11] results, but not in the AI overview. [55:13] That's usually a real lowhanging fruit [55:15] like quick tweak you can make to the [55:18] content to satisfy what the LLM wants to [55:21] say and reverse engineering giving it [55:23] its confirmation bias um with content [55:26] that's already ranking well so we're [55:28] using you know number of AI overview [55:31] rankings um and we're also we developed [55:34] like a percentage of visibility so like [55:37] the proportion of ones that exist that [55:39] were in versus not in that sense [55:45] I forgot to touch upon word count before [55:47] too. You mentioned word count. I wanted [55:48] to hit on that. Even before all of this [55:50] AI powered search, LLM search stuff. Um [55:54] the AI AI there was there's always been [55:57] AI in search in the in the form of [55:59] machine learning, right? Like the [56:01] backend kind of stuff, right? So it's [56:03] this more front-end like generative AI [56:06] chatbot type functionality that's new. [56:08] But the AI that's kind of baked into the [56:10] search engine algorithms that has been [56:12] there from for years from the machine [56:13] learning perspective thankfully even [56:15] before this over the last handful of [56:17] years has gotten better at determining [56:19] what's appropriate content length given [56:22] the topic. So we do still say just for [56:26] simplicity sake like a thousand words or [56:28] more,500 [56:29] or more but really you know when we're [56:31] really looking at an opportunity we're [56:34] looking at like what content is already [56:36] ranking for that topic. How long are [56:38] those top 10 articles? That's the [56:41] average word count that we need to go to [56:43] for that topic because certain complex [56:46] topics warrant a lot of words. Other [56:48] simpler ones don't. So thankfully that's [56:50] gotten like a bit dynamic. It's gotten [56:52] smarter at that. [56:57] >> Take one more question. We're getting [56:59] close to them kicking us out. So [57:02] >> we're getting close [57:05] out. [57:06] >> Um With the importance of experiential [57:10] content and the fragmenting of sort of [57:13] the formatting, [57:15] how is this affecting what was in the [57:18] past a high priority which is keywords [57:21] and keyword density in particular? [57:24] That's also a really good and [57:26] everchanging um point that that the the [57:31] AI the machine learning AI I think was [57:33] getting better at even before all of [57:36] this. Um but so let me let me rewind [57:39] then as far as keyword density what [57:41] we're talking about is like the number [57:43] of times you use your target keyword in [57:46] an article and your target secondary [57:48] keywords like your your semantic [57:50] supporting words. um largely as far as [57:54] like the primary phrases go that we're [57:57] working into the title and is the main [57:59] theme of the article that really hasn't [58:02] changed. We're still saying like one [58:04] time per 100 words, not to exceed two [58:07] times per 100 words so that you're not [58:09] stuffing. It's just kind of like you [58:11] have to say it enough like to make it [58:14] clear that that's what the topic is [58:16] about without being repetitive. from a [58:18] primary keyword perspective and from the [58:21] perspective of a one article still being [58:23] one thing in the eyes of traditional [58:25] search engines, nothing really has [58:27] changed. Um, other than like with the [58:31] secondary keywords, I think the AI [58:33] machine learning AI into the baked into [58:36] the back end of the search algorithm has [58:38] gotten better at understanding context. [58:40] So you don't have to be as literal with [58:42] like every single secondary phrase that [58:45] might be related to the topic. Like I [58:48] think it's it's gotten a little smarter [58:49] in that sense, but the chunking is [58:51] really the big difference. That's where [58:52] like it's almost like each section, each [58:56] building block of the article should [58:57] have its own primary keyword. We're [58:59] still calling them secondary keywords [59:00] because we don't want to confuse people [59:02] with like putting telling them to put [59:03] them in the title and all throughout the [59:05] article, but you can kind of think of it [59:08] as having a target keyword for each [59:10] chunk of the article. So, in that sense, [59:12] I'd say it's it's changed and it's still [59:16] changing. [laughter] [59:18] It's still evolving, I should say. I [59:19] just said in the beginning that I [59:20] wouldn't say that SEO changes, but [59:22] instead it changes. It evolves, gets [59:23] more complex. [59:26] >> I just five minutes. [laughter] [59:28] All right. [59:29] >> Thank you so much. [59:30] >> Thank you guys so much for having me. [59:31] [applause]