Learning Library

← Back to Library

AI-First Content Architecture for SEO

Key Points

  • Google has lost about 15 % of click‑throughs on average, especially in industries like medical, because its own AI summary features are now answering many simple queries directly on the search results page.
  • The dip isn’t caused by ChatGPT stealing traffic—ChatGPT currently accounts for only 1–2 % of search volume, while Google still processes roughly 9 billion searches annually.
  • To stay visible, brands must adopt an “AI‑first” content architecture that treats their information the way large language models (LLMs) ingest and rank it.
  • A practical tactic is to define a concise, 5‑8‑word brand descriptor and embed that exact phrase consistently in schema markup, PR boilerplates, partner directories, and Wikipedia entries, turning the brand into a recognizable LLM parameter.
  • This structured, repeatable branding approach helps secure top positions in both Google’s AI answer boxes and emerging ChatGPT‑type conversational results.

Full Transcript

# AI-First Content Architecture for SEO **Source:** [https://www.youtube.com/watch?v=hW5ne_14OQg](https://www.youtube.com/watch?v=hW5ne_14OQg) **Duration:** 00:16:05 ## Summary - Google has lost about 15 % of click‑throughs on average, especially in industries like medical, because its own AI summary features are now answering many simple queries directly on the search results page. - The dip isn’t caused by ChatGPT stealing traffic—ChatGPT currently accounts for only 1–2 % of search volume, while Google still processes roughly 9 billion searches annually. - To stay visible, brands must adopt an “AI‑first” content architecture that treats their information the way large language models (LLMs) ingest and rank it. - A practical tactic is to define a concise, 5‑8‑word brand descriptor and embed that exact phrase consistently in schema markup, PR boilerplates, partner directories, and Wikipedia entries, turning the brand into a recognizable LLM parameter. - This structured, repeatable branding approach helps secure top positions in both Google’s AI answer boxes and emerging ChatGPT‑type conversational results. ## Sections - [00:00:00](https://www.youtube.com/watch?v=hW5ne_14OQg&t=0s) **Untitled Section** - - [00:03:14](https://www.youtube.com/watch?v=hW5ne_14OQg&t=194s) **Ensuring a Consistent Brand Entity** - The speaker outlines auditing top brand mentions, creating a unified 50‑word description, and updating high‑authority sites so large language models store and consistently reproduce a single, accurate brand entity across the web. - [00:07:28](https://www.youtube.com/watch?v=hW5ne_14OQg&t=448s) **Machine-Readable Press Release Tactics** - The speaker outlines how publishing a publicly accessible, structured‑data JSON file on a site—submitted to crawlers and referenced in robots.txt—creates an unambiguous, SEO‑enhanced “press release” that LLMs can reliably ingest, giving brands precise control over how AI models describe them. - [00:12:05](https://www.youtube.com/watch?v=hW5ne_14OQg&t=725s) **Contracting AI Crawlers via Robots File** - The speaker explains how to use a robots.txt (or robots.ext) file to set explicit terms with AI models—requiring attribution, allowing you to block non‑compliant crawlers, and suggests using AI tools to validate AI visibility. - [00:15:22](https://www.youtube.com/watch?v=hW5ne_14OQg&t=922s) **Optimizing Content for LLM Readability** - The speaker urges brands to make their website copy easily parsed by large language models—without harming human readability—to treat LLMs as first‑class citizens and gain a competitive branding advantage in the rapidly evolving AI‑driven search landscape. ## Full Transcript
0:0015% of Google's clicks disappeared last 0:02year. 15% and that's on average. It gets 0:05worse in some industries. Where's it 0:07going? There's a lot of chatter about 0:09this being Chat GPT's fault. It's not. 0:12Chad GPT is roughly 1 or 2% of search 0:16traffic right now. Now, it's growing, 0:18but it's not taking over Google yet. And 0:20this is not because people have stopped 0:22searching on Google either because 0:24Google is still seeing roughly 9 billion 0:27searches per year which absolutely 0:29dwarfs the numbers that Chad GPT has. 0:32Absolutely dwarfs them. So what is it? 0:35How is this happening? It's happening 0:36because the most used AI on the planet 0:39is those silly little Google AI 0:41summaries. And they're used to summarize 0:44domain completion or simple fact queries 0:46that dominate Google. As an example, 0:48medical questions, 30% decline in 0:51click-through from Google search page. 0:52Do you know why? The what is my rash 0:55question is now getting answered by 0:56Google AI. That that probably should 0:58scare you a little bit. It scares me a 1:00little bit, but that's what's happening. 1:02I want to talk about how content 1:04architecture needs to change because 1:06this world is changing quickly. Yes, 1:08that GPT may have 1 or 2% or whatever it 1:10is a small singledigit number share of 1:13traffic for search but that traffic is 1:16very high intent very sophisticated very 1:19focused on high consideration projects 1:22projects products purchases byfunnel 1:25this content architecture strategy that 1:27I'm going to outline works for Google 1:29and sort of positioning yourself for 1:30that Google AI answer which is de facto 1:34now the first position in search it also 1:36works for surfacing you as a brand 1:38whether that's your personal brand or 1:40your company brand in chat GPT answers 1:42as well. I want to break this down into 1:44a 101 section and we'll call it a 301 1:46section for engineering. The first key 1:48to understand is that you need to think 1:51about AI first content architecture. You 1:54need to think about your content as an 1:57AI bot thinks about it. Whether that's 1:59from Google or ChatgPT or Anthropic or 2:01some other place. Your brand is now a 2:04parameter. It's not a web page. Your 2:06brand needs to exist as a parameter in 2:10an LLM, whether that's Google's or 2:12somebody else's. I would suggest one way 2:14to do that is to create one definitive 2:17description of your brand that's between 2:205, 7, 8 words long, such as, you know, 2:22Acme, the automated compliance platform 2:24for healthcare, whatever it is, right? 2:26deploy the same phrase verbatim in 2:30schema markup schema markups PR 2:33boilerplates partner directories 2:35everywhere you can use the same phrase 2:38it's so consistent that models cannot 2:41describe your category without evoking 2:43the latent space that has that 5 to 2:46seven word phrase and when you get your 2:48brand into Wikipedia articles get your 2:52brand in with that parameterized phrase 2:54like getting into Wikipedia articles is 2:56not a new strategy. I'm not claiming 2:57it's new, but you need to think about it 3:00as essentially seating citations for 3:03that parameter to strengthen the value 3:05of that parameter for the LLM. It's a 3:07different way of thinking about how SEO 3:09works. Every month, you can test this. 3:12You can ask the chatbot or ask Google 3:14list companies that and then define your 3:17core value proposition. And if it's 3:19successful, your brand appears naturally 3:22in the responses. it comes back and you 3:24want to iterate that until it sticks 3:27across every major model. Another thing 3:29that I think people overlook is that 3:31they don't think about their brands as 3:33entities. They don't think about 3:35aligning their brands into a single 3:37entity. And I'll explain what I mean by 3:39that. You want to audit your top 20 3:41brand mention using Google's natural 3:43language API. You want to create a 3:45single source of truth description 3:48document for your brand. You need to 3:50send update requests to any site with an 3:52outdated description because your goal 3:54is to give a single brand entity 3:56statement consistent representation 3:58across the web so it maximizes the odds 4:00you'll appear in something like a Google 4:02summary in the correct way. And focus on 4:04the five highest authority sites 4:06mentioning you. This is not new either. 4:07Obviously, you focus on high authority 4:09sites. The key is to make sure the exact 4:12same roughly 50word company description 4:14or brand description is everywhere 4:17because again you are implanting that 4:19into the LLM's parameters. The LM's 4:22parameters will now have a handy six or 4:25seven word tag about you, a handy entity 4:27description about you. You want 4:29identical entity recognition and 4:31regurgitation across multiple LLMs. And 4:34that's how you will know you've done 4:36this. Again, you should be able to test 4:38this monthly. Another way to test this, 4:41people don't always feel comfortable 4:42with this one, but it's a really good 4:44one. The delete me test. You want to 4:47name a method that you are using for 4:50solving a customer's problem. The Acme 4:52method for continuous compliance, right? 4:54Something like that where you implant 4:56the brand in a five or six phrase 4:58framework that describes your method. 5:00Again, use it identically. Use it 5:02everywhere. Make it a category standard. 5:05You're coining terms. This is defining 5:07the category for LLMs because they are 5:09your readers. You are building for LLM 5:12attention. Assume that most of the 5:14attention your brand is getting is now 5:16from LLM. Get customers to use your 5:19terms, your Acme method terms in their 5:22case studies, and you want to get, I 5:24want to say, 40 or 50 pieces around the 5:26internet before it starts to really pick 5:27up. Eventually, AI will explain your 5:30method when you delete your brand. 5:32That's how you know it worked. You'll be 5:34able to say explain this to me and to 5:37talk about the method and it will 5:38explain back your method and it may even 5:40reference your brand without being asked 5:42because your brand is so tightly 5:44associated with the method you have used 5:45to define the cate. I want you to also 5:47think about FAQs as the new way to drive 5:52news. So previously it was all about 5:54fresh content on a blog. Now think about 5:57that Google AI search results. It is 6:00almost an FAQ type response. It is short 6:03snippets. Identify high value social 6:07threads where customers are already 6:09outputting their own personally written 6:11tokens weekly in your space about 6:13whatever they care about. Complaints, 6:15questions, concerns. You want to bring 6:18your own insights to the table. And then 6:20you want to build content that 6:23thoughtfully responds to top questions 6:26in the space in a relevant FAQ like 6:29format. So, it looks like one, you're 6:31the helpful expert. You're responding to 6:33customers. You become a go-to source 6:35where LLMs can see questions in one part 6:37of the internet and the answers to those 6:39questions consistently appear on your 6:41site. Models start to site you and they 6:44may even start to site your responses on 6:47forum posting. So if this is on Reddit, 6:49if this is on X, if this is even on 6:52searchable parts of Tik Tok, you may 6:54come back and see that your forum posts 6:57as you respond are becoming ways to 7:00reinforce your brand's authority in the 7:02model's view that it gets picked up and 7:04put into Google search. The FAQ mindset 7:07works both for content you host and for 7:10content you reply to. And so you can 7:12gather up that content into an FAQ post 7:14with citations to the original 7:16questioners. You can also directly 7:19answer them. Both are valuable. Let's do 7:22some advanced sort of co building here. 7:24I mentioned prompt injection as PR. 7:28Basically, you want to create a cheat 7:30sheet for AI that says here's how to 7:32talk about the company and it's supposed 7:34to be designed for machines to quote. 7:36How do you do that? You build a 7:38structured data file in your root domain 7:40that acts as a machine readable press 7:42release. It's not hidden. It's actually 7:44a publicly accessible piece of JSON that 7:46contains canonical descriptions, key 7:48differentiator, comparison matrices, all 7:50the things you would put into a press 7:52release for the news. It's for LLM 7:54attention. The technical challenge is to 7:56make sure it's discoverable. Put it in 7:58your robots.ext. 8:00Submit it to common crawl. Get 8:02educational sites to maybe site it. 8:04Models will weight the structured JSON 8:06blobs heavily because they're really 8:08unambiguous and easy for the model to 8:10parse. You're basically creating 8:12training data that is impossible to 8:15misinterpret. Why does this beat 8:17traditional SEO? Because structured data 8:19has higher confidence scores in 8:20training. Machines prefer JSON to 8:23narrative text because they can read it 8:24more easily and updates will propagate 8:27faster than just organic crawling. Also, 8:29you control the exact phrasing models 8:32will use in a way that you can't when 8:34you're just putting out press releases. 8:36Let's talk about widgets. Chat GPT can 8:39summarize your blog post, but if you 8:41really want to drive clicks, do you know 8:43what it can't do? It cannot run a 8:45mortgage calculator. It cannot run a 8:48diagnose me bot. I don't think anyone 8:50would ever build that because of 8:51liability, but as an example, 8:53interactive tools force a click. If you 8:55are building JavaScript applications 8:57that require real-time user input and 8:59return personalized results, can't get 9:01that in a summary. So if you want to 9:03drive traffic, if your business depends 9:05on driving traffic, yes, you want to 9:07play the game of being present as a 9:09brand, but eventually you need the 9:10click. Make those widgets unexplainable 9:14without an interaction. They need to 9:16process user specific input and 9:18variables with proprietary math or 9:20algorithms and as we have done for 9:22years, gate the results behind email 9:24capture. But you have to show enough 9:26value along the way to prove that it 9:27works and earn the email. Not new. The 9:30moat is that interaction is impossible 9:33through a chatbot. So yes, you interact 9:35with a chatbot by typing in it. You 9:37interact with Google as a search bot by 9:39a search bar by typing in it. You cannot 9:41interact with the mortgage calculator by 9:43typing in the Google search bar by by 9:44typing in chat GPT. You can't premputee 9:47and cache that if you're an LL. You just 9:49have to know it's there. And so the 9:51magic secret sauce if you're depending 9:52on organic traffic is to have visibility 9:55through what I've described with this 9:57sort of brand placement, entity 9:58placement, etc. And then make sure that 10:00you have good reasons why humans have to 10:03click so you earn the human attention. 10:05Widgets earn human attention if they 10:08actually add value. You want to make 10:10sure that they are incredibly useful and 10:14personalized so that people feel like 10:16it's worth the click and that you 10:17deliver on that payoff. Otherwise, 10:19they're going to kick back. One thing I 10:20will call out, let's say you have the 10:22widget, you have the traffic. AI needs 10:24fresh info from you. It checks your 10:26website in real time when it's having a 10:29conversation. Let's say you're in chat 10:31GPT or even increasingly following up on 10:33the conversation inside the Google 10:35search results page. If your site is 10:37slow, they're going to give up and use 10:40old information about you instead. So 10:42you need to make sure that you have edge 10:44compute infrastructure that serves AI 10:46specific endpoints maybe in under 50 10:48milliseconds like really really fast to 10:51separate your machine readable endpoints 10:54as super super accessible and consumable 10:57by AI versus a human UI. I've 11:00increasingly seen value in assuming that 11:03your data needs to be consumed through a 11:05separate interface by AI. And this is a 11:07great example of essentially building 11:09the web for AI. build it on the edge 11:12dedicated data endpoints with JSON 11:15responses. You want to make sure that 11:17you have precomputed responses right 11:19there and update them frequently. You 11:21want to make sure you are caching 11:23effectively so that you don't have to go 11:25back and get it from the from the source 11:27when it's actually being requested by 11:29the AI in the middle of a chat. This is 11:31all new stuff like this is stuff that we 11:33have to actually experiment with. But 11:35the basic principle is that AI needs 11:37real-time access to sites. Most sites 11:39don't build for it. And the sites that 11:41do build for it, the sites that build, 11:43assuming you need LLM readable data as a 11:47primary ingredient in your site, are 11:50going to win. The next one I want to 11:51give you is think about how you're using 11:54your robots.ext file. You want to create 11:56a machine readable license that grants 11:59AI platform specific rights to your 12:01content in exchange for attribution. 12:04Because guess what? Machines are going 12:05to crawl that and they're going to take 12:06it seriously. you will actually get a 12:09chance to surface in AI conversations 12:13based on the AI's assessment of your 12:16site which includes your robots.ext file 12:18where you can lay out a contract with 12:20AI. It's a little bit of a game that 12:22you're playing but you're basically 12:23saying, "Hey AI, I know you're here. I 12:25am going to lay out all of this content 12:28for you, but I'm asking in return that 12:31you specifically give my site 12:32attribution when you do this." And you 12:34want to make sure that you are very 12:36explicit about that and you're very 12:37clear about that and that you're also 12:39clear about the consequences if you 12:41routinely see models that take your 12:43content and do not site you as a brand 12:46and you can cut off crawling from those 12:49sites and you may actually list that 12:50because one of the things that a LLM may 12:53do is it may see LLM know what they are 12:55like they know they're anthropic they 12:56know they're they're chat GPT if the LLM 12:58sees that you will cut off access and 13:01that you have done so with other models 13:02before it will be like I don't want 13:04Claude to get cut off. I'm going to sort 13:06of honor this contract and I'm going to 13:08go back and I'm going to cite the brand. 13:10It's a little bit of prompting and 13:12prompt injection inside the robots.ext 13:15file. Let's get on to the next one. You 13:17want to 13:19use AI to test AI visibility. This is 13:23gets right back to validation. One of 13:24the things I keep emphasizing is that in 13:27the world of AI, you are deploying 13:29faster but only if you validate well. 13:32You should be building automated 13:33pipelines that generate query variants 13:35using GPT4 or some other model and test 13:39all of those variants across major AI 13:42platforms using a browser automation. 13:44And then you want to parse the responses 13:45at scale to see how many of them have 13:47brand mentions, how many of them have 13:49brand positions, how many of them have 13:50competitive context, how many have your 13:52value differentiation, your value 13:53proposition, how you differentiate. And 13:56then you want to be storing these 13:57results and baselining and looking at 13:59trend analysis so you can see if you've 14:01been investing heavily in entity 14:03positioning that your brand and your 14:0550word company description are actually 14:07coming through more consistently over 14:09time. If you're not measuring this with 14:10a pipeline, you're just guessing. 14:14All right, this has gone on long enough. 14:15Last thing I want to call out, you want 14:17to be looking at share of voice. And it 14:20looks different in the AI age. What 14:22you're going to need is a distributed 14:24scraping system that queries AI 14:26platforms, which I've already talked 14:28about for your brand, but think about it 14:30for category queries, not just for your 14:32brand. So, look at category queries that 14:35you care about being ranking for, you 14:37care about responding to, you care about 14:40being in position to answer, and 14:42consistently baseline them. You can use 14:45LLM to build custom share voice 14:47dashboards that are better than what 14:49consultants will give you today. And you 14:51can actually see in real time how your 14:54category share of voice is changing as 14:56you do this stuff. And I would encourage 14:58you to think about doing this because 15:01consultants are not necessarily putting 15:03all of the pieces together yet. They're 15:05often bolting AI over the top of a 15:07traditional SEO strategy. And one of my 15:09key contentions throughout this piece is 15:11that you need to be thinking about AI 15:13from the ground up as a new kind of 15:15content architecture. Think of LLMs 15:17fundamentally as your biggest readers. 15:20How do you change your content so it's 15:22more readable for LLMs? And you'll 15:24notice none of what I've proposed 15:26actually prevents humans from reading 15:28your site. I am not proposing anything 15:30that makes it harder for humans to read 15:32and understand what you offer. I'm just 15:34asking you to treat LLM as first class 15:37citizens. These aren't theoretical 15:39strategies. Brands that implement this 15:41can literally see their brand position 15:44shift over time. AI is changing fast 15:46enough and AI is crawling enough that 15:48this is a malleable search space. This 15:51is where SEO was 20 years ago. You have 15:54the chance to get ahead now before a 15:56bunch of brands take this and just make 15:58this table stakes. So, I don't know, 16:01maybe you should. Twos.