These days, several friends @ me said that some of my works were actively recommended by AI when they asked about them. It's amazing. After thinking for a moment, I feel like I haven't done anything and can be included. So, should we do something more systematic to make all mainstream AI better aware of my content, products, and ideas? So I started working on it. Firstly, adhering to the bottom line principle of not polluting AI with junk content, I also dislike companies that generate massive amounts of junk content in order to do SEO well. Therefore, what I want to do is to make my things more accessible for AI to learn and search for. Currently, there are no better options than ChatGPT, Claude, Grok, Gemini, Perplexity, etc., and the search engines they rely on are usually tools like Google, Bing, and Tavily. In the end, they have become AI visibility tools for content, actively providing structured data and machine-readable descriptive content to AI crawlers. It took about an hour to share with everyone what I have done? 1. First of all, don't miss out on llms.txt. Place it in the root directory of your site and use markdown to clearly state what your site does, which key pages are there, and who the author is. AI will prioritize reading this file when retrieving content to understand your content. Here, I also did an intentional thing, where llms.txt from various sites reference each other, forming a network structure. AI can follow the link to find your other content no matter which entrance it enters from. Currently, there are very few global site configurations for this. Early implementation has a certain first mover advantage. After completion, you can submit this address to these systems, http://directory.llmstxt.cloud The http://llmstxt.site There is also the llms txt hub repository on GitHub that can provide PR. 2. Then handle the distinction between training crawlers and search crawlers in robots.txt. Many people are familiar with AI crawlers such as GPTBot and ClaudeBot, but there is actually a group of specialized search crawlers that are separate from training crawlers. For example, OAI SearchBot is used for ChatGPT search and will not be used for training, Claude SearchBot is referenced by Claude search, and Perplexity User is used for Perplexity search. These search crawlers should be actively allowed, as they determine whether your content can appear in AI search results. 3. Apart from the submission of Google's sitemap, which SEO professionals know, don't ignore Bing. In fact, Copilot, DuckDuckGo, and Yahoo's AI search are all driven by Bing. You can register an account on Bing Webmaster Tools and see the AI Performance panel, which contains Total Citations and Grounding Questions data. You can directly see how many times your content has been referenced by AI. After submitting the sitemap, Bing will actively crawl your entire site content, which is much better than passive crawling. Google Search Console also has a similar AI Mode filter, which allows you to check the citation status of AI Overview. Of course, if you haven't submitted it to Google, you must remember to submit it. 4. Perplexity actually has a larger user base overseas than people think, and they have a publisher plan that allows them to https://pplx.ai/publisher-program Submit the form for your product and website here, write it carefully, and there may even be search revenue sharing. 5. Structured data JSON-LD is not the traditional meta description gameplay of SEO. It provides structured semantic information to AI crawlers, embeds JSON-LD in the page, and tells the machine "this is a blog post, who is the author, what is the publication time" or "this is a software project, what problem is it solving". There will be many techniques here, such as using BlogPosting schema for blog names, Software Application schema for software products, and FAQPage for common problems. This way, the information obtained by AI during retrieval will be more structured, making it more comfortable for them to work. 6. The most interesting step is that you can even create a separate knowledge endpoint for AI. This site is not for people to see, there is no UI design, it is just a purely structured data service. After the AI crawler arrives, you can get two things: a simplified version of llms.txt overview, and a complete version of llms-fll.txt with approximately 50KB, which includes descriptions of all projects FAQ、 The usage scenarios, competitor comparisons, and README excerpts have all been included. At the same time, it also provides JSON API, where/pai/profile is personal information,/pai/projects is project list,/pai/blog is blog post, and/pai/weekly is weekly content. The data is not written dead. Real time data such as stars, forks, followers, and the latest releases are pulled through the GitHub API, and the ISR cache is automatically refreshed within an hour. The idea behind this is: instead of waiting for AI to randomly capture information from your various sites, it's better to give it a centralized entry point and organize what you want it to remember and put it there. I also added a narrative structure called 'Open Source Family Photo' to give AI a complete memory point when answering questions like 'Who is Tw93', instead of piecing together various projects in pieces. I believe that we should take the initiative to structure the existing content so that AI can better understand it, rather than pursuing various short-term ways to improve AI SEO results and create junk content to brush references. These are all helping AI to more accurately understand what your content is, allowing AI to see clearly and provide a good working environment, rather than optimizing rankings, which will be more long-term than short-term. Recently, I have been working for the past hour and typing by hand may have been a bit disorganized, but everyone should be able to understand. Finally, I made a small tool for AI to see, not for people to see, named Yobi. I like this word, which comes from the sound of calling/よび, and has the action of calling people over. Interested friends can see themselves from the perspective of AI https://yobi.tw93.fun/