SEO is dead (again)? Why I optimize my SaaS for LLM scraping, not Google clicks
If your main KPI is clicks in Google Search Console, you are optimizing for the past. Recently Google rolled out analytics for AI Overviews and the reality is exactly what we expected. CTR of the first position drops up to 60% if there is an AI answer hanging above it.
Users get the solution right in the search results without clicking anything. Fighting for clicks is getting too expensive and inefficient. I stopped chasing clicks and switched to Citation Rate (how often my product is mentioned) inside neural network answers.
The shift from blue links to Share of Voice
The game used to be about finding keywords and writing SEO texts just to get top rankings. Now the goal is totally different.
When a user types into ChatGPT, Perplexity or Gemini "recommend the best tool for X", the AI needs to pull from its database and recommend your product. This is GEO (Generative Engine Optimization). You are not competing for a spot in search results anymore. You are competing for the share of voice in LLM outputs.
What I am actually implementing in the code
Adapting for AI agents requires architectural changes, not just rewriting text. Here is what I do in practice.
1. JSON-LD instead of text walls LLMs rely on knowledge graphs. Its really hard and resource heavy for them to parse abstract walls of text on landing pages. I shifted my focus entirely to microdata. Instead of basic HTML tags I use dense structures with FAQPage, Product and strict Entity extraction. When an AI agent or crawler visits the page, it should grab ready facts in key-value format from JSON in milliseconds, rather than trying to guess the meaning of some marketing fluff.
2. llms.txt files (the hype that works) There is a growing trend in the community to put an llms.txt file in the root of the site. It is basically a robots.txt for AI that serves clean Markdown. Google said this file itself isnt a ranking factor for them. But here is the thing: third-party RAG systems, Perplexity crawlers and other AI parsers eat it right up. It is the cheapest and most reliable way to feed an LLM a dry summary of your product without JS garbage and CSS layouts.
3. Content chunking for RAG Long paragraphs just dont work anymore. The text on the pages should be broken down into short, semantically independent blocks based on a simple rule: one intent = one specific answer. AI agents use vector search, pulling the piece of text that is mathematically closest to the users prompt. If your page is a mess of meanings, the algorithm will just ignore your content because it cant extract a relevant vector.
While big content sites and media are panicking over lost search traffic, this is a great window of opportunity for indie devs. Technically adapting a project for AI agents is much clearer and cheaper than classical SEO stuff.
To be clear, im not saying you should stop building links. Buying links and building a strong backlink profile still remains a very important factor for domain authority and even for AI trust signals. We still have to do that or competitors will eat us alive.
