Ranking #1 on Google used to be the whole game. It isn't anymore. Recent analysis of AI Overviews and chat-based search shows the overlap between top-ranking pages and the sources an AI actually cites has fallen from roughly 70% to under 20%. Perplexity cites about 22 sources per answer on average; ChatGPT cites about 10; the two share only about 11% of the same domains. Ranking well is no longer enough — your content has to be something a model can find, parse, and quote with confidence.
This is what people now call Generative Engine Optimization, or GEO. It isn't a replacement for SEO — Google's own guidance calls it "still SEO" — but it shifts the emphasis: from ranking position to citation, from keywords to extractable, quotable answers. Here is the concrete checklist we used to get fanpino.com itself into shape for this, in the order we'd recommend doing it.
1. Check you aren't accidentally blocking the crawlers you want
The most common own-goal in GEO is a robots.txt file that blocks AI crawlers by accident — usually copied from another site, or a blanket Disallow added for an unrelated reason. Audit yours line by line. There are two distinct families of AI bot worth knowing apart:
- Training crawlers — GPTBot, ClaudeBot, CCBot, Google-Extended, Applebot-Extended, Amazonbot. These feed model training and give nothing back in the moment.
- Retrieval crawlers — OAI-SearchBot, ChatGPT-User, Claude-SearchBot, PerplexityBot, Perplexity-User. These fetch pages live to answer a real question and are what actually gets you cited.
You can opt out of the first group without losing anything from the second — they're separate user agents specifically so publishers can make that choice. What you should not do is block both by accident with one careless rule.
2. Add an llms.txt file
llms.txt is a plain-text file at your site root that gives language models a short, structured map of what your site actually offers — products, key pages, docs — instead of making them guess from a sitemap built for search engines. It won't move your Google ranking, but it's a cheap, direct way to hand an AI system a clean summary instead of the crawl guessing your value proposition from your homepage hero copy.
3. Structured data is how a model understands your page without reading all of it
JSON-LD is the only structured data format Google recommends, and it's the format AI systems parse most reliably too. The types worth prioritizing:
- Organization / WebSite on every page — identifies who you are.
- Product (or SoftwareApplication) on pricing pages — lets a model answer "how much does X cost" directly from the markup.
- FAQPage — Google retired the visual FAQ rich result in mid-2026, but the schema still strongly helps AI systems extract direct Q&A pairs. Write the FAQ content first; the schema is just a wrapper around real answers.
- BreadcrumbList — cheap, still boosts click-through in traditional search, and gives a model your site's category structure for free.
- Article / BlogPosting with a real published date and author — a page with no visible date is a page a model can't trust for anything time-sensitive.
4. Submit through IndexNow, not just a sitemap
A sitemap tells a crawler what exists; it still has to come back and check it, on its own schedule. IndexNow is an open protocol — co-developed by Microsoft, free to use, no account required — that pushes a URL to Bing (and other participating engines) the moment it changes, instead of waiting for the next crawl. This matters more than it sounds: ChatGPT's web search runs on Bing's index, not Google's. A page Bing hasn't seen cannot show up in a ChatGPT answer, no matter how well it ranks on Google.
5. Write for the question, not the keyword
A generative engine doesn't return ten blue links for one query — it fans a question out into several sub-questions and pulls the best answer to each from wherever it can find one. A page that thoroughly answers one specific sub-question can get cited even if it never ranks on page one for the head term. This favors direct, self-contained answers over content built to rank for a single phrase: state the answer in the first sentence of a section, then support it, rather than building up to it.
6. Measure citations, not just clicks
Referral traffic from AI assistants is real but easy to undercount — several strip the referrer, so a share of it lands in your analytics as "Direct." A dedicated channel grouping (or at minimum, checking your server logs for the retrieval bot user agents above) gives a truer picture than session counts alone. The metric that actually matters going forward is simpler to state than to measure: does your brand or URL show up inside the answer, not just somewhere in a list of links below it.
None of this is exotic. It's a robots.txt audit, one small text file, structured data you were probably already halfway to having, an instant-indexing ping, and writing answers instead of copy. We rolled all of it out on this site over the course of a few days — it's exactly why this article exists.
