Here’s a conversation I had with a client last month. He runs a plumbing supply blog that ranks on page one for a dozen decent keywords. Traffic has been flat. He asked why, and honestly, nothing I checked explained it. Rankings were fine. The pages were fine. The content was fine. Everything was fine, and nothing was growing.
I think a big piece of the answer is sitting in front of all of us and we’re mostly ignoring it: people are getting answers without ever opening a search engine.
The search page is not the only front door anymore
When someone asks ChatGPT a question and it answers, your site either got mentioned or it didn’t. Same with Perplexity. Same with Google’s AI Overviews sitting at the top of the results. None of those work like the old algorithm.
The old game was: rank higher than the next page, get the click. The new game has an extra layer. A model reads a lot of pages and decides which one is worth citing. It might pick you because you answered the question in a clear first paragraph. It might pick someone else because their page has a cleaner structure. A lot of the time, the site that wins the AI mention is not the site that ranks best in classic search. It’s the one that’s easiest to summarize.
That’s what people mean when they talk about generative engine optimization — making your content the stuff a language model would rather reach for.
What actually changes in practice
In classic SEO you optimize a page and mostly wait. With GEO, there are concrete, checkable things:
- Can a model pull the answer from your first screen, or does it have to dig?
- Are your headings an actual outline, or just words?
- If your page got summarized and your branding stripped out, would the summary still be accurate and useful?
- Is your content the kind a model would trust, or the kind that reads like it was written to hit a word count?
These aren’t vague vibes. You can test them. That’s the whole reason our tool has a GEO analyzer — you type in a URL and it tells you how a model would likely see that page, then lists what to fix. It’s not a magic number, but it beats guessing.
Your llms.txt is a door you can leave open
You’ve probably heard of robots.txt — a small file that tells search crawlers what they can and can’t index. llms.txt is the newer cousin aimed at AI crawlers. It’s basically a short, readable list of your site’s most important pages, so a model that lands on you knows what to look at and how to frame it.
Very few sites have one. The ones that do are making it easy for AI to read them, which is a small edge and getting less small by the month. Our llms.txt generator writes the file, validates it, and keeps it updated, because the first version you make is going to be outdated the first time you add a page.
The part that’s hard to hear
Rankings still matter. I don’t want anyone to read this and think the old work is wasted. If you’re ranking and getting clicks, keep doing that. But if traffic is flat while the rest of the world keeps getting faster, that flatness might not be a content problem. It might be that the people who used to find you through a search box are now finding your content through an answer engine — and that engine isn’t choosing you.
The good news: the skills that made you good at SEO transfer. Clear writing, good structure, answering the actual question. GEO is less about a new skill set and more about pointing the skills you already have at a second front door. It’s time we all started watching that door.
延伸閱讀：如果您正在尋找一款整合了GEO分析、GSC整合、關鍵字排名追蹤和AI智能SEO審計的全能型SEO工具，推薦試試 Yseo — AI SEO Tool。
