AI Search Is Eating Your Clicks — Here's What Actually Helps
You've probably noticed: search results don't look like they used to. Google shows AI Overviews, people ask ChatGPT and Perplexity instead of typing a query into a box. When an AI answer appears right there, users often don't click through at all.
Reported numbers back this up. When AI summaries show up, users click links only about 8% of the time versus 15% when there's no summary. On pages where AI Overviews appear, traffic from search can drop 30% or more. E‑commerce sites are already reporting roughly a 22% drop in search traffic where AI-generated suggestions eat the click. That's not speculation, the shift happening now.
As a website owner, what do you do about it? A lot of the buzz is around Generative Engine Optimization (GEO) — the idea of optimizing for AI systems the way we once optimized for Google. The AIO/GEO market is projected to go from about $886M in 2024 to $7.3B by 2031 (around 34% annual growth). Over half of marketers are already using generative AI in their SEO workflow, and three-quarters of digital agencies have launched GEO offerings. The question isn't whether the ground is moving, it's what actually works.
Here's what I've learned from doing the work: LLMS.txt, on-page AIO, and the rest.
This Isn't Just "SEO Again"
Classic SEO is about ranking for keywords and getting the click. GEO is about being the source AI systems *cite* — so when someone asks a question, your content is in the answer. That means different priorities: clarity, structure, and making it obvious what your page is about. AI systems (and users) need to understand you in one pass.
Google AI Overviews showed up in over 13% of U.S. desktop searches by March 2025 and had 1.5 billion+ monthly users in Q1 2025. So even if you're not thinking about ChatGPT or Perplexity yet, Google alone is already surfacing AI answers in a big way. If your content is vague, buried in jargon, or structured for bots from 2010, you're leaving visibility on the table.
What Moves the Needle
The stuff that helps AI (and humans) find and trust you:
Clear structure. Headings that state the topic. Paragraphs that answer one idea at a time. Lists and definitions where they make sense. If a crawler or an LLM can't quickly infer "this page is about X and says Y," you're harder to cite.
Entity-rich, intent-aligned content. Name the things you're talking about. Answer the question the page promises to answer. AI systems are better at citing content that clearly matches a query and uses consistent, recognizable concepts.
E-E-A-T without the theater. Experience, expertise, authoritativeness, trustworthiness — not keyword stuffing. Author bylines, clear attribution, and content that shows you know what you're talking about. That helps both search and AI.
Technical basics. Fast, crawlable pages. Sensible URLs and internal links. Mobile-friendly and accessible. If your site is a mess under the hood, no amount of "AI optimization" will fix it.
Schema and structured data. JSON-LD (Organization, Article, FAQPage, and the like) gives search engines and AI systems an explicit map of what your page is about. It doesn't replace good content, but it makes it easier to parse and cite. Use it where it honestly describes your content — not stuffed or fake.
FAQs on the page (and in schema). Pages that answer real questions in a clear Q&A format are easier for AI to quote and attribute. Adding an FAQ section with real questions your audience asks, plus FAQPage schema so crawlers can see question/answer pairs, is one of the highest-leverage moves. This site does it on every article: the takeaways at the bottom are also output as FAQ schema so both humans and systems get the same structure.
LLMS.txt and metadata. I've added LLMS.txt to this site and keep it updated when I publish. Studies so far don't show a measurable lift in AI citations from LLMS.txt alone — adoption among top sites is still small. I treat it as low-cost future-proofing: a signal about who we are and what we offer, and one less excuse for a crawler to ignore us. Not the main lever.
If you want a practical list of what to do (and what to skip), I wrote The GEO Checklist I Use — same ideas, in checklist form.
The Honest Take
AI search is eating your clicks. GEO is real, and the market is growing fast. But the levers that work today are mostly the same ones that have always made content findable and useful: structure, clarity, and technical hygiene. LLMS.txt is worth doing because it's cheap and will likely matter more later.
Focus on making your content easy to understand and easy to cite. Then add the metadata and the LLMS.txt file. In that order.
---
