Five years ago SEO was mostly one question: how do you rank higher on Google. In 2026 that's only half the picture. The other half: how an AI model cites you when someone doesn't even open Google, but asks ChatGPT or Perplexity instead.
These are two separate channels with two separate logics — and most sites are prepared for neither. Let's break down what matters.
1. The technical foundation nothing works without
Before any „AI optimization": the technical basics aren't optional. Without them, even the best content misses.
- Speed: Core Web Vitals isn't marketing fluff. A slow site loses in both ranking and conversion.
- Clean, semantic HTML: the crawler and the AI model read from the same source. Messy structure means poor understanding.
- Correct canonical, hreflang, sitemap: boring, but it decides whether the right URLs get indexed at all.
- Structured data (JSON-LD): Organization, Article, Product, FAQ — this is how a machine reads unambiguously what's what.
Plenty of sites pour money into content marketing while their robots.txt blocks key AI crawlers, or the sitemap has been stale for six months. Plumbing first, furniture later.
2. What AI discoverability actually means
When ChatGPT or Google AI Overviews answers, it cites sources. Whether it cites you is not random. A few things that measurably matter today:
- Clear, stated claims. A model struggles to cite a page that talks around the point. If you're the best at X, say it plainly.
- Question-answer structure. Content that gives tight answers to real questions is far more citable than a flowing essay.
- Freshness and dates. Models favor recent, dated sources. A „2024 guide" is dead weight in 2026.
- Consistent entity. If your name, data and description match everywhere (site, JSON-LD, external profiles), the model identifies you with more confidence.
3. llms.txt — robots.txt's AI-era sibling
More and more sites — ours included — publish an /llms.txt file. It's a simple, machine-friendly summary of what the company does, its services, pricing and principles. No magic, but exactly the kind of structured context a language model likes.
robots.txt tells the machine where it can't go. llms.txt tells it what's worth knowing about you.
Important: llms.txt doesn't replace real content and structured data. It complements them. But it's cheap, quick to produce, and points exactly where search is heading.
4. What to do tomorrow morning
If you do one thing: run an honest technical audit. Not a PDF full of green checkmarks — one that tells you where you're losing traffic today. On most sites the biggest win isn't a new tactic, it's fixing existing mistakes.
For us, every marketing engagement starts with exactly this kind of technical SEO audit — because it's pointless to produce content for a site the machine half-understands.