Getting found in AI search: a practical GEO checklist
Answer engines read your site differently than classic search engines. Entities, readable HTML and honest structured data — the concrete checklist that still works.
Daniel Martínez Marqués
independent developer & software builder
Search engines used to be the only machine readers that mattered. Now your content is also read by AI systems: answer engines that summarize instead of link, AI agents that browse on behalf of users, and models that get trained on your pages. "Generative Engine Optimization" (GEO) is the discipline of making content work for these readers.
The good news: most of it is classical SEO done properly. The bad news: a few habits from the growth-hacking era actively hurt.
Own your entity slot
AI systems build understanding from entities — stable names with consistent attributes. For a person or a company, that means:
- One canonical name, used consistently. Nicknames and rebrands should map cleanly to the main entity.
- A consistent factual description of what the entity does, repeated in the same words across pages. Variation is good writing; ambiguous identity is bad data.
- Structured data that matches what the page visibly says. An Organization or Person schema that contradicts the visible content teaches the wrong lesson.
- Links between related entities: person → product → studio → writing. This site tries to do exactly that — people, tools, a studio and a blog connected by clear labels.
Make the page readable without JavaScript
AI systems crawl pages the way search engines do. A page that renders content only after JavaScript execution is a gamble: some systems execute, some don't, and none reward the risk. The essentials — who you are, what you build, what this page is about — should be in the initial HTML.
Semantic HTML is the same story. Headings, nav, article, time and table semantics form a machine-readable outline of your page. An outline beats a blob.
Write for extraction
Answer engines don't "read" like humans; they extract. The practical implications:
- Put the answer near the question. A paragraph that opens with the answer — then explains — gets extracted far more often than a paragraph that builds to it.
- Use concrete language. "Fast, secure and maintainable" is meaningless. "Ships as a static site with pre-rendered HTML" is extractable.
- Label your claims. Dates, scopes and numbers make content more trustworthy to extractors — and to humans.
- Be a source, not a summary. Original technical content — measurements, code, experiments, explained trade-offs — is what AI systems struggle to generate themselves. That's the durable moat.
Structured data that tells the truth
Schema.org types are how you hand AI systems a map of your page. Use them accurately:
- Article for articles, with real publication and modified dates.
- BreadcrumbList for navigation context.
- SoftwareApplication for software products, with real metadata about platforms and categories.
- Person for yourself, with links to platforms you actually use.
Do not add schema for things the page doesn't show. Mismatched structured data trains systems to ignore you.
GEO is just engineering discipline
The pattern across everything that works: consistent identity, readable semantic HTML, extractable prose, honest structured data and internal links that connect related entities. None of it is magic, and all of it compounds. The signals that search engines and AI systems look for today are the same signals a careful engineer would ship by default.
I built SEO Engine around this idea — deterministic checks for the parts that are checkable, so the discipline doesn't depend on remembering it.