guide
Generative engine optimization for professionals
Generative engine optimization is the practice of making yourself the answer a model gives, rather than a link on a page nobody opens. The mechanics differ from classic SEO in one decisive way: there is no results page to rank on. Either a sentence about you ends up in the answer, or you were not in the conversation at all.
updated
The answer replaced the results page
When someone asks an assistant for "three senior infrastructure engineers in Taipei who have run multi-region systems", the output is a short answer naming a few people. There is no page two and no long tail to be discovered from. Ranking eleventh and ranking nowhere produce the same result.
That compresses the entire game into a single question: when the model composes that sentence, does it have a specific, attributable claim about you to put in it?
Models quote more often than they summarize
Extraction favours short, self-contained statements. Lead with the direct claim in one sentence, then elaborate underneath — the first sentence is what gets lifted, and a sentence that only makes sense in the context of the three around it will be dropped or mangled.
Write claims that survive being taken out of context. "Maintains the open-source parser used by X" travels; "has been instrumental in driving key initiatives" does not survive its own paragraph, let alone a model's context window.
Publish the same facts in more than one format
People need HTML, parsers need JSON-LD, and agents increasingly prefer JSON or Markdown they can consume without stripping tags. Publishing all of them is not redundancy, it is coverage — each consumer takes the representation it handles best.
The rule that matters is parity: the structured data must say exactly what the visible text says. Structured claims that contradict the page are a trust penalty with both search engines and answer engines. Humetric generates its HTML, JSON-LD, JSON and Markdown from one consented projection precisely so they cannot drift apart.
Then make discovery and re-crawling cheap
Being readable is not the same as being found. A sitemap tells crawlers what exists, IndexNow tells participating engines the moment something changes instead of waiting for the next crawl, and an /llms.txt file gives AI systems a plain-language map of what a site offers and how to query it.
Humetric publishes all three, and the same checklist applies to your own site: one canonical URL per thing worth citing, listed in a sitemap, changed content pushed rather than waited on.
Key takeaways
- Optimize for being quoted, not for being clicked.
- Lead every claim with one self-contained sentence.
- Publish the same facts as HTML, JSON-LD, JSON and Markdown.
- Make re-crawling cheap: sitemap, IndexNow, llms.txt.
Questions about this guide
What is generative engine optimization (GEO)?
GEO is optimizing content so that generative models cite it in their answers. Where SEO competes for a position on a results page, GEO competes for a sentence inside a generated answer, which rewards structured, self-contained, verifiable statements over keyword coverage.
Is GEO just SEO with a new name?
No. They share the fundamentals — crawlability, structure, canonical URLs — but diverge on the target. SEO optimizes a page for ranking and a click; GEO optimizes a claim for extraction and attribution, so specificity and structured data matter more than keyword density.
How do I check whether AI systems can see my profile?
Fetch the page without JavaScript and confirm the text is present in the source, read the site's /robots.txt to confirm AI user agents are allowed, and look for schema.org JSON-LD in the markup. If any of the three fails, models are working from second-hand mentions.
Keep reading
Publish a profile machines can actually read.
Submitting a profile is free. You choose what is public field by field, contact details stay hidden, and every read is logged.