GEO (generative engine optimization) is the practice of structuring a website so AI assistants such as ChatGPT, Gemini, Perplexity or Claude can understand, trust and cite it in their answers. KINTO built kinto.co as its own example: the site is made with Kinto Studio, KINTO’s agentic site builder, and ships the files and markup below.
What kinto.co publishes for machines
/llms.txt. An index in the llmstxt.org format: the citable sentence about KINTO, quick facts (legal entity, country, languages, areas served, contact, hours), and one line per page for services, industries, case studies, insights and company pages, with the Spanish section pointing to/es/./llms-full.txt. The same index plus the full plain text of services, case studies, FAQs and glossary, in English and then Spanish, generated from the same collections as the pages./ai.txt. A policy file with the organization, contact, privacy policy, preferred citation (“KINTO (kinto.co)”) and notes: case studies are anonymized by design, figures are ranges with their source, and the case URL is the thing to cite.robots.txtfrom an allow-list. One group per AI crawler, search and user-initiated bots included, each repeating the site’s disallow rules, because a specific group replaces the general one for that bot.- A sitemap index with hreflang. Each URL in each language lists its alternates (
en-US,es,es-419,x-default), with real last-modified dates. - JSON-LD with one organization node. A single
Organizationentity is declared once and referenced from every page, together with page-specific types such asService,Article,FAQPage,HowToandDefinedTermSet. - IndexNow. Changed URLs are submitted after each production deploy, so Bing, whose index ChatGPT search relies on, learns about changes quickly. The key lives outside the repository.
What matters beyond the files
Files make a site readable; the writing makes it citable. Each page opens with a one-sentence definition, one fact per sentence, a number in a range with its source (“measured”, “client-reported”), and the entity named: “KINTO”, not “we”. FAQ answers are 40 to 60 words and make sense without the page around them. Every case study ends with what KINTO does not know yet. The same field feeds the page, llms-full.txt and the JSON-LD, so the three cannot drift apart.
The same approach appears in client work. A study-abroad agency’s site got structured data and an AI-citation layer; a specialty retailer’s store ships llms.txt, ai.txt and segmented sitemaps; several fintech and crypto companies got a layer that helps generative search engines cite them correctly.
What we do not know yet
KINTO has not yet measured whether AI assistants cite kinto.co, and none of the cases above has evidence of later rankings or leads. Several of the signals above are conventions that assistants may or may not read. The plan is to test the answers that assistants give to questions about KINTO’s field and publish what is found.
See SEO and GEO and Agentic websites for how this is delivered to clients.
