The technical work that decides whether an AI engine can read you
Short answer
Two conditions decide most of it. The answering crawlers have to get through your CDN and your robots.txt, which means allowing OAI-SearchBot, Claude-SearchBot, PerplexityBot and Bingbot, and the content they came for has to exist in the raw HTML rather than being assembled by JavaScript in the browser. Training crawlers such as GPTBot, ClaudeBot and Google-Extended are a separate and lower-stakes decision, and blocking the wrong one removes you from answers while protecting nothing. Schema is worth having because it states filterable attributes as data, though Ahrefs' matched study of 1,885 pages that added JSON-LD against 4,000 controls found no lift in AI citations, and llms.txt is close to inert for discovery, with Ahrefs finding 97 percent of the published files it located across 137,210 domains had received zero requests.
Crawler access, server-rendered HTML, schema and llms.txt: which technical work an answer engine needs, and which of it does nothing at all.
The technical half of this work is unglamorous, cheap and finite, and it is also the half that most often explains why a business is missing from AI answers entirely. A menu or a price list that only exists after JavaScript runs is invisible to the crawlers behind ChatGPT, Claude and Perplexity, and that single template decision outweighs every content tactic sold on top of it. NovaTechRay puts the crawler and rendering checks first because they are binary: either the page can be read or nothing else you do matters. The part we will concede, which most vendors will not, is that two of the most-sold technical deliverables in this field, schema as a citation lever and llms.txt, have evidence pointing at roughly nothing. Ship them if you want them for other reasons, and do not let anyone bill you for them as the reason you get cited.