Getting cited by ChatGPT, Claude and AI Overviews
AI visibility is whether AI assistants can retrieve your pages and are willing to quote them. It is a narrower and more tractable problem than the category's marketing suggests: a large part of it is crawler access and page structure, both of which you can inspect and fix, and a smaller part is measurement, which is mostly sampling and should be described that way.
These guides are written from evidence and cite what each platform has actually said — including where that contradicts the prevailing advice.
llms.txt
What the file is, which platforms have actually committed to reading it, and whether shipping one does anything.
AI crawlers
Training bots and retrieval bots are not the same thing, and blocking one to stop the other is the most common own goal in AI SEO.
Can it be measured?
What AI visibility tools are really sampling, why their numbers disagree, and what we refuse to estimate.
AI Overviews
Google's own position is that there's no separate optimization. What that leaves you actually able to influence.
Four things that hold across every surface
Google's Overviews, ChatGPT, Claude and Perplexity use different indexes and different retrieval paths. These properties travel across all of them.
Access is the prerequisite
A retrieval crawler that cannot fetch a page cannot cite it. Most accidental invisibility traces to a robots rule written before training and retrieval bots were distinguishable.
Extractability beats length
Assistants quote passages, not pages. A section that stops making sense once lifted out of its context is not going to be used, however well researched it is.
Verifiable beats persuasive
Unsupported superlatives are a liability to whoever cites them. Specific, checkable statements get quoted; adjectives get skipped.
Most of the measurement is sampling
No assistant publishes per-site citation data. Any share-of-voice figure is inferred from a prompt sample, and should be read as a survey rather than a count.
What Zylx does here, stated exactly
Zylx scores pages for AI answer readiness from real on-page signals and returns the specific blocks a page is missing, and it assesses AI crawlability as part of its site crawl. Pages carrying unsupported claims are blocked from scoring as answer-ready, on the reasoning that an answer engine has no good reason to quote something it cannot verify.
Zylx does not currently measure whether AI answer engines cite you. No citation-measurement source is wired into the product, and rather than estimate a figure, the system records every source as unsupported and says so. Why that is a deliberate choice is worth reading if you are comparing tools in this category.
Start with what you can control
Crawler access and page structure are deterministic, inspectable and fixable. Everything else is a survey.