What AI visibility means: selection, citation and absorption

Take a customer who asks ChatGPT, Gemini or Claude which accountant in Lisbon handles freelancer taxes. Before the answer appears, the assistant may search the web, collect a handful of pages, and write a reply from them. Your business can take part in that answer in several ways, and each one means something different.

A page can be cited without being absorbed: the link sits under the answer, but the text came from somewhere else. A business can be recommended without being cited at all, because the assistant drew on a directory, a review site or its own training data. Treating all of this as one "AI visibility score" hides the problem you need to fix.

The idea that content changes can shift what generative engines show comes from the GEO paper presented at KDD 2024. It measured changes inside its own benchmark. That is a useful signal that content matters, not a promise that a specific edit will move a specific assistant.

The split between selection and absorption is set out in a 2026 preprint by Zhang, He and Yao. It defines selection as a platform choosing a source and absorption as a cited page contributing language, evidence, structure or facts to the final answer, and reports that engines differ: some cite many sources lightly, others cite fewer sources that shape the answer more. It has not yet been peer reviewed, so treat it as a strong framing rather than settled fact.

A separate observational study of LLM-based and traditional search found that LLM search cites a different, more diverse set of domains than traditional results. Again, that describes what was seen, not why.

Each outcome points to a different fix:

• Not selected at all. Start with access. If search crawlers can't reach the page, or it is blocked from snippets, nothing else matters. See our guide to AI crawler controls.
• Selected but not cited. The page is found but something else is preferred as the source. Compare it with the pages that are cited for the same question: do they state the fact more directly, with a date and a named source?
• Cited but not absorbed. You get the link but the answer uses someone else's facts. Look at what the answer actually says and whether your page states it clearly and first-hand.
• Absorbed but not recommended. Your information informs the answer, yet a competitor gets named. That is usually about how your business is described across the web, not about one page.

Scope saves every answer it collects, with the question, assistant, market, language and date. From those saved answers it reports whether you were recommended, mentioned, or left out, which competitors were named, and which pages each assistant cited. Every finding opens to the answer behind it.

The paid report also checks whether the answers actually use your facts, and sorts each one: cited and used, cited but not used, used without credit, or not present. It flags facts the assistants get wrong about you, too. That keeps absorption separate from citation, so a link under an answer isn't mistaken for influence over it.