Search Engine Land has published an analysis asking a direct question: what do 9 million AI-generated answers actually reveal about who stays visible in search?

The number is the story. Nine million answers is not a panel study or a handful of spot checks. It is a dataset large enough to show patterns in how AI systems — the chatbots and answer engines now sitting between publishers and their audiences — select, summarize, and attribute the sources behind their responses.

For operators, the question is practical rather than academic. Traffic that once arrived through a ranked list of blue links increasingly arrives through a synthesized paragraph, or does not arrive at all. When an AI answer satisfies the user, the click may never happen. What the 9-million-answer analysis tracks is the next-order problem: when these systems do pull from the web, which sites do they pull from, and how often do they name them?

Search Engine Land's piece frames this as a visibility question rather than a ranking question. Traditional SEO optimizes for position on a results page. AI answers redistribute that logic. A source can be quoted, paraphrased without credit, or skipped entirely — and each outcome has a different commercial consequence for the publisher.

The scale of the sample matters for confidence. Small audits of AI answers go stale quickly and overfit to a handful of prompts. A 9-million-answer corpus smooths out anecdote and exposes baseline behavior: how consistently the systems cite, which domains recur, and where the citation patterns diverge from classic search rankings.

For news publishers, e-commerce operators, and anyone whose revenue depends on organic discovery, the actionable takeaway is measurement. If AI intermediaries are becoming a parallel discovery channel, operators need to know whether their content appears in the answers, whether it appears with attribution, and how that presence trends over time. Studies of this size are the first credible yardstick for that work.

The analysis arrives amid a broader shift in how search platforms themselves handle AI content — experiments with AI-generated summaries, changes to attribution formats, and ongoing disputes over how publishers get credited, or compensated, when their material trains or feeds an answer.

We have not independently verified the dataset or its methodology, and operators should treat headline findings from any single-vendor study with the usual caution. But the scale here — 9 million answers — puts the research in a different weight class from typical AI-visibility surveys, and the underlying question it asks is one every publisher relying on organic traffic will need to answer.

The full analysis is available at Search Engine Land.