Generative AI Search
AI Search / Generative Search
What is generative AI search?
- Search that gives a synthesized 'answer' instead of a list of links
- Usually uses RAG to find current documents and generate an answer grounded in them
- Beyond SEO, optimizing for AI citations (GEO/AEO) is the new challenge
What generative AI search really changes is not the quality of answers but the path traffic takes. Search results pages used to be a gateway to your site; now the results page itself is the destination. After reading the three paragraphs AI has summarized, users have little reason to click the links below. So rankings stay the same while clicks drop first, and the person in charge stares at the ranking report and misses the cause.
There are three main forms: the embedded type, which places a summary at the top of search results (Google AI Overviews); dedicated engines that replace search with conversation (Perplexity); and the tool type, which searches the web only when needed (the web search feature in chatbots). All three pick documents to cite by similar criteria. A document is more likely to be chosen when the question and answer sit together in one block, when figures, criteria, and conditions are written as text rather than images, and when it was updated recently. The key point is that search ranking and citation probability are separate things. It's common for a page ranked in the top 3 to be ignored entirely while an FAQ page outside the top 10 gets cited, because Q&A fragments are easier to excerpt than long narrative pages.
Two things most often go wrong in practice. One is the myth that 'if we publish enough posts, we'll eventually get picked up.' Ten pages with exact prices, specs, and conditions beat 100 near-identical posts when it comes to citations. The other is blocking AI crawlers in robots.txt during a security review, then wondering why 'we're the only ones missing from AI answers.' If you block them, you never even become a candidate. Conversely, if you open access and then neglect your content, a three-year-old price list or a discontinued course can end up in the answer as is, and customers send inquiries based on those terms. That's why clearing out outdated pages comes before writing new ones.
John KimInstructor John Kim, author of B2B Marketing Design and CEO of Nanum Consulting
Seolyeong ParkGenerative AI and AX training specialist with 17 years of hands-on experience and 200+ lectures, helping non-technical staff put AI to work right awayAs RAG, which combines search with text generation, matured, and tools such as Google AI Overviews and Perplexity went mainstream in 2023–2024, AI search became established as 'the next form of search.'
Consider a hypothetical case. Company A makes industrial filters, with 120 employees and KRW 32 billion (≈ US$23,941,300) in annual revenue. It hasn't cut ad spend and its search rankings haven't moved, yet organic search traffic to its website fell from 8,000 to 5,200 visitors a month in six months. When the CEO asked, 'What did our competitors do?', the marketing team set out to trace the cause.
- When Kim, the marketing team lead, opened Search Console, impressions were steady, but click-through rate had been cut in half, from 3.1% to 1.7%. The first report was, 'Our ranking hasn't changed, but people aren't coming to the site.'
- When they entered their 12 main keywords directly into Google AI Overviews and Perplexity, only 3 competitors were cited, across 9 of the answers, and Company A's pages did not appear even once.
- The development team found that during last year's security review, someone had blocked GPTBot and PerplexityBot entirely in robots.txt. The manager in charge said, 'I thought blocking every crawler was the safe thing to do.'
- They lifted the block and rewrote 30 product pages in a 'one-line question, one-line answer, three lines of evidence' format. All 18 spec tables that had existed only as images were converted to text tables.
- They took down 7 pricing pages for older models posted in 2021 and added an update note, 'As of March 2025,' to the pages they kept. That ended the problem of AI answering with outdated prices.
- When we re-measured 8 weeks later, Company A was cited in AI answers for 7 of 12 keywords, monthly visits rose from 5,200 to 6,800, and quote requests grew from 18 to 29 a month.