AI Overviews · SGE
Google AI Overviews / Search Generative Experience
What are AI Overviews?
- Google's generative AI summary at the top of search results
- The experimental name SGE, renamed when the feature became official
- Being cited matters more than ranking first
What AI Overviews really change is not the ranking table but who owns the first screen of search results. It used to be that the #1 link sat at the top of the screen. Now the summary card takes up nearly the whole first screen, and the blue links appear only after you scroll. So even at the same #3 position, traffic differs sharply between a page cited in the card and one that isn’t, and you can end up with a ranking report that looks fine while traffic keeps dropping.
It helps to tell three easily confused things apart. A featured snippet lifts a sentence from a single page as is, whereas an AI Overview is an answer rewritten and merged from multiple documents, so the original sentences don’t survive intact. An AI Overview is also a card layered on top of the existing results, while AI Mode is a conversational search screen in its own right. Triggering is selective too: overviews show up more often for explanatory queries such as 'what is X' or 'X vs. Y' than for brand names or queries close to purchase, and the card usually carries three or four citation links. Citations are not picked in ranking order. A document outside the top 10 is often pulled in simply because its definition sentence is clean.
The most common problem is the belief that 'responding to AI Overviews means buying a new tool.' In reality, most of the work is editing the intros and subheadings of existing documents. At the other extreme, blanket-applying nosnippet or Google-Extended because 'AI is scraping our assets for free' removes you from citations and cuts regular search visibility too. AI Overviews also appear or disappear depending on account, region, and timing, so concluding 'our keywords don’t trigger one' because it didn’t show up on the team lead’s PC means you will misdiagnose a traffic decline as a ranking drop. Raising keyword bids before you find the real cause is the most expensive mistake.
John KimInstructor John Kim, author of B2B Marketing Design and CEO of Nanum Consulting
Hyerang ParkSpecialist in content storytelling and AI-driven brand contentAt Google I/O, its annual developer conference, in 2023, Google unveiled SGE (Search Generative Experience), an experiment that combined generative AI with search. In 2024 it rolled the feature out generally and renamed it 'AI Overviews.'
Let’s look at a hypothetical case: Company A, a B2B manufacturer of industrial automation parts with 70 employees and annual revenue of around KRW 18 billion (≈ US$13,467,000). Its 120+ technical blog posts, built up over three years, had been generating 12,000 sessions a month, but traffic fell to 8,000 sessions within six months even though most of its Google rankings stayed the same. The marketing team lead started digging into the cause: "Our rankings are fine, so why aren’t people coming?"
- The head of marketing searched 40 key keywords one by one in an incognito window. An AI Overview appeared for 27 of them, and Company A's documents were cited in only 3.
- When the content editor took apart the 3 cited posts, the common thread was a two-line definition in the first paragraph. The conclusion: "Not one post that bragged about the company and buried the definition in the fourth paragraph got cited."
- They picked the top 30 posts, rewrote each intro as a one-sentence 'X is Y' definition, turned subheadings into questions like 'What's the difference between the two?', and republished them over six weeks.
- They asked the dev team to add FAQ schema and author credentials. When someone internally proposed blocking crawlers, the team lead drew a line: no Google-Extended block.
- Separately from Search Console metrics, the team built a sheet where they logged by hand, every week, whether each of 40 keywords was cited. The team lead checked only the changes on Monday mornings.
- Three months later, cited keywords had grown from 3 to 14, the blog had recovered to 11,000 sessions a month, and technical contact-form submissions had risen from 9 to 16 a month.