How an AI Overview works
On some queries Google places a written summary above the ordinary results, generated by its own models. The summary is assembled from pages Google has crawled and indexed, together with what the model already holds, and it carries links to a small number of the sources it drew on. Those links are the only route from the summary back to a website.
It does not appear on every search. Whether one shows varies by query, by country and language, by device and over time, and the same words typed twice can produce a different answer or none at all. Eligibility to be cited follows from ordinary indexing: your page has to be crawlable and indexed. The usual snippet controls apply too, so a page carrying a no-snippet instruction can be kept out — at the cost of losing its normal search snippet as well, which is rarely a trade worth making.
Why AI Overviews matter
They change the shape of the page. Ordinary results move down the screen, and a searcher whose question is fully answered in the summary may never scroll at all. That pressure falls hardest on straightforward informational queries — definitions, how-to steps, quick comparisons — where the answer is short enough to be summarised without loss.
It falls much more lightly on queries with a decision behind them. Someone choosing a supplier, checking prices or booking something still needs a page, a person and a form. Being cited in the summary is worth having because it puts your name in front of the searcher at the moment they are forming an opinion, even when the click goes elsewhere.
Where it goes wrong
The first mistake is treating citation as a product that can be bought. There is no submission, no markup and no bid for a place in an AI Overview, and anyone selling a guaranteed spot is selling something they do not control.
The second is measuring it like a ranking position. It is not a position, it appears inconsistently, and a single check from one browser tells you very little. The third is reacting to a fall in clicks without looking at which queries lost them. If informational pages lost traffic while enquiries held steady, the business result may be unchanged; if commercial queries lost clicks, that is a different problem with a different fix.
What to do about it
Answer questions properly and early. A page that states its answer plainly near the top, in language a person would use, is the kind of page that gets quoted — and it also reads better for the human who arrives. Keep facts checkable and attributed, keep pages crawlable, and make sure who you are and where you operate is stated clearly on the site, since summaries lean on sources they can identify.
Then measure sensibly. Watch impressions against clicks in Search Console by query group rather than in total, and track enquiries as the outcome that matters. Where AI answers are a large part of your market, treat AI Overviews optimisation as one strand of search work rather than a replacement for it, and expect the detail to keep moving. This is the fastest-changing part of search, and anything stated too confidently about it tends to age badly.
