An AI Overview is the generated answer at the top of a Google results page, and the sources beside it are pages Google’s systems chose to build that answer from. Getting cited is now a visibility goal in its own right, because for a growing share of queries the Overview is what a searcher reads instead of the ten blue links beneath it.
The useful question is not how to game that selection. It is what genuinely influences it, and what does not, because a substantial industry has grown up selling the difference — which we take apart in generative engine optimization: what it is and what it isn’t.
What Google actually says
Start with the primary source, because it settles more than most commentary does. Google’s guidance on AI features states plainly that there is nothing special site owners need to do to appear in AI experiences, and that the same practices which make a page eligible for good search visibility apply. The pages surfaced come from a wide set of helpful results.
Read that carefully and two conclusions follow. First, any product promising “AI Overview optimisation” as a distinct service is selling a technique Google says does not exist. Second, and more usefully, the work that earns citation is the work that already earned rankings, which means nobody has to start over.
How the selection actually works
An Overview is built by retrieving relevant web content and generating a summary grounded in it. That retrieval step is the part worth understanding: the model is not searching the open web freely, it is drawing on pages the search system already considers relevant and trustworthy for the query.
This has a practical implication that surprises people. Ranking on page one is close to a prerequisite, but it is not a guarantee, and position one does not translate into citation one for one. Overviews routinely cite pages from further down the results, and often pull in formats the classic ranking does not favour.
We checked this while writing. For the query “what is generative engine optimization,” Google returned an Overview built from six sources, and three of those were YouTube videos, alongside a course platform, a software vendor and an SEO tool. None of the traditional SEO publications you would expect to own that term appeared in the citation list at all. That is worth sitting with: the sources an Overview picks can look quite different from the sources that rank.
What you can actually influence
Four things, in order of how much they matter.
Be visible in normal search. Since retrieval draws on ranking content, everything that earns rankings earns eligibility: relevance, coverage, technical health, and authority. Nothing here is new, which is precisely the point.
Make passages liftable. An Overview quotes a self-contained chunk of text. A page that states the answer plainly, under a heading that matches the question, provides an obvious candidate. A page that spends four paragraphs building to the point provides nothing to take. This is the single most underrated adjustment, and it costs nothing but editing.
Answer the adjacent questions too. Overviews synthesise across several sub-questions. A page that covers the follow-ups as well as the headline query gives the system more reason to draw from you repeatedly, and often earns the citation over a page that answers only the narrow question.
Be a credible entity. Author identity, consistent brand information, real expertise and evidence all feed the trust signals the retrieval step depends on. Google’s helpful content guidance is the reference here, and its emphasis on demonstrable experience matters more when a system is deciding which handful of sources to stake an answer on.
What you cannot influence
Whether an Overview appears at all is Google’s decision, and it varies by query, by intent and over time. Which sources get chosen changes between searches and between users. There is no submission process, no markup that requests inclusion, and no setting that improves your odds.
Accepting that boundary is more useful than it sounds, because it redirects effort toward the things that do respond: coverage, clarity, structure and credibility.
Measuring something Google does not report
Search Console does not break out AI Overview appearances. Impressions and clicks from Overviews are folded into overall Search performance with no filter to isolate them, which means anyone quoting you a precise “AI Overview traffic” number is estimating.
Three practical approaches work. Check manually and on a schedule: run your priority queries, record whether an Overview appears and which sources it cites, and watch that list move over months. Watch the shape of your Search Console data: rising impressions with flat clicks on informational queries is the pattern that usually accompanies an Overview absorbing the answer. And shift the success measure downstream, because if Overviews are removing the visits that were never going to convert, sessions will fall while conversions hold, and only the second number was ever the point.
What this means for how you publish
The strategic shift is smaller than the noise suggests, but it is real. Content whose entire value was summarising what is already known has lost most of its purpose, because summarising is exactly what the Overview does, instantly and for free. Content carrying something that cannot be synthesised — original data, first-hand experience, a defensible opinion, evidence from work actually done — becomes more valuable, because it is both harder to replace and more attractive to cite.
If you want the local-business version of this question, we covered how AI Overviews choose which local businesses to name separately, since proximity and profile data change the calculation. The platform-specific version for OpenAI is in ChatGPT SEO. And if you would rather have the whole programme handled, that is our SEO and AI search service.
