AI Overviews Audit: Which Queries You Can Actually Win

Across the Australian health, property and local-services query sets we have audited, about 40 percent of the keywords a client hands us never trigger an AI Overview at all, and roughly half of the ones that do are grounded in sources nobody is displacing this year: health.gov.au, a state regulator, a hospital network, Wikipedia. This audit tells you which queries are left, before you commit a content budget to the other two groups. What we run: Every query in your set, logged out, on desktop and mobile, from an Australian IP and from the second market if you have one, sampled across ten separate days rather than once. Overviews do not fire consistently for the same query, and a single-day snapshot will hand you a fire rate that is wrong by twenty points in either direction. For each query we record whether an Overview fired and how often, which domains it linked, and, the part that takes the time, which paragraph on those pages the Overview wording actually tracks. Passage matching is what turns an audit into instructions. Then we check your side of it: whether the pages you want cited are indexed at all, when they were last crawled, where they sit organically for the same query, and whether the passage you would want lifted exists as a self-contained unit or is spread across three paragraphs and a subheading. What you receive: A sheet keyed by query, carrying the fire rate over the sampling window, the cited domains, the cited passage quoted verbatim, your organic position, and a verdict on each one: winnable this quarter, winnable only with third-party corroboration, or not winnable. For the winnable set, a rewrite specification at paragraph level, naming the question each section must answer, in which words, with which figure and which source. Then a call to walk the sheet with whoever owns the content. For AHPRA-regulated clients we flag the queries where advertising rules constrain what the page may say: the testimonial prohibition under section 133 of the National Law, and therapeutic claims that need evidence on file. There is no point specifying a rewrite the practice cannot lawfully publish, and we would rather find that at audit than after the writers have been paid. For agencies and developers the equivalent constraint is state underquoting law, where price-guide language on a listing page is regulated and an Overview quoting the wrong figure is a compliance problem rather than a win. What this is not: Not a monitoring subscription. One measurement window, with a decision at the end of it. Not a score out of a hundred. Not an implementation project either. We specify the rewrite here; we do not perform it. Who should not buy this: If your pages sit outside the top thirty for most of the query set, AI Overviews are downstream of a problem this audit would only restate in more expensive language. Google grounds Overviews in passages from pages it already ranks, and there is no side entrance. We will say that on the first call and decline the work rather than sell you a document whose conclusion is fix your search rankings. The same answer applies if the query set is mostly transactional. Overviews fire on explanatory and comparative questions far more than on booking or purchase intent, so a set assembled from your paid keywords comes back mostly blank and you will have paid us to discover it. Send the questions your intake calls actually open with, the ones reception answers six times a day, and the audit will be worth reading. If nobody has written that list down, spend a fortnight on it before you spend anything with us.