AI Overview tracker: what 40 buyer queries show about who Google cites

An AI Overview tracker records, for each query you care about, whether Google shows an AI Overview and which pages it cites. We ran 40 buyer queries through one on 2026-10-24, from the United States, on desktop, in English. All 40 showed an AI Overview. YouTube was cited in 23 of them and Reddit in 14, and the page ranking first was cited in only 24.

Rank tracking alone no longer tells you what a buyer reads first. The AI Overview sits above the blue links and cites its own short list of pages, and about half of those pages are not in the organic results you track. This post gives the numbers from our sample, what they mean for the pages you write, and an agent prompt that runs the same check on your own queries.

How did we run the AI Overview sample?

We wrote 40 queries a buyer of B2B software or data APIs would type, 10 in each of four shapes: "best X", "X vs Y", "X alternatives" and "X pricing". Each query went through looot's Google AI Overview endpoint from Cloro once, with country US and language English, and returned the Overview text, its cited sources and the first page of organic results.

Examples of each shape: "best web scraping api", "apollo vs zoominfo", "bright data alternatives", "reddit api pricing". One run per query is a sample, not a score: AI Overviews change between runs, so treat each count as what Google showed that morning.

How often did Google show an AI Overview?

Every time. All 40 buyer queries returned an AI Overview, in every shape. The Overviews were long: the median ran 262 words, the shortest 156 and the longest 519. They also fed the "People also ask" box: 140 of the 156 questions there were themselves answered by an AI Overview.

For a buyer query in our sample, the Overview was the first thing on the page every time. If your report only has positions, it is missing the block people read before position one.

Which sites does the AI Overview cite?

Video and forums first, then a long tail. Across the 40 answers we counted 177 different cited domains. YouTube appeared in 23 answers and Reddit in 14; no other domain appeared in more than 6. The rest were comparison posts, vendor blogs and docs pages, most of them cited once.

The shape of the query changed who got cited. YouTube was in 8 of 10 "best" answers but only 2 of 10 "pricing" answers. Reddit was in 6 of 10 "vs" answers and 1 of 10 "pricing" answers. Comparison queries pull in opinions; pricing queries pull in the vendor.

Does ranking first get you cited?

Often, but not reliably. The page in organic position one was cited in 24 of the 40 Overviews, so in 16 the top result was left out. Going the other way, about half of the cited domains (52%) also appeared in that query's first page of organic results; the other half came from pages that did not rank on page one at all.

Two patterns stood out:

1. Pricing queries cite the vendor. For 8 of 10 "pricing" queries, the vendor's own site or docs was among the sources, for example serpapi.com for "serpapi pricing" and firecrawl.dev for "firecrawl pricing". The two exceptions were "reddit api pricing" and "linkedin api pricing", where third-party pages answered instead.
2. Alternatives queries do not. For 9 of 10 "X alternatives" queries, the vendor in the query was not cited. Its competitors' comparison pages were.

So the page that wins a pricing Overview is usually the vendor's own pricing page, and the page that wins an alternatives Overview is usually somebody else's comparison. If you sell against a bigger vendor, the alternatives query is where a page of yours can be cited.

What should an AI Overview tracker record?

Per query and per run: whether an Overview appeared, the cited URLs and domains, whether your domain is among them, the domains cited instead, and the organic top 10 next to it. The last one is what turns a list of citations into a to-do list, because it shows whether you need to rank, or to be the page that answers.

Keep the raw answer of every run. When a citation appears or disappears, the text tells you which claim it supported.

Run your own AI Overview check with an agent

The same endpoint we used is in looot's catalog, next to a cheaper one from AnyAPI, and both are paid per call. The table reads the cheapest live price from the catalog when this page loads.

Paste this into Claude Code, Codex or any agent that can call looot:

Searching the catalog and pricing a run is free; only the runs are charged. Saving each run's raw answer means you can rerun the same prompt next week and compare. Running it on a timer needs scheduled runs, not available yet, so for now you start the rerun yourself. For ChatGPT, Perplexity and Gemini answers too, see how to build an AI visibility tracker, and the endpoints for each engine are on the AI search platform page.

What this sample does not show

• It is 40 queries on one morning from one country. Your market, your country and mobile results can differ, so run your own queries before you plan around these counts.
• The queries are about B2B software and data APIs. Consumer, health and news queries behave differently, and Google shows fewer Overviews on some of them.
• One run per query. We did not measure how often the cited list changes between runs.
• looot was not cited in any of the 40 Overviews, and did not rank on page one for any of them. This is a study of who Google cites, not a claim about looot.
• The sample names domains only. We did not judge whether each cited page was correct.

The per-job pages are AI search visibility for the use case and the best SERP APIs if you also need the organic results.

Last checked 2026-10-24 against the 40 runs described above and the live looot catalog.