We asked Google nine buyer-intent questions about rank tracking tools on 15 August 2026 and kept every source URL attached to the answers. Eight questions returned an AI Overview, filling 70 citation slots across 44 distinct URLs. Not one of those 44 was a homepage. Forty-nine of the 70 slots went to a title containing the word "best", 36 to a title containing a year, and 11 to Reddit, YouTube or Quora.
This is a description of a cited set, not a formula. Eight questions on one topic on one day cannot tell you what causes a citation, and this post does not claim to. What it can do is show you, with real URLs, what the pages Google actually pulled looked like at the page level. Everything here is Google AI Overviews. Not ChatGPT, not Perplexity, not Claude.
The short answer
- 8 answered questions, 70 citation slots, 44 distinct URLs.
- 0 of 44 were homepages. Every cited URL was a specific interior page.
- 53 of 70 slots went to a "best" or "top" title; 36 carried a year; 31 opened with a number.
- 11 of 70 slots went to Reddit, YouTube or Quora.
- Four URLs took 20 of the 70 slots. The same pages get pulled into answer after answer.
The method, briefly
Nine buyer-intent templates, asked live through Bright Data's SERP API with gl=us and hl=en, at $0.0015 a question: best X, what is the best X, best X for small business, how to choose X, top X compared, which X should i use, best X for beginners, cheapest X, is X worth it. Eight of the nine calls completed. One failed at the transport layer and is excluded. The full walkthrough, including why a question with no AI Overview is excluded rather than counted as a miss, is in how to measure AI visibility.
Nobody's homepage was cited
The clearest single result in the set, and the easiest to act on.
All 44 distinct cited URLs were interior pages. Forty-one were articles, guides or forum threads. The three exceptions were product and tool pages, and the interesting part is which questions pulled them in:
seranking.com/position-tracking.html cited on "how to choose rank tracking tool"
seobility.net/en/rankingcheck/ cited on "how to choose rank tracking tool"
marketingtracer.com/rank-tracker/cheap-rank-tracker cited on "cheapest rank tracking tool"
That last one is worth staring at. The question was "cheapest rank tracking tool" and the cited page is literally titled "Cheap rank tracker". A product page can be cited, but in this sample it happened when the page answered the specific qualifier in the question rather than describing the product in general.
If your plan for AI visibility is to make your homepage more persuasive, this is the result that should redirect it. The homepage was not the unit of retrieval for a single one of these eight answers.
The titles
Counting across all 70 citation slots:
| Pattern in the cited title | Slots (of 70) |
|---|---|
| Contains "best" | 49 |
| Contains "best" or "top" | 53 |
| Contains a year (2025 or 2026) | 36 |
| Opens with a number | 31 |
Real examples, exactly as they came back in the references[] array:
15 Best Keyword Tracking Tools For 2026: AI & Pixel Rank
15 Best Rank Tracking Software (Tested on Real Campaigns)
10 Best Rank Trackers for SEO in 2026 (Ranked and Reviewed)
28 Best SEO Rank Trackers in 2025
The 12 best rank tracker tools
The reading that is available here is modest and the reading that is tempting is wrong.
The tempting one is "write listicles with the year in the title". That does not follow. On a query set built entirely from best X and top X compared, a corpus of "N best X in 2026" articles is what exists to be cited. The sample is measuring the content market for these questions, not Google's preference.
What does transfer is the property underneath the format. Every one of those pages contains an itemised, comparable set with an explicit answer to the question as it was asked. A synthesised answer needs items it can name, compare and attribute. A page that argues its way to a conclusion in prose gives the system nothing quotable at the granularity it is working in. That is a structural observation about extractability, and it holds regardless of whether you number your headings.
The same pages, over and over
Citation is not spread evenly. Four URLs took 20 of the 70 slots:
7 of 8 answers yotpo.com/blog/best-keyword-tracking-tools/
5 of 8 answers reddit.com/r/SEO/comments/1b3q9a4/what_are_the_best_rank_tracker_tools/
4 of 8 answers onelittleweb.com/top-tools/best-rank-tracking-software/
4 of 8 answers tryanalyze.ai/blog/best-rank-tracking-tools
One URL was cited in seven of the eight answers. Nine other questions could be asked of this topic, and that page would probably be in most of them. It is the page the system has decided answers this category.
For a small site the consequence is not "produce more pages". It is that a single page which becomes the canonical answer for a category is worth more than ten pages that are each somebody's fourth choice. It also means the measurement is stable enough to be useful. If a page like that appears in your citation list, it will keep appearing, and the week it stops is a real signal rather than noise.
Reddit, YouTube and Quora took 11 of 70 slots
Community and video sources are not a footnote in this sample.
5 slots reddit.com/r/SEO/comments/1b3q9a4/what_are_the_best_rank_tracker_tools/
2 slots reddit.com/r/SEO/comments/1ducj5j/any_accurate_rank_tracking_tool/
1 slot youtube.com/watch?v=WgLj-Mt8fn4
1 slot youtube.com/watch?v=_-2VXX0X96g
1 slot youtube.com/watch?v=FgkMGO8BdYI
1 slot quora.com/Are-paid-rank-tracking-tools-worthwhile-for-owners-of-one-website...
Reddit appeared in seven of the eight answers, more often than it appeared in the organic results for those same queries, which we broke down in ranking first does not mean cited. Both Reddit threads have titles that are questions, phrased almost exactly as the query was phrased.
The uncomfortable implication for a vendor is that one of the most reliably cited pages about your category is a forum thread you do not control and cannot edit. The useful implication is that this is measurable. You can see it in your own scan rather than guess at it.
The citation attaches to a claim, not to the page
One more detail from the API response, because it changes how to think about the unit of work.
The AI Overview object carries a references[] array for the answer as a whole, and the answer text itself carries links on specific spans inside sentences. In a separate single call on the same day, for "best ai visibility tools", the first paragraph named several products. It attached a link to the individual product names inside the sentence, one of them pointing at tryprofound.com/blog/choosing-ai-visibility-provider.
So there are two things happening. The page gets into the source list, and a specific claim inside the answer gets attributed to a specific URL. The second one is the interesting target, and it is a sentence-level property: a clear, self-contained, attributable statement that is true and checkable on your page.
Two things follow. The paragraph that answers the question has to be able to stand on its own, because it may be lifted away from everything around it. And the page has to be readable as text, since a claim that only exists in an image, a chart or a client-rendered component cannot be attributed. Our free AI content extractor shows what is actually in the HTML rather than what your browser paints, and the answer readiness checker tests whether the page clears Google's eligibility floor, which is that it must be indexed and eligible to show a snippet (Google Search Central).
What this does not tell you
Said plainly, in the same section as the finding.
Eight answered questions on one topic, one country, one language, one day, through one provider. It is a probe.
Nothing here is causal. We can see that 53 of 70 cited titles said "best" or "top". We cannot see whether that made any difference, because we have no comparison group of uncited pages for the same queries. A trait shared by cited pages is only a factor if uncited pages lack it, and that check is not in this data.
It also says nothing about a different query shape. Every question in the set was commercial. Informational, navigational and news queries will produce a different cited set, and possibly no AI Overview at all.
What to do at the page level
- Point at a page, not a homepage. Every cited URL in this sample was an interior page that answered a specific question. If your best answer to "how to choose X" lives on your homepage, it does not exist for this purpose.
- Answer the qualifier literally. "Cheapest", "for agencies", "for beginners" are separate questions and in our sample they produced substantially different citation lists. The page that won "cheapest" was the one whose title said cheap.
- Give the answer an itemised, comparable shape. Named things, with the differences between them stated. A synthesised answer needs pieces it can name and attribute, and that is the reason, rather than anything magic about list format.
- Make the key claim self-contained. Assume the paragraph will be read without the two around it.
- Make sure it is text in the HTML. Check with the AI content extractor rather than your browser. If you are auditing markup at the same time, our structured data checker covers what still earns something, and which structured data still matters explains why the list is shorter than it used to be.
- Then measure, on your own questions. All of the above is a hypothesis until your own citation list moves. How to measure AI visibility is the method, and what is actually known about how AI Overviews pick sources is the line between what Google has published and what people have assumed.
The scan that produced everything on this page cost $0.0135. Run the same nine questions against your own category. That citation list is the one worth acting on.