Key findings
What we found
- The citation panel is almost always exactly three sources. 259 of 290 AI Overviews showed three references — no more, no fewer.
- Vendor blogs and listicles are more than half of everything cited. They take 53% of top-3 citations — more than three times the share of any other source type.
- Reddit holds the opening slot in about a third of all answers: 33 first-position citations across ~97 panels, more than any other domain.
- Review aggregators are effectively absent. 89 of ~97 searches have no G2, Capterra or TrustRadius citation in the top 3 — and Forrester and IDC were cited zero times.
- YouTube has huge reach but little prime position: cited somewhere in 68 of ~97 panels, yet inside the top 3 for only 20.
- And audience size is a weak gate — the smallest video holding a top-3 citation had 87 views.
What gets cited, by source type
| Source type | Citations per pull | Share of top 3 |
|---|---|---|
| Vendor blog or listicle Round-ups and “best X” posts published by software vendors. | 148 | 53% |
| Forum or community Almost entirely Reddit, with a handful of Quora threads. | 42 | 15% |
| Vendor product page A vendor’s own homepage, product page or category page. | 24 | 8% |
| Video YouTube — reviews, walkthroughs and “I tested N tools” comparisons. | 22 | 8% |
| News or media Mainstream and trade press: PCMag, Forbes, TechRadar, CNBC. | 19 | 7% |
| Analyst Gartner only. Forrester and IDC were not cited once. | 15 | 5% |
| Review aggregator G2, Capterra, TrustRadius, Software Advice and peers. | 7 | 3% |
| Professional, org or gov Trade bodies, professional associations and government sites. | 4 | 1% |
| Social or pro network LinkedIn, Medium and similar. | 1 | 0% |
Based on 282 top-3 citations per pull across ~97 searches that returned an AI Overview. Figures are the mean of three pulls.
The shape of the panel
Three slots, and only three
Google shows a near-fixed set of three references beside its AI Overview. That is the entire shortlist a buyer sees, which makes the third slot the whole game.
3
References shown per panel
Median across every search that returned an AI Overview.
6.4
Distinct sources per search
Counting the per-section citations inside the answer as well as the visible list.
97/100
Searches with an AI Overview
Per pull. All 100 returned one on at least one of the three pulls.
How many references each panel showed
- 0 refs4
- 2 refs12
- 3 refs259
- 4 refs9
- 5 refs2
- 6 refs1
- 17 refs3
Three is an editorial choice, not a technical limit. One search — best product analytics software — returned 17 references on all three pulls. For everything else, Google settles on three. The panel is also not fixed in time: 41% of these searches returned a different top 3 when we simply asked again, so treat any single snapshot, including this one, as having an error bar.
The video signal
YouTube is everywhere, and cheap to enter
YouTube is the most-cited domain after Reddit, appearing somewhere in roughly 68 of ~97 panels. We profiled all 174 cited videos across 119 channels to see what separates the ones Google puts in the top three from the ones it buries.
4.1×
View advantage
Median 4,476 views for videos reaching the top 3, against 1,093 for those cited further down.
87 views
Smallest video in a top-3 slot
8 of 31 top-3 videos have under 1,000 views. Audience size is a weak gate.
2.2×
Subscriber advantage
Median 25,600 subscribers in the top 3 against 11,700 below it — a weaker signal than views.
The channels supplying the most cited videos
| Channel | Videos cited | Median views |
|---|---|---|
| CodeLucky | 8 | 224 |
| Business Solution | 8 | 1,517 |
| OneHourProfessor | 5 | 1,422 |
| Software Connect | 5 | 17,246 |
| Team Collaboration Solution | 4 | 96 |
| HR Toolbox | 4 | 6,665 |
| Uplatz | 3 | 57 |
Several of the most-cited channels are high-volume uploaders whose videos average only a few hundred views. Publishing a dedicated video per software category appears to matter more than any single video performing well. 18 of 174 cited videos are Shorts.
Full dataset
Explore the data
Every search, every cited domain and every cited video. Find your category and see exactly what Google put in front of buyers.
| Top 3 cited | Source types | Repeatable |
|---|
No searches match these filters.
| Source type |
|---|
No domains match these filters.
| Reached top 3 |
|---|
No videos match these filters.
Free content audit
Want to know what AI cites for your own keywords?
Three slots per search is a brutal shortlist. We’ll check which sources Google’s AI Overview cites for the searches that matter to your product — including the Reddit threads and vendor listicles doing the work — and show you where your content can realistically earn one.
How we ran it
Method and limitations
- 100 “best <software>” keywords with real US search volume, spanning sales, marketing, HR, finance, security, dev tools, AI/ML and vertical SaaS.
- Pulled through DataForSEO’s SERP API (
/v3/serp/google/organic/live/advancedwithload_async_ai_overview=true) on 19 October 2026, US, English, desktop. - Each search was pulled three times. Every figure is the mean across those three pulls, so one unlucky snapshot cannot drive a headline.
- “Top 3” means the first three entries of the AI Overview’s reference list, in the order Google returns them, not de-duplicated.
- Citations appear in two places in Google’s response: a short top-level reference list and per-section lists inside the answer. The top-3 analysis uses the top-level list; the distinct-sources figure counts unique domains across both.
- Source types are assigned by rule from the domain, with vendor pages split into listicle/blog and product/category by URL and title. Checked at 97.6% agreement against a manually labelled set.
- YouTube metrics come from DataForSEO’s YouTube API and are point-in-time as of 19 October, not as of the moment each video was cited.
- Limitations: US desktop only, and a single day. AI Overviews are non-deterministic — 41% of these searches returned a different top 3 across our own three pulls, so treat any single snapshot as having an error bar.
This repeats the design of our earlier work on AI Overview citations: AI Overview citations for “best product” searches and how YouTube affects AI Overviews for SaaS.