This report covers one category, AI visibility and answer-tracking software, measured against one brand in that category. That is a narrow base, and we say so up front rather than in a footnote. What makes it worth publishing is the verification step: we did not infer presence from the answers, we opened the cited pages and read them.
Measured 20 August – 11 September 2026 across ChatGPT, Perplexity, Gemini, Claude, Grok and Google AI Overviews.
What the engines actually cited
Across 23 scans, the six engines returned source URLs resolving to 25 distinct third-party hosts. Grouped by what kind of page they are, the evidence set is not balanced — it is dominated by other people's comparison content.
| Source type | Hosts | Citations | Share | What these pages are |
|---|---|---|---|---|
| Roundups & comparisons | 18 | 1,061 | 51.5% | "Best X tools", "alternatives to Y" listicles on independent blogs |
| Community | 3 | 599 | 29.1% | Reddit threads, LinkedIn posts, Quora answers |
| Video | 1 | 288 | 14.0% | YouTube, including Shorts |
| Independent articles | 3 | 113 | 5.5% | Single explainers, not listicles |
Two of those rows deserve attention because they are where budget rarely goes. A single host, Reddit, was cited 411 times — about twice the most-cited dedicated roundup site, and nearly four times the vendor's entire own-domain total. YouTube supplied 288 citations on its own, 14% of the third-party evidence, from one host. Neither is a place most software companies treat as a distribution channel for buying-stage content.
The twelve hosts that carry the category
| Host | Type | Citations | Engines | Brand on page? |
|---|---|---|---|---|
| reddit.com | Community | 411 | 5 | Absent |
| youtube.com | Video | 288 | 4 | Absent |
| getairefs.com | Roundup | 201 | 6 | Absent |
| linkedin.com | Community | 139 | 4 | Absent |
| integrate.io | Roundup | 93 | 3 | Absent |
| therankmasters.com | Roundup | 82 | 5 | Absent |
| tryanalyze.ai | Roundup | 79 | 4 | Absent |
| get-ryze.ai | Roundup | 79 | 4 | Not read |
| seoturtle.com | Roundup | 68 | 3 | Absent |
| getpassionfruit.com | Article | 58 | 4 | Absent |
| shanikaw.medium.com | Roundup | 53 | 5 | Absent |
| chosenly.com | Roundup | 50 | 3 | Not read |
Only one host in the whole set — getairefs.com — was cited by all six engines. It is a blog most people in this category have never heard of, and it outranks every vendor's own domain as a source of evidence about that category's products.
The engines do not behave alike
Averaging the six engines together hides the most actionable finding in the dataset. They differ enormously in how much outside material they will pull in.
| Engine | Hosts cited | Of 25 | Reading |
|---|---|---|---|
| Perplexity | 23 | 92% | Most open. Third-party placement moves it fastest. |
| Claude | 22 | 88% | Nearly as broad; leans on roundups and community alike. |
| Google AI Overviews | 17 | 68% | Broad, but weighted to established publishers. |
| Grok | 15 | 60% | Mid-range; noticeably fond of community threads. |
| Gemini | 8 | 32% | Narrow. Concentrates on a handful of large hosts. |
| ChatGPT | 3 | 12% | Most closed. Third-party work barely reaches it. |
This matters for sequencing. If the goal is to be cited in Perplexity or Claude, getting onto the roundups and threads in the table above is the direct route, and it is measurable within weeks. If the goal is ChatGPT specifically, that route is mostly closed — it cited three outside hosts in three weeks — and the work is slower and more structural. Selling those two as the same engagement is how AEO retainers lose money; the agency playbook covers how to scope them separately.
What the vendor's own pages achieved
The vendor publishes steadily on its own domain, so it is fair to ask what that produced. Across the same period the engines cited the vendor's own pages 106 times, against 2,061 for third-party hosts.
| Page format | Citations | Share of own | Note |
|---|---|---|---|
| Comparison pages /compare/* | 58 | 55% | "X alternatives", head-to-head pages |
| Homepage | 18 | 17% | Cited as an entity reference |
| Pricing page | 11 | 10% | Engines quote published prices directly |
| Guide articles /learn/* | 14 | 13% | The largest library on the site, the smallest return |
| Product docs | 5 | 5% | Developer-facing pages |
The guide row is the expensive lesson. The guide library is the biggest content investment on the vendor's site, produced steadily and to a consistent standard, and it returned 14 citations — 0.7% of the total evidence set. The comparison pages, a much smaller and less loved body of work, returned four times as many. Engines quote pages that resolve a decision, not pages that explain a topic. It is the same split between ranking and being quoted that runs through From SEO to AEO.
What this means for sequencing
On this evidence, the ranking of effort for a company trying to enter generated answers in a competitive software category is: get onto the third-party roundups engines already read; publish comparison and pricing pages that state real figures; participate honestly in the community threads that carry 29% of the evidence; and only then expand the owned guide library. The vendor measured here had that order close to backwards.
Method
Anyone can repeat this with a tracker that stores the source URLs behind each answer; AI visibility tools compared sets the main trackers side by side.
- Fixed question set. 22 buying-stage prompts a real purchaser would type — "best AI visibility tracking tool", "which tool includes Google Search Console", "how much do AI visibility tools cost", and named-competitor alternative queries. The set was fixed before measurement and not tuned afterwards.
- Daily runs across six engines. ChatGPT, Perplexity, Gemini, Claude, Grok and Google AI Overviews, 23 scans between 20 August and 11 September 2026, using a commercial AI visibility tracker running daily scans. Every answer's source URLs were stored.
- Roll up by host. URLs grouped by host and counted. Rival-owned domains and engine-owned properties were excluded, so a mention on a competitor's own blog does not inflate the set.
- Open and read every cited page. The verification step. Each host's cited pages were fetched and read in full to determine whether the measured brand appears. Results recorded as present, mentioned-without-link, absent, or not-yet-read. Only the third counts as a gap.
- Report the unread separately. Five hosts had not been fetched at cut-off. They are excluded from the gap count and reported on their own so the headline figure is not inflated.
Questions people ask
Where do AI engines get the information in their answers?
In our measurement of one software category across 23 daily scans, six engines drew on 25 distinct third-party hosts, citing them 2,061 times, against 106 citations of the vendor's own website. Third-party roundups and comparison articles supplied 51.5% of those citations, community platforms 29.1%, video 14.0% and independent articles 5.5%. The vendor's own domain accounted for 4.9% of the total evidence set.
Does publishing more content on your own site get you cited by ChatGPT?
Not directly, on this evidence. The vendor's own pages made up 4.9% of the citations in the category we measured. Within that small share, comparison pages were cited 58 times and long-form guide articles 14 times, despite the guide library being far larger. Publishing volume on an owned domain was the weakest lever we measured; presence on the third-party pages engines already read was the strongest.
Which AI engine cites the most third-party sources?
Perplexity, by a wide margin. It cited 23 of the 25 hosts we tracked (92%), followed by Claude at 22 (88%), Google AI Overviews at 17 (68%), Grok at 15 (60%) and Gemini at 8 (32%). ChatGPT cited just 3 of the 25 hosts (12%), making it the most closed of the six engines on these questions and the hardest to influence through third-party placement.
How much do AI answers rely on Reddit?
More than on any other single host in our measurement. Reddit was cited 411 times, roughly twice the most-cited dedicated roundup site and nearly four times the vendor's entire own-domain total of 106. Reddit, LinkedIn and Quora together supplied 599 citations, 29.1% of all third-party evidence. Reddit appeared in answers from five of the six engines.
How do you find out which pages AI engines cite about your category?
Run the same buyer questions repeatedly across every engine, store the source URLs each answer returns, and roll those URLs up by host. The step most teams skip is verification: open each cited page and read it to confirm whether your brand is actually on it. In our set, 20 of 25 hosts were verified absent by reading the page in full. Five had not yet been read, which is not evidence of absence and is reported separately here.
We run this measurement for clients
The useful output is not the score. It is the list of specific third-party pages that already decide your category's answers, and whether you are on them. That list is different for every business, and it is the part you can act on this quarter.
Get a free visibility check