We checked 17,651 buyer answers for pages on the vendors' own sites
The sample is every answer Proofsource collected from October 4 to 23, 2026, for a buyer question (one that doesn't name the brand) on ChatGPT, Perplexity, Claude and Google AI Overviews: 17,651 answers for 115 brands in 76 workspaces.
For each brand we know its own site and the sites of the competitors it tracks. So for every cited page we can tell whether it sits on one of those vendors' sites, and whether that vendor was named in the same answer.
We called a page a best-of list when its address carries "best" or "top", and a head-to-head page when it carries "vs", "versus" or "alternatives". Of 45,576 cited pages of either kind, 3,287 (7.2%) were on a tracked vendor's own site. The rest belonged to publishers, review sites and vendors nobody in the workspace tracked.
38% of answers that cited a vendor's own best-of list left the vendor out
A vendor's own best-of list was cited in 1,590 answer and vendor pairs, covering 242 vendors and 610 questions. The answer named that vendor in 982. In the other 608, 38.2% (95% interval 35.9% to 40.7%), it didn't. In 564 of those, the vendor's name appears nowhere in the answer text.
Other pages did better. When the cited page on the vendor's site was a product page, a doc, a homepage or anything else that wasn't a comparison, the vendor was left out 21.4% of the time (2,672 of 12,488). Head-to-head and alternatives pages sat in between at 23.6% (223 of 945). When none of the vendor's pages was cited at all, it was named in only 11.3% of answers.
On the same question, the list did no better and no worse
That gap is mostly about which questions a best-of list gets pulled into. We took the 236 cases where the same engine answered the same question, citing the vendor's best-of list on some days and another page of theirs on other days. Averaged over those cases, the vendor was named 78.8% of the time with the list and 80.0% with the other page.
So the list wasn't punished, and it wasn't rewarded either. It gets a vendor's site read on broad category questions its product pages never reach, and on those questions the engine builds the shortlist from other evidence. The list supplies names. It doesn't decide which ones make it into the answer.
Perplexity and Claude dropped the publisher most often
Share of answers that cited the vendor's own best-of list and left the vendor out, with the same figure for its other non-comparison pages:
- Perplexity: 44.5% (293 of 659), against 35.0% for other pages.
- Claude: 44.0% (234 of 532), against 38.2%.
- Google AI Overviews: 23.7% (46 of 194), against 14.9%.
- ChatGPT: 17.1% (35 of 205), against 8.1%.
Earlier studies counted the citation, not the answer
Peec tracked 232,000 citations on software review prompts from December 2025 to February 2026 and found about 11% came from self-promotional listicles, 3.6% on ChatGPT against roughly 10% on Google AI Mode and Perplexity. Ahrefs found that when ChatGPT recommended a software brand, that brand's own blog list was among the sources in 34% of responses.
Both start from the citation or from the recommended brand. Neither reports how often the list was cited and its publisher went unnamed, which is the case a vendor paying for the page cares about.
What to check before you write another list
- Count the answers that cited your list and named you, not the citations. On our sample, more than one in three didn't.
- Compare against your other pages on the same questions. If the list doesn't beat them there, it's buying reach on questions where you weren't a pick anyway.
- Look at who the engine named instead. Those are the companies your own page is helping.
How Proofsource handles it
Proofsource keeps the full text and cited pages of every answer, along with the tracked brands it named. You can open the answers that cited your own pages and see, answer by answer, whether you were in them and who was.
Where this could be wrong
Page types come from words in the address, not from reading the page. A list without "best" or "top" in its address is missed, and some addresses with those words aren't lists. We also don't know whether the vendor ranked itself first, which is how Peec defines a self-promotional list; some of these are retailers' gear guides.
Vendors are the brand and the competitors it tracks. Lists on untracked vendors' sites count as third party here.
Names come from our own extraction. Text matching agrees on the large majority: the vendor's name was absent from 564 of the 608 answers we scored as leaving it out.
The sample is 20 days, four engines and the questions our users chose to track, mostly software. The paired comparison rests on 236 cases, and none of this is an experiment: the engine chose which page to cite.
