TL;DR
- A vendor's own best X list gets cited by AI engines but rarely makes that vendor the pick.
- Vendors ranked themselves first on 81.8% of the 462 vendor lists the engines cited.
- When an answer cited one of these lists, it left the vendor out 46.2% of the time.
- A first place on another site's list led to a recommendation 85.7% of the time and a first pick 35.2% of the time.
- Niche brands gain the most from their own list, which lifted their inclusion from 4.5% to 26.1%.
A “best X” list that ranks your own product first will get cited by AI engines, and it will not get you recommended nearly as often as a #1 spot on someone else’s list. Across 1,920 answers from ChatGPT, Microsoft Copilot, Google Gemini and Google AI Mode, engines cited a vendor’s own self-ranked list 775 times. In 46.2% of those cases the answer left the vendor out entirely.
Self-promotional listicles became one of the most copied tactics in generative engine optimization this year. The logic is simple: engines lean on list pages, so publish your own list and put yourself on top. We took the 8,272 citations behind our 1,920-answer engine comparison, fetched the cited pages, and checked who wrote each list and where it ranked its author. Then we checked whether the engine actually recommended that author.
Vendors rank themselves first on 82% of their own lists
Half of all citations (50.9%) in these answers pointed to list pages. Of the 2,097 list pages cited, 462 were written by a company that sells a product in the same category. Those vendors ranked themselves #1 on 378 of them, or 81.8%. Another 37 put themselves second or third. Only 10 left themselves off their own list.
The format is not limited to big names. 142 of the 378 self-ranked lists came from vendors that none of the four engines recommended in any of the 12 answers for that prompt. Their list got pulled into the answer as a source, and their product still never made the shortlist.
Gemini leans on vendor lists the most
The engines use these pages very differently. Gemini cited a self-ranked vendor list in 43.8% of its answers, and nearly one in four of its citations (23.3%) went to one. Copilot cited them least, in 16.5% of answers. It leans on directory and aggregator sites instead, such as toolradar.com, which it cited 195 times. ChatGPT sits in between, but it cites vendors’ own pages of every kind more than any other engine: 49.1% of its citations went to a page published by a brand in the category.
| Engine | Answers citing a self-ranked vendor list | Share of citations to these lists | Vendor left out when its list is cited |
|---|---|---|---|
| Gemini | 43.8% | 23.3% | 46.6% |
| Google AI Mode | 28.5% | 7.4% | 57.5% |
| ChatGPT | 19.2% | 8.5% | 23.3% |
| Microsoft Copilot | 16.5% | 4.4% | 46.5% |
| All four | 27.0% | 9.4% | 46.2% |
The engine cites the list, then skips its author
When an answer cited a vendor’s self-ranked list, it recommended that vendor 53.8% of the time. It made the vendor its first pick 10.9% of the time. When the vendor did make the answer, it landed at position 3.8 on average, while its own list had it at 1.
A list from anyone else that ranked a brand first did much more for that brand. Across 2,342 such cases, the engine recommended the brand 85.7% of the time and put it first 35.2% of the time, at an average position of 2.6. Engines read vendor lists for the category overview and the competitor names. They mostly don’t carry over the vendor’s ranking of itself.
Google AI Mode was the least willing to follow a vendor’s list, leaving the author out 57.5% of the time. ChatGPT was the most willing, at 23.3%. Google’s two engines pull in opposite directions here: Gemini cites vendor lists the most, and AI Mode skips their authors the most. Our comparison of AI Overviews and AI Mode found a similar split in the sources Google cites for the same query.
Your own list mostly helps small brands get named
Pooled numbers mix big brands that would be recommended anyway with small ones that would not. To separate them, we compared the three reruns of each prompt on the same engine. Answers vary from run to run, as our ChatGPT citation variance study showed, so some runs cite a given page and others don’t. For each brand, we compared runs that cited its self-ranked list with runs that cited nothing the brand owns.
Across 298 of these brand, engine and prompt combinations, runs that cited the brand’s own list recommended it 45.1% of the time, against 22.3% for runs that didn’t. Runs that cited another site’s list ranking the brand first recommended it 82.5% of the time, against 57.9% without it.
The gap is widest for niche brands, the ones named in two or fewer of the nine answers from the other three engines. A niche brand’s own list lifted its inclusion from 4.5% to 26.1%. A third-party list ranking it first lifted inclusion from 21.3% to 66.9%. Neither got niche brands picked first often: 1.9% with their own list, 7.6% with a third-party #1. For established brands the difference mostly disappears, since they make the answer about 80% of the time regardless.
Eight lists, eight outcomes
Each prompt got 12 answers: three runs on each of four engines. These self-ranked lists were among the most cited. Some turned every citation into a recommendation, and some turned none.
| Vendor list | Prompt | Answers citing it | Answers recommending the vendor |
|---|---|---|---|
| Buffer | best social media management tools | 7 | 7 |
| UptimeRobot | best uptime monitoring tools | 7 | 7 |
| LingQ | best language learning app | 7 | 7 |
| Artisan | best sales engagement platform | 7 | 7 |
| Kixie | best sales engagement platform | 7 | 1 |
| Airwallex | best bank for small business | 6 | 0 |
| Cube | best dashboard software | 5 | 0 |
| Parseable | best log management tools | 5 | 0 |
The four lists that converted fall into two groups. Buffer and UptimeRobot were recommended in all 12 answers for their prompt, whether their list was cited or not. Artisan and LingQ were recommended in exactly the 7 answers that cited their list and in none of the other 5. For those two, the list is the only visible route into the answer.
Kixie, Airwallex, Cube and Parseable got cited again and again and almost never named. On the small-business banking prompt, the engines settled on Chase, Bluevine and Bank of America, whatever Airwallex’s page said.
What this means for brands
- Publish the list for the citation, not for the ranking. A vendor list gets cited readily, and the engine takes the category and competitor names from it. Expect your own #1 to count for little.
- Put your effort into third-party lists. A #1 spot on someone else’s list went with a recommendation 85.7% of the time and a first pick 35.2% of the time. The same spot on your own list returned 53.8% and 10.9%.
- Small brands have the most to gain. A niche brand’s own list moved it from 4.5% to 26.1% inclusion. That is the one group where the tactic clearly pays.
- Check each engine separately. Gemini cites vendor lists in 43.8% of answers, and AI Mode skips the author 57.5% of the time. A single blended number hides both. A manual AI visibility audit or one of the AI visibility tools will show where you stand on each.
- Treat the citation as the win. Engines cited self-ranked vendor lists 775 times, and a cited page puts your domain in front of the reader even when the answer names someone else. A comparison page built for AI search earns that citation whether or not the engine agrees with your ranking.
This connects to what we found about AI Overview citations on “best X” queries and about ChatGPT’s search fan-out. Engines search wide and cite many list pages, then build their own shortlist. For the wider picture of how these engines pick sources, see our state of AI search report.
How the study was run
- Answers: 160 “best X” buyer prompts across software, consumer products, developer tools and business services, each sent three times to ChatGPT, Microsoft Copilot, Google Gemini and Google AI Mode from the US in September 2026. 1,920 answers, with the recommended brands extracted from each.
- Citations: 8,272 cited URLs across those answers, 4,180 unique. YouTube, Reddit and similar sites were typed by domain. The rest were fetched.
- List pages: a page counts as a list when its headings name at least three brands from the prompt’s category. Its ranking is the order those headings name them.
- Vendor lists: a list published by a company that sells a product in the category. Publishers were matched to brands by domain and site name, and publishers that no engine recommended were checked by hand. The classification was also checked on a random sample of pages.
- Rerun comparison: for each brand, engine and prompt, runs that cited the brand’s page were compared with runs that cited nothing the brand owns. Brand tiers use how many of the nine answers from the other three engines named the brand.
- Limits: engines search, cite and recommend in one pass, so these numbers show what goes together, not proof that a page caused a recommendation. The rerun comparison removes differences between prompts and engines but not that.