Onsite Listicles Still Work (If You Put in the Offsite Work)

AI Engines Answer Buying Questions From "Best of" Lists, and Companies Have Started Writing Their Own

Ask ChatGPT which CRM to buy, which accountant to use or which agency to hire, and there's rarely one right answer for it to look up. It does what a buyer would do on Google instead: it finds pages that already list the options and builds its answer from those.

Companies have noticed. Plenty of them now publish a "best of" list for their own market on their own site, usually with themselves in first place. Ranking Atlas has one too, so I wanted to know whether these lists work.

There's good research on the wider problem already. Kevin Indig's ghost citation analysis (Search Engine Journal, April 2026; 3,981 domain appearances across 115 prompts and four engines) found brands were cited in 74.9% of appearances and named in 38.3%. A Writesonic study reported on Search Engine Land put the share of citations that don't name the brand at about 40%, highest on Perplexity. Both show the same thing: an engine can use your page as a source and still leave your name out of the answer.

The closest study to this one is Lily Ray's. She looked at self-promotional "best of" lists on Google AI Overviews across 100 "best B2B software" searches. In June 2026, the brand that ranked itself first was left out of the recommendation 69% of the time. By September it was 83% (PPC Land, 5 October 2026).

Lily's study is the clearest evidence I'd seen that ranking yourself first on your own list doesn't win you the recommendation, and she showed it getting worse between June and September. It made me want to look a level deeper. Does the same thing happen on ChatGPT, Gemini and Perplexity? Does it hold in a different market? And when the company that wrote a list does get named, what's different about it?

To find out, I tested a market I know well: AI SEO agencies. Ranking Atlas's Search Console shows what buyers type when they're looking for one, so I took 22 questions from it word for word and wrote 4 more in the same form. I asked ChatGPT, Gemini and Perplexity all 26, twice each, on two dates between 5 and 7 October 2026. They gave 262 answers and cited 2,413 sources.

Agencies are the test case here. The question is the same for any company with a "best of" list on its own site. The full method, and what this data can't tell you, are at the end of the report.

In the Agency Market, Half the Sources the Engines Cited Were "Best Agencies" Lists, and Six in Ten of Those Were Written by Agencies

Of the 2,413 sources, 1,205 were "best agencies" lists. Of those, 745 went to a list on an agency's own website: an agency ranking agencies, usually with itself at the top. Of the 30 lists cited most often, 24 put the agency that wrote them in first place.

Table 1. Where the engines look for agency recommendations

Measure	Figure
Share of all citations that are "best of" lists	50% (1,205 of 2,413)
Answers citing at least one list	83% (217 of 262)
Lists cited from a company's own site	97 companies, 745 citations
Companies that put themselves first on their own list (30 most-cited lists)	24 of 30

Data in this table

Measure	Figure
Share of all citations that are "best of" lists	50% (1,205 of 2,413)
Answers citing at least one list	83% (217 of 262)
Lists cited from a company's own site	97 companies, 745 citations
Companies that put themselves first on their own list (30 most-cited lists)	24 of 30

Source: Ranking Atlas sample of 262 AI answers to 26 buyer questions, two runs per engine, 5 to 7 October 2026. The self-ranking count comes from reading the 30 most-cited lists on 5 October 2026.

In all, 97 agencies had their own list cited at least once, so writing a list does get your page in front of the engines. What I wanted to know was whether it gets your name into the answer.

An Engine That Cites Your Service Page Names You 86% of the Time; One That Cites Your Own List Names You 44% of the Time

When an engine cited an agency's own service page, the answer named that agency 241 times out of 280 (86%). When it cited the agency's own "best agencies" list, the answer named that agency 330 times out of 745 (44%).

Table 2. Cited is not the same as named

What the engine cited	Citations	Company named in the answer
The company's own service page	280	86%
A "best of" list on the company’s own site	745	44%

Data in this table

What the engine cited	Citations	Company named in the answer
The company's own service page	280	86%
A "best of" list on the company’s own site	745	44%

Source: Ranking Atlas sample, 262 answers, ChatGPT (gpt-5.5, web search), Gemini (gemini-2.5-flash, Google Search grounding), Perplexity (sonar), 5 to 7 October 2026.

More often than not, the engine read the list, took other agencies' names from it, and left out the agency that wrote it.

I think the reason is simple. Your service page is about you, so when ChatGPT cites it, it's going to say your name. Your list is about ten or fifteen companies. The engine picks the ones it wants for its answer, and on your own page you're one name among many.

My numbers point the same way as Lily's, on three other engines and in a different market: the agency that wrote the list was left out of 56% of the answers that cited it. If a tracker tells you your list was cited, don't count it as a recommendation. In more than half of those answers, the buyer read other companies' names and not yours.

Perplexity Cites the Most Lists and Names the Agency Behind Them Least

Perplexity cited agency-written lists 425 times and named the agency behind the list in 30% of those answers. ChatGPT cited 100 and named the author 65% of the time. Gemini cited 220 and named the author 62% of the time.

All three did far better when they cited an agency's own service page: ChatGPT named the agency 94% of the time, Gemini 84% and Perplexity 79%.

Table 3. Perplexity cites the list and leaves the host out

Engine	Own list cited	Company named	Own page cited, company named
ChatGPT	100	65%	94%
Gemini	220	62%	84%
Perplexity	425	30%	79%

Data in this table

Engine	Own list cited	Company named	Own page cited, company named
ChatGPT	100	65%	94%
Gemini	220	62%	84%
Perplexity	425	30%	79%

Scroll for more →

Source: Same 262 answers. Gemini's second-day run holds 2 answers because DataForSEO rate-limited the Gemini route on 7 October; its first-day run of 36 answers is complete.

The answer to my first question is mixed. The gap between service page and list showed up on all three engines. But on ChatGPT and Gemini, an agency's own list still got it named more often than not. On Perplexity it mostly didn't.

This data can't tell me why Perplexity behaves differently. What I can see is that it returned exactly ten sources with every answer in this sample, so lists end up in its sources even when the answer doesn't use the names on them. If your buyers use Perplexity, don't count on your list to get you named there. Your service page still worked: when Perplexity cited it, it named you 79% of the time.

Two Days Later, the Result Held

Before going further, I wanted to check this wasn't a fluke. I ran the same 26 questions again on 7 October and compared ChatGPT and Perplexity only, because Gemini's second run was cut short. The agency behind a cited list was named 37% of the time on both days. That's lower than the 44% above because Gemini, which names list authors more often, isn't in this comparison. A cited service page got its agency named 87% of the time on the first run and 86% on the second.

Table 4. Two days apart, the same answer

Run	Own list cited, company named	Own page cited, company named
5 to 6 October 2026	37% (95 of 259)	87% (103 of 118)
7 October 2026	37% (98 of 266)	86% (92 of 107)

Data in this table

Run	Own list cited, company named	Own page cited, company named
5 to 6 October 2026	37% (95 of 259)	87% (103 of 118)
7 October 2026	37% (98 of 266)	86% (92 of 107)

Scroll for more →

Source: Same prompts on 5 to 6 October and 7 October 2026; ChatGPT and Perplexity only, so both days compare like with like.

AI answers move around a lot. In a related sample of mine, only 55% of the websites cited in one run were cited again in a second run minutes later. In a study of how AI names experts in tech, changing only the wording of the question moved the result from 11.8% to 27.8%. Which pages get picked changes every time you ask. How the engines treated a list once they'd picked it stayed the same.

Agencies Already Named in Other Answers Get Named From Their Own Lists Three Times as Often as Agencies That Aren't

That leaves the question I most wanted to answer. Some agencies did get named when their own list was cited. What was different about them?

I took every agency whose list was cited three or more times and counted how many other answers named it without citing its list. That's my measure of offsite corroboration: how often the engines name an agency when its own list isn't in front of them. Then I checked how often each group got named when its list was cited.

Agencies that no other answer named got named from their own list 22% of the time. Agencies named in one to four other answers: 44%. Five or more: 68%.

Table 5. The more a company is named elsewhere, the more its own list gets it named

Company also named in other answers	Companies	Own list cited	Company named when its list is cited
Never	46	316	22%
1 to 4 answers	14	127	44%
5 or more answers	25	285	68%

Data in this table

Company also named in other answers	Companies	Own list cited	Company named when its list is cited
Never	46	316	22%
1 to 4 answers	14	127	44%
5 or more answers	25	285	68%

Scroll for more →

Source: Same 262 answers; companies whose list was cited three or more times. No company is identified; the bands carry the finding.

The table at the top of this report breaks the same groups down by engine, and the pattern holds on all three. ChatGPT named an agency from its own list 30% of the time when no other answer named it, and 94% when five or more did. Gemini went from 46% to 75%, and Perplexity from 13% to 58%.

The simplest reading is that the engines name companies they can back up from more than one source. Your list is one source, and you wrote it, so on its own it got an agency named about one time in five. Once other people name you too, on directories, review sites, other people's lists, press coverage and customer case studies, your list becomes one more page that agrees with what the engine has already seen.

I can't prove from this data that the coverage elsewhere causes the naming. I didn't measure other differences between the agencies, such as size or how long they've been trading, and those may play a part too. But the pattern held on all three engines, and it's what you'd expect if an engine checks a name against other sources before repeating it. That's the answer to my third question, and it's the finding I'd most like to see someone test in other markets.

What I'd Do If I Were Publishing a List

• Write it, and keep it indexed. Half the sources in this sample were lists, so a list is worth having. Check the URL in Search Console after every core update. If it drops out of the index, the engines that search the web can't find it.
• Write your own entry so it can be quoted as it stands. Your exact company name and one sentence on what you do and who it's for.
• Spend more time getting named on pages you don't own than on the list itself. Other people's lists, directories, review sites, press coverage, customer case studies. Writing a list is quick. Getting named elsewhere is slow, and in this data it's what made the difference.
• Track "named" and "cited" as two numbers. If your tracker only counts citations, it will tell you the list is working while the engines read it and name someone else.
• Treat Perplexity separately. If your buyers use it, your service page does more for you there than any list.

“Write the list. It gets your site into the pages the engines read when a buyer asks what to choose. Then spend the next few months getting named on sites you don't own. In this data, that was what moved a company from cited to named.”

Daniel Grainger, Ranking Atlas

Indexing Is the Aim, Corroboration Is the Game

If you want an AI engine to name your company, two things need to happen, and they're different jobs.

First, the pages that mention you have to be in the search indexes the engines use, and stay there through updates. That covers your own site and other people's. Your list is part of this job, and a useful part: the engines find it, cite it and build answers from it.

Second, other people have to name you on pages you don't control, which is how you build citation equity. In this data, that's what separated the agencies that got named from their own list from the ones that didn't, and your own list can't do it for you.

Do them in that order. Coverage elsewhere doesn't help if the engines can't find the pages it's on. And indexed pages with no coverage elsewhere get you cited, but in this sample they got a company named only about one time in five.

Lily's numbers come from B2B software brands on Google, and mine from agencies on three other engines. In both, the company that wrote the list was left out more often than not. I haven't tested the corroboration part outside agencies yet, but if you sell anything buyers compare on "best of" lists, I'd plan as if it applies to you.

Download the Data

The tables in this report and the 26 questions, as two CSV files. I've left out the agency-by-agency rows (whose list was cited, and how often they got named), because the point is the pattern, not a league table of named companies.

Citation: Ranking Atlas, "Onsite Listicles Still Work (If You Put in the Offsite Work): data analysis of 745 listicle citations across ChatGPT, Gemini and Perplexity," October 2026. Research by Daniel Grainger.