# 72 Statistics on Which Pages AI Search Cites When Buyers Choose

> Are listicles, best-of list pages, the most cited content type in AI search? Lists sit under most AI answers when buyers are choosing and almost none when they are learning, and someone else's list didn't get the company named.

By Vlad Shvets, Partner @ Qvery · Published 2026-03-29 · Updated 2026-11-09 · 11 min read
Canonical: https://qvery.ai/blog/listicles-most-cited-content-type-ai-search/

Somebody has put a proposal on your desk for twenty "best [your category]" pages, and the slide that justifies the budget says listicles are the most cited content type in AI search.

If you run a company of a few hundred people, ask it what you would ask of any percentage on a slide: most cited of what, and what does a citation get the company?

An answer that cited only other people's lists named the company it was asked about no more often than an answer that cited no list at all. When the list was the company's own, the answer named it about four times in five. These are pages the answers cited and companies they named, not purchases.

> 📊 **Highlights** 
> - **52.32%** of ChatGPT's sourced answers, the ones that showed their sources, and **76.91%** of Google AI Mode's cited at least one list (July 2026).
> - On ChatGPT, product and service pages outnumber lists about **two and a half to one** in the sources.
> - Buying questions cited a list **about ten times as often** as how-it-works questions on ChatGPT (60.27% against 6.22%).
> - With only other people's lists in the sources, ChatGPT named the company **25.85%** of the time; with no list, **25.28%**; with its own list, **77.80%**.
> - Only **5.2%** of ChatGPT conversation sessions produce an outbound click.[^1]

## Lists Lead Google AI Mode's Sources and Trail Product Pages on ChatGPT

From the GEO Agent's daily records, July 2026: each page cited in the answers to the buyer questions Qvery's GEO Agent asks ChatGPT and Google AI Mode, Google's chat-style search, every day carries a tag for its kind: a list, a product or service page, a guide, or a discussion.

On Google AI Mode, lists were the largest kind of cited page, **32.16%** of links, ahead of product and service pages (25.34%) and guides (20.81%).

On ChatGPT, product and service pages took **50.33%** of links and lists **20.52%**. A company's own product, pricing and home pages now outnumber lists in ChatGPT's sources about two and a half to one.

**Product pages lead ChatGPT's sources; lists lead Google AI Mode's** (% of cited links)

On ChatGPT product and service pages were 50.33% of cited links and lists 20.52%; on Google AI Mode lists were 32.16% and product pages 25.34%.

| Label | ChatGPT | Google AI Mode |
|---|---:|---:|
| Product and service pages | 50.33% | 25.34% |
| Lists | 20.52% | 32.16% |
| Guides | 8.44% | 20.81% |
| Discussions | 4.94% | 6.46% |
| Documentation | 4.93% | 1.00% |
| Versus pages | 3.69% | 2.01% |
| Reviews | 1.78% | 1.18% |
| Videos | 0.07% | 7.86% |

Share of cited links by kind of page, of the links the records had sorted by kind, on the buyer questions the GEO Agent asks every day, July 2026. Read each engine against itself: they show different numbers of links per answer. Source: Qvery's GEO Agent, tracked buyer questions (July 2026)

Count answers instead of links and lists look bigger, because one list among a dozen links is enough. **76.91%** of Google AI Mode's sourced answers cited at least one list, and **52.32%** of ChatGPT's. A list was the first link in 23.42% and 20.07% of them.

The two engines show different numbers of links per answer, so read each one against itself, not against the other.

On the same questions, asked every month from February to July, lists fell from **32.26%** of ChatGPT's links to **21.40%**, while product and service pages went from 49.93% to 53.95% (60.95% in June). Google AI Mode's list share slipped more slowly, from 37.18% to 31.68%.

**On the same questions, ChatGPT's list share fell by a third; Google AI Mode's slipped** (% of cited links that were lists)

Lists fell from 32.26% of ChatGPT's cited links in February to 21.40% in July, and from 37.18% to 31.68% of Google AI Mode's.

| Label | Google AI Mode | ChatGPT |
|---|---:|---:|
| Feb | 37.18% | 32.26% |
| Mar | 37.28% | 30.73% |
| Apr | 36.22% | 31.08% |
| May | 35.20% | 24.24% |
| Jun | 34.13% | 18.21% |
| Jul | 31.68% | 21.40% |

Lists as a share of cited links, by month, on the same buyer questions the GEO Agent asked in every month from February to July 2026. Later months left out: the capture changed in August. Source: Qvery's GEO Agent, tracked buyer questions asked every month (February to July 2026)

> A slide that says listicles are the most cited content type is quoting a number that moves: on the same questions, lists fell from 32.26% of ChatGPT's cited links in February to 21.40% in July. Before a budget rides on it, ask which engine the figure comes from and which month, and plan to re-read it every quarter, because a mix that shifted by a third in five months can shift again.
>
> Vlad Shvets, Partner @ Qvery

## Buyers Who Are Choosing Get Answers Built on Lists

Our September research asked both engines two kinds of question across ten industries, mostly in the US, with the UK, Canada and Australia beside it.

Buying questions are the ones people type when they are choosing: "best lenders for FHA loans", "cheapest attorney services with payment plans". How-it-works questions are the ones they type to understand something: "how does the eviction process work for landlords", "what is the difference between Chapter 7 and Chapter 13 bankruptcy".

ChatGPT cited a list in **60.27%** of its answers to buying questions and in **6.22%** of its answers to how-it-works questions. Google AI Mode: 70.60% against 8.43%.

**Lists sit under buying answers, almost never under how-it-works answers** (% of sourced answers citing a list)

ChatGPT cited a list in 60.27% of its answers to buying questions and 6.22% to how-it-works questions; Google AI Mode in 70.60% and 8.43%.

| Label | Value |
|---|---:|
| Google AI Mode, buying | 70.60% |
| ChatGPT, buying | 60.27% |
| Google AI Mode, how it works | 8.43% |
| ChatGPT, how it works | 6.22% |

Share of answers that cited any source and cited at least one list page, by kind of question, ten industries, mostly US, September 2026. A list page ranks or lists several options for one need, read from its title and address. Source: Qvery research: ChatGPT and Google AI Mode answers in ten industries (September 2026)

That is about ten times as often on ChatGPT and eight times on Google AI Mode.

The how-it-works answers leaned on the people who write the rules instead: **58.29%** of ChatGPT's cited a government page, against 14.49% of its buying answers.

A University of Toronto preprint, posted on October 10, tested the format more directly.[^2]

It compared pages of the same website, which holds the site's reputation fixed, on questions about US home internet asked of three assistants (ChatGPT among them) every day from June to September, and repeated the work in a market of small vendors selling to businesses.

Lists and tables on a page went with no change in how often it was cited (−1.0% and −1.4%, neither passing the study’s tests).[^2]

A closer match between a page’s best paragraph and the questions people ask went with **15.5%** more citations, and a closer title match with 14.1% more.[^2]

**What went with more citations, comparing pages of the same website (University of Toronto)**

| Page property | Change in citations |
|---|---|
| Best paragraph matches the questions | +15.5%[^2] |
| Title matches a question | +14.1% |
| Lists on the page | −1.0% |
| Tables on the page | −1.4% |
| Headings | −2.7% |
| Questions-and-answers block | −3.4% |

The Toronto study measured lists within a page, not listicles as a genre. Its findings are associations, not a test of rewriting a page.

## Ecommerce Buyers Get a List in Four ChatGPT Answers of Five

Within buying questions, the industry moves the number a long way. On ChatGPT, a list sat under **82.63%** of the ecommerce buying answers and **73.84%** of the software ones, and under **26.59%** of the healthcare ones. Real estate (53.37%) and legal (47.75%) sat below the middle.

On Google AI Mode, fintech (financial technology) led at **95.35%**. Ecommerce came last at 33.75%.

**On ChatGPT, lists sit under most ecommerce and software buying answers and a quarter of healthcare ones** (% of sourced buying answers citing a list)

A list sat under 82.63% of ChatGPT's ecommerce buying answers and 73.84% of its software ones, and under 26.59% of its healthcare ones.

| Label | Value |
|---|---:|
| Ecommerce | 82.63% |
| Software | 73.84% |
| Travel | 66.48% |
| Fintech | 65.74% |
| Automotive | 64.17% |
| Crypto | 61.39% |
| Education | 60.93% |
| Real estate | 53.37% |
| Legal | 47.75% |
| Healthcare | 26.59% |

Share of ChatGPT's answers to buying questions that cited any source and cited at least one list page, by industry, mostly US, September 2026. Source: Qvery research: ChatGPT answers to buying questions in ten industries (September 2026)

The lists are not the same kind of page either. Who writes them decides whether a page of your own competes with them or a placement on them is the work.

In fintech, two comparison desks carry the lists: NerdWallet's were under **27.41%** of ChatGPT's buying answers and CNBC Select's under 17.59%.

In legal, the city pages of lawyer directories (Avvo, Justia and Super Lawyers) lead, so the list that counts is a directory profile to keep accurate.

In software, no single site's lists were under more than **8%** of answers, which leaves room for a vendor's own page.

**Whose lists the buying answers cite, by industry**

| Industry | The lists the answers cite | What that makes the work |
|---|---|---|
| Fintech | NerdWallet and CNBC Select | Earning a place on two publishers' lists |
| Legal | Lawyer directories' city pages | Keeping your directory profiles right |
| Software | Spread across many sites | Writing a page of your own |

None of the ten industries was logistics, so freight buyers' questions are untested here.

## Only the Company's Own List Went With a Higher Named Rate

A cited list is a page under the answer. The buyer reads the answer itself.

So in the GEO Agent's records we took the questions that do not contain the company's own name and asked, for each answer, what kind of list it cited and whether it named the company the GEO Agent measures.

On ChatGPT in July, an answer that cited no list named the company **25.28%** of the time. An answer whose only lists were other people's named it **25.85%** of the time.

An answer that cited the company's own list named it **77.80%** of the time. Google AI Mode: 30.14%, 23.81%, and 82.29%, respectively.

**Someone else's list did not raise the named rate; the company's own list did** (% of sourced answers naming the company)

ChatGPT named the company in 25.28% of answers citing no list, 25.85% citing only other people's lists and 77.80% citing its own list; Google AI Mode in 30.14%, 23.81% and 82.29%.

| Label | ChatGPT | Google AI Mode |
|---|---:|---:|
| No list cited | 25.28% | 30.14% |
| Only other people's lists | 25.85% | 23.81% |
| The company's own list | 77.80% | 82.29% |

Share of answers that cited any source and named the company the GEO Agent measures, by what kind of list the answer cited, on tracked questions that do not contain the company's name, July 2026. Which came first, the list or the name, the answers cannot show. Source: Qvery's GEO Agent, tracked buyer questions (July 2026)

June, the month we checked against, showed it more sharply: with only other people's lists, ChatGPT named the company 14.23% of the time against 28.07% with none, and Google AI Mode 17.94% against 23.14%.

Across the four readings, someone else's list never lifted the named rate by more than about half a point. The answers cannot tell us which lists included the company.

> We read only the answers to questions without the company's name in them, because a question that names you gets your name back. What those answers cannot show is whether you were on the other lists they cited. To see that for your own brand, open each cited list and record whether it names you, then compare those answers with the ones citing no list. That one field turns the 25.85% from an average into your number.
>
> Piotr Babel, Partner @ Qvery

The Toronto study describes the mechanism the same way: the engines cite the pages their own search finds, and they name the brands they remember, so a page can be cited while the brand that matters to you is not named.

ChatGPT named brands without any supporting page in **46%** of its mentions.[^2]

The own-list number has two readings: your list may get you named, or an answer already heading toward your company may fetch your page; the answers can't separate the two, ours or anyone's.

Our [study of software buyers' questions](https://qvery.ai/blog/saas-self-listicle-playbook/) measured it more carefully, comparing repeated askings of the same question with and without the company's roundup, and found a smaller gap: 52.69% against 30.18%. That comparison is closer to what the page itself adds, though still an association, and the gap is still large.

> 📌 At Qvery, we help companies stop paying for pages no answer reads, with custom AI agents that check each comparison draft against your buyers' questions and report the gaps.

## Few Readers Click Out of the Answer

The last step of the old funnel was the click, and on ChatGPT it is rare. A study of US desktop browsing found that ChatGPT produced an outbound click, a visit to another website, in only **5.2%** of conversation sessions far below Google's referral rate.[^1]

A citation that does not come with your name is a link few buyers open. OpenAI's own help page warns that its search results and citations can be incomplete, out of date or wrong, and tells users to open the source to check.[^3]

For any company whose buyers ask an assistant first, this is a change to how it gets found: the shortlist is written inside the answer, and a list page is one of the things it is written from.

![Two panels: a page gets cited when the assistant’s search finds it and it matches the question; a company gets named when it is on the page the answer reads or the assistant knows the brand.](https://qvery.ai/blog/listicles-most-cited-content-type-ai-search/illustration-cited-vs-named.webp)

*A page can sit under an answer that names other companies; the left panel is the citation, the right is the part the buyer reads.*

Those are averages across other companies' questions. Yours are the ones that set your budget: which of your buyers' questions are choosing questions, which lists their answers cite, and how often the answers name you when your page is among them.

GEO Agent: https://qvery.ai/geo-agent/

## Write the Twenty Pages Only Where Buyers Are Choosing

The twenty pages are worth writing only where your buyers' questions are choosing questions and the answers to them lean on lists, and each draft is worth checking against those questions before it goes live.

Comparison Drafts Checked Against Your Buyers' Questions by Qvery's Agents: https://calendly.com/qvery-vlad/30min

Once a page of yours is live, it is worth keeping for as long as the answers that cite it name you.

## Frequently asked questions



### Are listicles still the most cited content type in AI search?

On Google AI Mode, yes: in the GEO Agent's July 2026 records lists were 32.16% of its cited links, the largest kind of page. On ChatGPT, no: product and service pages were 50.33% of links and lists 20.52%; on a fixed set of questions its list share fell from 32.26% in February to 21.40% in July.

### What is a listicle?

A best-of list page: a page that ranks or lists the products or companies in a category, such as "best lenders for FHA loans". The answers that cite them are mostly answers to choosing questions.

### Does Google AI Mode cite lists more than ChatGPT?

Yes, in the GEO Agent's July records: 76.91% of Google AI Mode's sourced answers cited at least one list, against 52.32% of ChatGPT's, and lists were its largest kind of cited page.

### Should we publish our own best-of list?

Test a list where your buyers ask choosing questions. ChatGPT named the company in 77.80% of answers citing its own list, against 25.28% with no list. Track whether answers name you before and after publication, alongside inquiries mentioning an assistant, before commissioning another page.

### Does getting onto someone else's list help?

ChatGPT named the company in 25.85% of answers citing only others’ lists, versus 25.28% with none, but the records do not show which lists included it. Pursue a placement when the list names you and answers to your buyers’ questions cite it.

### Do AI answers cite listicles for how-to questions?

Rarely. In our September research across ten industries, 6.22% of ChatGPT's answers to how-it-works questions cited a list, against 60.27% of its answers to buying questions. The how-it-works answers cited a government page in 58.29% of cases.

### Does formatting a page as a list help it get cited?

Not on this evidence. A University of Toronto preprint comparing pages of the same website found lists and tables on a page went with no change in citations, while a paragraph or a title that matched the questions people ask went with 15.5% and 14.1% more.[^2]

### How many people click the sources in an AI answer?

Few. A clickstream study, of what a panel of US desktop browsers clicked, found ChatGPT produced an outbound click in 5.2% of conversation sessions.[^1] Count whether the answer names you, not only whether it cites you.

### Which industries' AI answers lean most on lists?

In our September research, a list sat under 82.63% of ChatGPT's ecommerce buying answers and 73.84% of its software ones, and under 26.59% of its healthcare ones. On Google AI Mode fintech led at 95.35%.

### Whose lists do fintech answers cite?

Mostly two comparison desks: NerdWallet's lists were under 27.41% of ChatGPT's fintech buying answers and CNBC Select's under 17.59%, so the work there is earning a place on those lists.

### What should a law firm do about lists in AI answers?

Keep its directory profiles accurate. In legal, the lists the answers cite are the city pages of lawyer directories such as Avvo, Justia and Super Lawyers.

### Is a best-of page of our own worth writing in software?

It can be. In software no single site's lists were under more than 8% of ChatGPT's buying answers, which leaves room for a vendor's own page. Across the GEO Agent's July records, answers citing a company's own list named it 77.80% of the time.

### Why do link shares and answer shares differ?

A single answer can cite several pages, so one list is enough for the answer to count as citing a list. Measure the share of links when comparing page types, and the share of answers when asking how often buyers encounter a list.

### Does a citation mean the answer recommends our company?

A cited page can sit below an answer that names other companies. Record the company names in the answer alongside the cited pages; counting links alone does not tell you whether a buyer saw your name.

### Do these figures prove that publishing a list brings more mentions?

They show an association, not cause and effect. An answer already heading toward your company may retrieve its comparison page, so the observed gap cannot tell you how many mentions a new page will add.

## Sources

[^1]: arXiv, "Answering Without Referring: How AI Search Rewrites the Web's Economic Bargain" (2026), https://arxiv.org/abs/2607.07652
[^2]: University of Toronto (arXiv), "What Drives AI Citations? Content, Retrieval and Model Knowledge in Generative Engine Optimization (GEO)" (2026), https://arxiv.org/abs/2610.15033
[^3]: OpenAI Help Center, "Searching the Web With ChatGPT" (2026), https://help.openai.com/en/articles/9237897-searching-the-web-with-chatgpt
