Key takeaways
- Across all classified ChatGPT citations, product pages are the most cited type at 22.4%, then landing pages at 16.8% and listicles at 13.6% (Promptwatch Data, Sep 28 to Oct 4, 2026).
- Older claims that "Best X" listicles or directories dominate ChatGPT come from narrower samples. They are still useful for buying questions, but they are no longer the whole story.
- Durability varies a lot. Landing pages, docs and comparisons keep their citations longest, while news loses half its pages within 36 days.
- Being named on someone else's list does not reliably get you named in the answer. A list you own, written for the questions your buyers ask, does much better.
- Start with your own product, landing and pricing pages, then add comparison pages, then pursue third-party roundups and directories.
Why the "most cited type" question is harder than it looks
Somebody in your company has probably shown you a slide saying listicles are the most cited content type in AI search. It sounds like a measurement. Then you ask the obvious questions: most cited of what, in which engine, on which prompts, and measured when?
The answers change the conclusion. I found three different pictures in the research for this guide, and they disagree because they sample different things:
- A widely shared LinkedIn post claims 61% of what ChatGPT cites is a directory or a "best of" list, and only 1% is Reddit. It tracked 2,490 citations across 30 buying questions. That is a narrow sample of purchase queries.
- Qvery's GEO Agent, looking at buyer questions in software, services, insurance and travel in July 2026, found product and service pages at 50.33% of ChatGPT links and lists at 20.52%.
- Promptwatch Data, covering all classified ChatGPT citations from Sep 28 to Oct 4, 2026, puts product pages at 22.4% and listicles at 13.6%.
The gap between Qvery's 50% and Promptwatch's 22% is not a contradiction. Qvery asks buyer questions only, and the two use different classifiers. Treat Promptwatch's numbers as the broad baseline and Qvery's as what happens when the user is shopping.
Two caveats before the numbers. Citation share changes month to month, so check the period on any stat you quote. And a citation is not a sale. It is a page the answer pulled in, which is a prerequisite for being named, not proof of it.
What ChatGPT cites: the content type breakdown
Here is the split of ChatGPT citations by content type for all prompts, from Promptwatch's content type report (Sep 28 to Oct 4, 2026):
| Content type | Share of ChatGPT citations |
|---|---|
| Product pages | 22.4% |
| Landing pages | 16.8% |
| Listicles | 13.6% |
| How-to guides | 8.7% |
| Documentation | 7.5% |
| General articles | 6.1% |
| Comparisons | 5.9% |
| News | 5.7% |
| FAQ pages | 3.3% |
| Reviews | 1.7% |
Product and landing pages together are 39% of citations. That is the single most useful line in the table. The pages you own and fully control are the biggest citation surface, and a lot of teams are still spending their budget on pages that sit further down the list.
Editorial formats (listicles, how-tos, comparisons, articles and reviews) add up to a large chunk too, at 43% combined including reviews. Listicles, how-tos and comparisons alone make up 28%. So third-party and editorial content matters. It just is not the main event.
How intent changes the mix
The same report splits citations by prompt type, and the differences are worth acting on.
| Prompt type | Top types | Notes |
|---|---|---|
| Commercial ("best CRM for small businesses") | Product 26.0%, landing 20.6%, listicles 15.0%, comparisons 7.9% | These four types are almost 70% of citations |
| Informational | Product 19.1%, landing 13.9%, listicles 13.6%, how-tos 10.7%, docs 8.5% | How-tos and docs matter more here |
| Branded | Product 23.3%, landing 22.8%, reviews 5.6%, FAQ 4.8%, social 3.0% | Reviews are over three times their all-prompt share |
So if your buyers ask comparison-style questions, comparison pages earn a bigger share (7.9% vs 5.9% overall). If people are asking about you by name, reviews and FAQ pages carry more weight than usual.
Product and landing pages: your main citation surface
If you do one thing after reading this guide, make your product, landing and pricing pages complete, factual and easy to compare.
One more data point supports this. A 180 Marketing analysis found brands were recommended 46% more often in ChatGPT when a product page from the same website was cited as a source (43.6% recommended when cited vs 29.8% when not). Category pages showed a +39% lift and homepages +37%. This is correlational, so I would not promise a 46% improvement from publishing a page. But the direction matches everything else.
What "complete and factual" means in practice:
- State prices, plans and limits in plain text, not behind a form or inside an image.
- Say who the product is for and who it is not for.
- List specs and features in a structure a model can lift without guessing, like a table or a clean list.
- Keep positioning consistent between your site, your docs and your profiles elsewhere.
- Make sure AI crawlers can actually read the page. If GPTBot or OAI-SearchBot hits errors, none of the above matters.
Pricing pages are classified as landing or product pages in Promptwatch's data, which is one reason that bucket is so large. Hiding your pricing is a real cost in AI search, not only in conversion.
Listicles: still useful, but be precise about how
Listicles are third in the broad data, and the share has been sliding in the buyer-question data. Qvery's monthly series shows lists falling from 32.26% of ChatGPT's links in February to 21.40% in July 2026, while product and service pages rose from 49.93% to 53.95% (with a 60.95% peak in June). On Google AI Mode lists still led at 32.16% of links in July.
Still, lists are what the assistants read when someone is choosing. Qvery found 52.32% of ChatGPT's sourced answers on buyer questions cited at least one list, and 76.91% for Google AI Mode.
The finding I think matters most is this one. An answer that cited only other people's lists named the company no more often than an answer with no list at all. When the list was the company's own, ChatGPT named it 77.80% of the time, against 25.28% with no list. That is Qvery's sample, from software, services, insurance and travel, so read it as a strong signal rather than a law.
The practical read:
- Appearing in a third-party roundup is worth pursuing, but do not assume it translates into being named.
- A "best [category]" page on your own domain, written for the questions your buyers ask, does real work.
- Do not build twenty of them. Count the questions where your buyers are choosing, and write for those.
Promptwatch's own advice is the same shape: get listed in the roundups ChatGPT cites for your category, and publish your own comparison pages.
Directories and marketplaces
The "61% directories or best-of lists" claim is the one you will see most often. It comes from a sample of 30 buying questions, so I would not generalize it. But the underlying point holds for specific verticals.
Promptwatch's vertical reports show it clearly:
- In travel, Skyscanner, Going and Tripadvisor lead, and booking and search sites earn more than half of citations (travel report).
- In software, GitHub, Microsoft's docs and G2 lead, with documentation and help centers earning the largest share (software report).
- In automotive, marketplaces like Autotrader and Cars.com outrank every car brand, and Chevrolet is the top manufacturer site (automotive report).
- In healthcare, government and public health sites earn almost half of citations, led by NIH, the FDA and the NHS (healthcare report).
So the right directory to target depends entirely on your category. In some verticals the aggregator is the answer. In others it barely registers. I did not research directories as a separate content type for this guide, so the way to find yours is to look at what ChatGPT actually cites for your prompts.
Docs, how-tos, reviews and social
Documentation is 7.5% of all citations but 8.5% of informational ones, and it is stable. ChatGPT cited 93% of docs pages again later, against 77% of news pages. For developer tools and SaaS, public crawlable docs are a straightforward win.
How-tos take 8.7% overall and 10.7% for informational prompts. They are worth building when your buyers are still learning.
Reviews are small overall (1.7%) but matter for branded prompts (5.6%). Collect them on sites ChatGPT already trusts in your category.
Social is weak in ChatGPT specifically. From Sep 28 to Oct 4, 2026, ChatGPT got 0.46% of its citations from social platforms, while Google AI Overviews got 12.2% (social citations report). Reddit's share of ChatGPT citations fell from about 4% to 0.5% on August 14, 2026 (Reddit report). If your plan relied on Reddit threads to win ChatGPT, rethink it. It may still matter for Google's surfaces.
How long citations last
Choosing a type is only half of it. Some types hold their citations and some do not. Promptwatch tracked every page ChatGPT cited on July 6, 2026 for 90 days (content half-life report):
| Type | Still cited after 30 days | Still cited after 60 days |
|---|---|---|
| Landing pages | 78% | 69% |
| Docs | 76% | 67% |
| Comparisons | 72% | 63% |
| Listicles | not reported | 60% |
| Product pages | not reported | 55% |
| Reviews | not reported | 55% |
| How-tos | not reported | 55% |
| News | not reported | 41% |
Product pages, reviews and how-tos reach their half-life at 74 to 77 days. Landing pages, docs, comparisons and listicles only cross 50% in the final week of the window, so their half-life is likely at or beyond 90 days. News is the outlier: almost a quarter of news pages are never cited again after day 0, and half are gone by day 36.
Two takeaways. Treat landing, docs and pricing pages as long-term assets. Treat product pages as pages that need regular updates, since they decay faster than you might expect for something so central. And do not bet a citation strategy on news.
Which type to target first
| Priority | Page type | Why | Effort |
|---|---|---|---|
| 1 | Product and pricing pages | Largest share (22.4%) and strongest link to being recommended | Low to medium |
| 2 | Landing pages | 16.8% share and the longest-lasting citations | Medium |
| 3 | Comparison pages on your domain | 7.9% of commercial citations, 63% still cited at 60 days | Medium |
| 4 | Your own "best of" list for choosing questions | Qvery found answers citing a company's own list named it 77.80% of the time | Medium |
| 5 | Public docs and how-tos | Docs are stable and how-tos reach 10.7% on informational prompts | Medium to high |
| 6 | Third-party roundups, directories and reviews | Helps in vertical-specific cases and branded prompts | High, depends on others |
| 7 | News and press | Short bursts only | Varies |
A caution on this ordering. It reflects broad data. If your category is travel or automotive, bump directories and marketplaces up.
A workflow you can run this month
- Pull 30 to 50 prompts your buyers really ask, split into choosing, learning and branded.
- For each, record which domains and which page types ChatGPT cites.
- Compare against your own pages. Where is your product page absent? Where do roundups cite competitors but not you?
- Fix your product, landing and pricing pages first.
- Build comparison and list pages only for the choosing questions you identified.
- Check crawler access. A page that AI bots cannot fetch cannot be cited.
- Re-measure monthly, because shares move.
Counting matters more than clicks here. Few readers click out of an AI answer, so the useful metric is whether the answer names you.
Tools that help
You need something that records citations by prompt, ideally with page-type classification. Platforms like Promptwatch do this across ChatGPT and other engines, and pair it with crawler logs and content gap analysis, which covers steps 2 to 6 above.
Plenty of lighter trackers cover the monitoring side if you only need to see who gets cited. Otterly.AI is one affordable option.
For sources specifically, MentionBird focuses on showing which sources AI cites in your category.
If you want to compare more options, the GEO software directory at bestgeosoftware.com lists a wide set of tools by category.
For the wider picture, Promptwatch Data publishes regularly updated citation research you can check against your own numbers.
Common mistakes
- Commissioning dozens of "best X" pages because a slide said listicles win. Count your choosing questions first.
- Treating a mention on someone else's list as a win. The data says it often does not get you named.
- Neglecting product pages in favor of blog content. They are the biggest share.
- Quoting one study as universal. The 61% directory stat, Qvery's 50% and Promptwatch's 22% all measure different things.
- Publishing news for citation value. Half the pages are gone in about five weeks.
- Ignoring that Google AI Mode and ChatGPT behave differently. Lists lead on AI Mode, product pages lead on ChatGPT.
Bottom line
Target your own product, landing and pricing pages first, because they make up the biggest share of ChatGPT citations and they last. Add comparison pages and one well-built list for each question where buyers choose. Chase third-party roundups and directories where your vertical clearly rewards them, and measure by whether the answer names you, not by how many pages got cited.