Key takeaways
- In Promptwatch's ChatGPT data for Sep 28 to Oct 4, 2026, product pages were the most cited type at 22.4%, followed by landing pages (16.8%) and listicles (13.6%).
- Studies covering several models, like Omniscient Digital's, put listicles first. The difference comes from the models covered, the classification scheme, and the time period, so treat any ranking as a snapshot.
- Intent changes the mix, but product pages stay on top for commercial, informational, and branded prompts.
- Content types decay at different speeds. Landing pages and docs stay cited longest, while news loses half its pages in about 36 days.
- You can check your own pages by classifying them, running real prompts, and comparing what gets cited against what you publish.
Why content type matters more than most people think
Most AI search advice focuses on writing quality, schema, or word count. Those matter less than people hope. One study cited by Onely measured 174,048 pages and found a correlation of 0.04 between word count and being cited in an AI Overview. Length is basically noise.
What seems to matter more is whether the shape of your page matches the question being asked. A person asking "best CRM for small teams" wants a comparison. A person asking "what is conversion tracking" wants an explanation. If you publish only one of those shapes, you are invisible for the other, no matter how well written it is.
So the useful question is not "is my content good?" but "do I have the page types ChatGPT actually reaches for, and are they the right ones for the prompts I care about?"
What ChatGPT cites most, by the numbers
Promptwatch classifies every ChatGPT citation by content type. In its citation share by content type report (Sep 28 to Oct 4, 2026), the overall split looked like this:
| 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% |
Two things stand out. First, product and landing pages together make up 39% of citations. That is a lot of weight on pages most content teams treat as an afterthought, handed to a designer and never touched again. Second, editorial formats (listicles, how-tos, articles, comparisons, news, reviews) combined come to about 43%. Blog-style content is still a big pool, but it is split across many formats, so no single one dominates.
Listicles, how-tos, and comparisons together earn roughly 28%. That is the "editorial that behaves like a product" group: structured, scannable, easy to lift from.
A caution about snapshots
These numbers move. Promptwatch's monthly trend reports show product pages leading July 2026 at roughly a third of citations, then slipping to about 25% in late August while how-tos more than doubled. Read the table above as a week's reading, not a permanent ranking. See the ChatGPT citation types over time report for the August trend.
Why other studies say listicles win
If you have read other research, you may have seen a different order. Omniscient Digital analyzed 23,000+ citations across ChatGPT, Google AI Mode, and Perplexity and found listicles at 21.9%, long-form articles at 16.7%, and product pages at 13.7%. Wix's AI Search Lab looked at 1,056,727 citations across the same three models and reached a similar ordering, with listicles first and articles and product pages behind.
These don't contradict the Promptwatch data so much as measure something different:
- Those studies blend three models. Promptwatch's report is ChatGPT only.
- Each study uses its own classification scheme. What one calls a landing page, another may call a product page or a category page.
- The time periods differ, and ChatGPT's retrieval behavior has changed several times in 2026.
The model differences are real, too. According to Wix's research, ChatGPT and Google AI Mode put articles in second place, while Perplexity elevates discussions (Reddit, LinkedIn, G2) to second, with about 17% of its citations. Onely's count of answers adds that ChatGPT cited 6.8 pages per answer, Perplexity 18.2, Gemini 2.7, and Google AI Mode 3.4. And ChatGPT leaned on the provider's own site far more than the others, citing it for 60.5% of its citations in one commercial-question test.
My read: if ChatGPT is your main concern, trust ChatGPT-only data first. If you care about several engines, use the multi-model studies as a sanity check, and expect listicles to matter more.
How intent changes the mix
Promptwatch's report splits ChatGPT citations by prompt type, which is where it gets practical.
Commercial prompts
For prompts like "best CRM for small businesses", product pages take 26.0%, landing pages 20.6%, listicles 15.0%, and comparisons 7.9%. Those four types make up almost 70% of commercial citations.
Informational prompts
For "how does a CRM work", product pages still lead at 19.1%, then landing pages (13.9%) and listicles (13.6%). But how-tos (10.7%), docs (8.5%), and articles (7.7%) all earn more than they do across all prompts. So educational content gets a bigger slice here, just not the top spot.
Branded prompts
For "is HubSpot good for startups", product pages (23.3%) and landing pages (22.8%) lead. Reviews reach 5.6%, more than triple their overall share. Social posts reach 3.0%, almost 4x their overall share. FAQ pages hit 4.8% and press releases 2.1%.
The pattern: product pages are the top type for all three intents. Only the rest of the mix changes. If your site has weak product pages, no amount of blogging fully compensates.
How long each content type stays cited
Citation share is only half the story. Promptwatch also tracked every page ChatGPT cited on July 6, 2026 for 90 days. The content half-life report found large differences:
| Content type | Still cited after 30 days | Still cited after 60 days |
|---|---|---|
| Landing pages | 78% | 69% |
| Docs | 76% | 67% |
| Comparisons | 72% | 63% |
| Listicles | n/a | 60% |
| Product pages, reviews, how-tos | n/a | 55% (half-life 74 to 77 days) |
| News | about half gone by day 36 | 41% |
News is the fast burner. Almost a quarter of cited news pages are never cited again after day 0. That is fine if you want a burst of visibility, and a bad bet if you want durable presence.
The less obvious finding is that product pages decay about as fast as how-tos. Prices and specs go stale, and a stale product page loses citations. Landing pages, docs, comparisons, and listicles only cross 50% in the last week of the window, when every line drops because the tracking period ends, so their real half-life is likely near or beyond 90 days. Pricing pages count as landing or product pages in this classification.
This lines up with third-party freshness data. Onely found 76.4% of ChatGPT's most-cited pages were updated within the previous 30 days, and Ahrefs found the average AI-cited page is 25.7% fresher than the average organic result. Freshness helps. It just isn't the whole story, since the same Ahrefs data shows the average cited page is still 2.9 years old.
What else affects citation, beyond type
Format is the first filter, not the only one. A few other findings from the research are worth knowing:
- Position on the page. Growth Memo's February 2026 analysis found 44.2% of citations come from the first third of a page, 31.1% from the middle, and 24.7% from the end. Put the definition or recommendation up top.
- Answer capsules. Search Engine Land found 72.4% of pages ChatGPT cites include a short, direct answer right below a question-style heading, typically 20 to 25 words.
- Titles and URLs. In the 1.4 million prompt study covered by CXL, readable slugs had an 89.78% citation rate versus 81.11% for opaque ones. A URL like /what-is-intent-data/ gives more signal than /resources/article-47/.
- Retrieval rank. A University of Toronto study found ChatGPT consulted 26 pages per question and cited 3, citing 43.8% of pages its search ranked first against 3.5% of those ranked 11 to 20. Being retrieved matters more than any on-page property.
- Schema. Ahrefs tracked 1,885 pages that added schema against 4,000 controls and saw citation changes indistinguishable from zero on AI Mode and ChatGPT. Add it for other reasons, not as a citation lever.
Also keep Google rankings in perspective. Ahrefs found only 12% of links cited by ChatGPT, Perplexity, and Copilot appear in Google's top 10 for the same question. Ranking and being cited are different systems.
How to check whether your pages match
Here is the audit I would run. It takes an afternoon for a small site and a week for a large one.
Step 1: Inventory and classify your pages
Crawl your site and label each URL with a content type using the same categories as the table above: product, landing, listicle, how-to, docs, article, comparison, news, FAQ, review. A crawler like Screaming Frog is enough to export URLs, titles, and word counts into a spreadsheet.
Count what you have. Most sites discover they have 200 articles, 12 product pages, no comparison pages, and no FAQ.
Step 2: Write the prompts your buyers actually use
Build a list of 30 to 50 prompts across the three intents: commercial ("best X for Y"), informational ("how does X work"), and branded ("is [your brand] good for Z"). Include a few competitor-comparison prompts.
Step 3: See what ChatGPT cites
Run the prompts and record every cited URL with its type. Doing this by hand works for a first pass. Beyond that, a tracking platform saves a lot of time. Promptwatch tracks real ChatGPT interface responses, classifies citations by content type, and shows which prompts cite which pages, so you can compare your mix against competitors without a spreadsheet.
Other trackers can cover the monitoring side if you only need basics, including Otterly.AI, Peec AI, and Profound. Check whether they classify content type, since many only report mentions.
Otterly.AI
Profound
Step 4: Compare your mix to the cited mix
For each intent, ask:
- Which types does ChatGPT cite for this prompt, and do I have that type?
- When it cites a competitor's product page, what does theirs have that mine lacks (specs, prices, positioning, FAQs)?
- Are the cited listicles third-party roundups I am not in?
That last one matters. In Wix's analysis of the top 1,000 cited URLs in professional services, 80.9% were third-party and only 19.1% self-promotional. For commercial prompts, getting listed in someone else's "best of" can be worth more than publishing your own.
Step 5: Check that the pages can be read at all
A page that matches the right type still gets nothing if AI crawlers cannot fetch it. Check robots.txt and your CDN rules, and look at server logs for GPTBot, OAI-SearchBot, and ChatGPT-User. Promptwatch's AI crawler traffic data shows what normal verified request volume looks like by crawler. For deeper log analysis, Siteline AI and Screaming Frog both help.
Step 6: Fix in priority order
I would work through gaps in this sequence:
- Product and landing pages. Make specs, prices, positioning, and use cases complete and factual. Put a direct answer in the first paragraph.
- Comparison and listicle pages for the commercial prompts where you are absent.
- How-tos and docs for informational prompts. Keep docs public and crawlable. Promptwatch's software citation report shows documentation and help centers earning the largest share for software prompts.
- An FAQ page covering brand questions, plus reviews on trusted third-party sites for branded prompts.
- A refresh schedule. Reviewing product pages monthly is reasonable given their 74 to 77 day half-life.
Tools for content-type analysis and refreshes
No single tool does the whole job. This is how the categories break down:
| Job | Example tools | Note |
|---|---|---|
| Track citations by content type | Promptwatch | Classifies citations, ties them to prompts and pages |
| Basic mention monitoring | Otterly.AI, Peec AI | Cheaper, less depth |
| Crawl and classify your own pages | Screaming Frog | Good for the inventory step |
| Content briefs and optimization | Clearscope, Surfer SEO, Frase | Help structure pages, no citation data |
| Change detection on competitor pages | Page Modified, Visualping | Useful for watching cited pages |
For a wider look at options, the GEO software directory at bestgeosoftware.com lists many more trackers.
Common mistakes
- Treating one study as gospel. Listicles lead in some multi-model data, product pages lead in ChatGPT-only data. Know which model you are optimizing for.
- Publishing more articles when the real gap is a missing comparison page or a thin product page.
- Letting prices and specs go stale. Product pages decay about as fast as how-tos.
- Chasing news for durable visibility. Half of cited news pages are gone within 36 days.
- Expecting Reddit participation to translate into citations. Promptwatch's data shows reddit.com's share of ChatGPT citations fell from about 4% to 0.5% on August 14, 2026 (see the Reddit citations report). It may inform answers, but it is rarely credited.
- Assuming being cited means being visited. Onely found about 1% of summary links get clicked, and its own cited page earned zero clicks from 5,544 impressions. Measure traffic and conversions, not just citations.
A short framework to remember
Match the shape of the page to the shape of the question, make the key answer easy to lift, keep it current, and make sure crawlers can reach it. Everything else is detail.
Start with the inventory. If you cannot say how many product, comparison, and FAQ pages you have today, you cannot know which gap to close first. For more data on how ChatGPT's behavior shifts month to month, the Promptwatch Data hub is updated regularly.