Key takeaways
- Product pages are the most cited content type in ChatGPT at 22.4% of classified citations (Sep 28 to Oct 4, 2026), and landing pages are second at 16.8%. Together that is 39% of citations going to pages that describe what a company sells.
- Product pages rank first for commercial, informational and branded prompts alike. They are not just a bottom-of-funnel play.
- ChatGPT quotes specs, prices and positioning directly, so complete, factual, current pages beat clever marketing copy.
- Product pages decay faster than landing pages and docs. About 55% are still cited after 60 days, so freshness is a maintenance job, not a launch task.
- Pair the page with comparison content, an FAQ, third-party reviews and open crawler access, then measure whether it actually gets cited.
Why this finding matters more than it looks
Most GEO advice I read in 2025 was about blog posts. Write the definitive guide, add some statistics, wait. The data says the pages doing the heavy lifting in ChatGPT are the ones most teams treat as an afterthought: the product page and the landing page.
Promptwatch Data's report on ChatGPT citation share by content type classified every ChatGPT citation from Sep 28 to Oct 4, 2026. Here is how it broke down:
| Content type | Share of ChatGPT citations |
|---|---|
| Product pages | 22.4% |
| Landing pages | 16.8% |
| Listicles | 13.6% |
| How-tos | 8.7% |
| Documentation | 7.5% |
| General articles | 6.1% |
| Comparisons | 5.9% |
| News | 5.7% |
| FAQ pages | 3.3% |
| Reviews | 1.7% |
One caveat. Other vendors publish different numbers. Evertune claims 50% of ChatGPT citations are listicles, and Wix Studio (via HubSpot) puts product pages at 13.7% of citations across all engines. Different samples, different classifiers, different time windows. I lean on the Promptwatch numbers here because they are ChatGPT-specific and dated, but the direction is consistent everywhere: pages that describe a product or an offer get cited a lot.
It holds across prompt intent
The part that surprised me is that product pages lead for every intent Promptwatch measured.
| Prompt type | Example | Product pages | Landing pages | Other notable types |
|---|---|---|---|---|
| Commercial | "best CRM for small businesses" | 26.0% | 20.6% | Listicles 15.0%, comparisons 7.9% |
| Informational | "how does a CRM work" | 19.1% | 13.9% | Listicles 13.6%, how-tos 10.7% |
| Branded | "is HubSpot good for startups" | 23.3% | 22.8% | Reviews 5.6%, FAQ 4.8%, social 3.0% |
Four types (product, landing, listicle, comparison) make up almost 70% of commercial citations. For branded prompts, product and landing pages together are almost half. If someone asks ChatGPT about your company by name, the single most likely source is a page you control. That is both an opportunity and a risk: if your page is thin, outdated or vague, ChatGPT will fill the gaps from somewhere else.
How ChatGPT decides which pages to cite
Ahrefs studied 1.4 million ChatGPT prompts and published the results in Why ChatGPT cites one page over another. The video walkthrough from Edward Sturm is a decent summary if you do not want to read the full study.
The findings that matter for product pages:
- ChatGPT retrieves dozens of sources per response but cites only about half of them.
- Titles, snippets and URLs are judged before the page is even opened. A vague title like "Solutions" loses to a descriptive one.
- Semantic relevance to the fan-out queries (the sub-searches ChatGPT runs behind the scenes) is one of the strongest factors.
- Natural-language URLs help.
- Older, established pages often beat newer ones for the same query. Freshness acts mainly as a tiebreaker for news.
That last point deserves a second look next to the half-life data below. Establishment helps you get picked. Stale facts make you drop out later.
Fan-outs are your real keyword list
A single prompt like "best project management tool for agencies" becomes several searches: pricing comparisons, integrations, reviews, specific features. Your product page needs to be semantically relevant to those sub-queries, not just the headline prompt. Promptwatch Data covers how this works in ChatGPT query fanouts, and the ChatGPT Query Fan-Out Generator is a free way to see likely sub-queries for any prompt. The free GEO Inspector Chrome extension shows the actual fan-outs and which sources ChatGPT fetched versus cited.
It is also worth knowing that on August 8, 2026, ChatGPT began using the site: operator at scale, going from about 0.4% to about 17% of fan-out queries overnight (Promptwatch Data). More of those searches now target specific domains, so having a clear, crawlable page on your own domain for each product is more valuable than it was a few months ago.
What a citable product page contains
Promptwatch's own recommendations from the content-type report are blunt: keep specs, prices and positioning complete and factual, because ChatGPT quotes them directly. I would add structure from the other studies. Here is the anatomy I would build.
1. A title and URL that name the thing
Use the product name plus the category in the title ("Acme Invoicing: Invoice software for freelancers"). Use a readable URL like /products/invoicing-software rather than /p?id=4821. Since titles, snippets and URLs get judged before a page is opened, this is cheap and it is first in line.
2. A plain-language definition in the first paragraph
One or two sentences: what it is, who it is for, what it costs or where pricing starts. Write it as if an assistant will lift it verbatim, because it might. Skip the slogan.
3. Real specs and real prices
Put pricing on the page as text, not behind a form or inside an image. If pricing varies, say so and give ranges or the entry tier. Include dimensions, integrations, plan limits, compatibility, whatever buyers actually ask. Hidden pricing is the most common reason I see product pages ignored in favor of third-party roundups that list the numbers.
4. A "who it is for" section
Commercial prompts are often phrased by persona: "best X for small teams", "best X for agencies". A short section naming the audiences you serve, and the ones you do not, gives ChatGPT something to match against.
5. A comparison table
AirOps analyzed 217,508 retrieved pages across 7,500 commercial prompts and found comparison content with tables earned 25.7% more citations, and pages that make shortlisting scannable earned 18.8% more. Put a table on the product page comparing your plans, and build separate comparison pages against alternatives.
6. Short, self-contained paragraphs
Evertune analyzed more than 40,000 cited URLs and found the median cited page runs about 941 words, with around 18 paragraphs, roughly 17 words per sentence and about four H2s. Treat these as medians, not targets. SE Ranking found pages over 2,900 words averaged 5.1 citations, so longer is not penalized when the content earns it. The practical rule: every paragraph should stand alone as a quotable unit of evidence.
7. An FAQ block in plain language
Branded prompts pull FAQ pages at 4.8%, well above their 3.3% overall share. Answer the questions people type about your brand: is it good for X, how does pricing work, does it integrate with Y, what are the downsides. Answer honestly. A dodgy FAQ is worse than none.
Product page or landing page?
The two types are close in the data (22.4% vs 16.8%), and Promptwatch notes pricing pages are classified as one or the other. I would not agonize over the label. What matters is that the page describes an offer completely. A landing page that is all emotional copy and a form is not going to behave like a product page; a landing page with specs, pricing and a clear audience will.
The half-life data gives a reason to have both:
| Content type | Still cited after 30 days | Still cited after 60 days | Note |
|---|---|---|---|
| Landing pages | 78% | 69% | Last longest |
| Docs | 76% | 67% | 93% are cited again after day 0 |
| Comparisons | 72% | 63% | True half-life likely 90+ days |
| Listicles | n/a | 60% | Cross 50% only at the window's end |
| Product pages | n/a | 55% | Half-life of 74 to 77 days |
| News | n/a | 41% | Half gone after 36 days |
That comes from Promptwatch Data's content half-life report, which tracked every page ChatGPT cited on July 6, 2026 through Oct 4, 2026. The takeaway is that product pages are heavily cited but decay faster than landing pages, docs and comparisons. If you let prices, specs and availability drift, you lose the citation. Put a quarterly (or monthly, for fast-moving catalogs) refresh on someone's calendar.
Build the surrounding ecosystem
A great product page alone is not enough. ChatGPT cites on average a lot of domains per response. Promptwatch Data found that almost half of ChatGPT responses cite 10 or more unique domains (Sep 28 to Oct 4, 2026). You are competing for one slot among many, and the other slots go to listicles, comparisons and reviews.
Get into the roundups
Listicles are 13.6% of ChatGPT citations and 15.0% of commercial ones. Find the "best of" pages ChatGPT cites for your category and pitch for inclusion. Also publish your own honest comparison pages. Comparison pages are among the most durable types in the half-life data.
Collect reviews on trusted third-party sites
Reviews are only 1.7% of citations overall but rise to 5.6% for branded prompts, over three times their normal share. For software, Promptwatch's software citation report shows G2 among the top cited domains alongside GitHub and Microsoft's docs. Directories like Trustpilot matter for the same reason.
Keep docs public
Documentation earns 7.5% of citations overall and 8.5% on informational prompts. If your specs live in a gated PDF or a login-only help center, ChatGPT cannot use them. Keep docs public and crawlable.
Do not expect authority to rescue you
Promptwatch Data shows sites with domain rank 46 to 75 earned almost half of ChatGPT citations in August 2026, while DR 91 to 100 fell to about 3% (report). A mid-sized brand with a complete product page can beat a giant with a vague one.
Make sure ChatGPT can actually read the page
This is the boring step that kills more pages than bad copy does. Check:
- robots.txt does not block GPTBot, OAI-SearchBot or ChatGPT-User
- Pricing and specs render in the HTML, not only through client-side JavaScript
- The page is in your sitemap and linked internally (no orphan product pages)
- Pages return clean 200 responses, not soft errors
Free helpers: the Robots.txt Generator for AI, the AI Crawler URL Inspector and the Agentic Readiness Check. For JavaScript-heavy storefronts, a prerendering service such as Prerender.io can help bots see real content.
A note on llms.txt and markdown: Promptwatch Data found markdown files make up just 0.05% of AI search citations (report). Do not spend your week on that. Spend it on the HTML product page.
Measure whether it worked
Publishing is the easy half. You want to know three things: is the page being crawled, is it being cited, and does it send visitors.
Promptwatch is built for that loop. Its Page Tracker shows which prompts and responses cite a given URL, its AI crawler logs show whether GPTBot and friends actually fetched the page and where they failed, and visitor analytics show traffic and conversions from AI engines. Its content gap analysis maps your pages against the prompts AI answers, which is how you find the product you forgot to describe properly. The Content Agent can draft and publish the missing pages to WordPress, Webflow or Framer. If you sell physical products, ChatGPT Shopping tracking shows which products ChatGPT recommends instead of yours.
Plans start at $95/month for Essential, and there is a 7-day free trial on monthly billing. Check promptwatch.com/pricing for current details. Crawler logs start on Professional ($245/month), and the Content Agent and Shopping insights start on Business ($579/month). If you just want a baseline first, the free AI visibility report scans how ChatGPT, Gemini and Perplexity talk about your brand with no login.
Other options exist. Monitoring-focused trackers such as Peec AI, Otterly.AI and Profound report mentions and citations, though most stop short of crawler logs or content execution. AirOps leans toward content engineering. If you want a wider survey, the GEO software directory at bestgeosoftware.com lists dozens.
Otterly.AI
Profound
A practical rollout plan
Week 1: audit
List your top products or offers. For each, check title, URL, whether pricing is visible in HTML, and whether the page is crawlable. Run the free AI Visibility Report or check Page Tracker data to see which pages ChatGPT already cites.
Week 2: rewrite the top five
Add the plain-language definition, specs, pricing, audience section, comparison table and FAQ. Keep sentences short. Remove vague claims you cannot back with a number.
Weeks 3 and 4: build the surroundings
Publish two comparison pages. Pitch for inclusion in two cited listicles. Ask satisfied customers for reviews on the platform ChatGPT cites in your category. Open up any gated docs.
Every month: refresh and check
Update prices and specs. Look at which prompts cite your pages and which competitor pages still win. Product pages hit their half-life in roughly 74 to 77 days, so a page you do not touch for a quarter is a page you are slowly losing.
Common mistakes
- Pricing "on request" with no range. ChatGPT then cites a roundup that lists numbers.
- One generic "Solutions" page covering ten products. Each product deserves its own URL.
- Specs inside images or PDFs.
- Marketing adjectives instead of facts. "Best-in-class" gives a model nothing to quote.
- Treating launch as the finish line. Citations decay.
- Blocking AI crawlers by accident through a copied robots.txt or a CDN bot rule.
The bottom line
The pages that earn the most ChatGPT citations are the ones that say plainly what you sell, who it is for, what it costs and how it compares. That is not glamorous work, and most competitors have not done it well. Pick your five most important product pages, make every fact on them complete and current, and check in 60 days whether they are still being cited. If you are going to hire help for the wider AI search program, the same discipline applies: ask to see citation and crawler data, not just rankings.