Content types by prompt intent: what ChatGPT cites for commercial, informational and branded questions
A side-by-side look at the page formats ChatGPT cited for commercial, informational and branded prompts in one week of Promptwatch data, and what to publish first.
Quick verdict
Across all three intents, the page type ChatGPT cited most was the product page. That is the headline of Promptwatch's data report on citation share by content type, which classified ChatGPT citations from September 28 to October 4, 2026. Product pages took 22.4% of citations across all prompts, and landing pages 16.8%. Put the two together and 39% of what ChatGPT cited was a page describing what a company sells.
The surprise is not that sales pages are cited. It is how much they lead the editorial formats that most content programs spend their time on. Intent changes the mix at the edges, but the top of the table barely moves.
What was measured
Promptwatch classified the pages ChatGPT cited into content types, then split the result by prompt intent. It is one vendor's dataset, for one week, in ChatGPT only. Editorial formats together (listicles, how-to guides, articles, comparisons, news, reviews and other editorial) account for 43% of citations across all prompts. Within that, listicles, how-tos and comparisons add up to 28%.
The classification is by page type, so a blog post about a product and the product page itself land in different buckets. We do not have the report's full labeling rules, so treat the boundaries between types as approximate.
The comparison table
| Content type | All prompts | Commercial | Informational | Branded |
|---|---|---|---|---|
| Product pages | 22.4% | 26.0% | 19.1% | 23.3% |
| Landing pages | 16.8% | 20.6% | 13.9% | 22.8% |
| Listicles | 13.6% | 15.0% | 13.6% | not stated |
| How-to guides | 8.7% | not stated | 10.7% | not stated |
| Documentation | 7.5% | not stated | 8.5% | not stated |
| General articles | 6.1% | not stated | 7.7% | not stated |
| Comparisons | 5.9% | 7.9% | not stated | not stated |
| News | 5.7% | not stated | not stated | not stated |
| FAQ | 3.3% | not stated | not stated | 4.8% |
| Reviews | 1.7% | not stated | not stated | 5.6% |
| Social posts | not stated | not stated | not stated | 3.0% |
| Press releases | not stated | not stated | not stated | 2.1% |
Source: Promptwatch data report, September 28 to October 4, 2026, ChatGPT. "Not stated" means the figure is not in the report summary we worked from, not that it is zero.
Reading it by intent
Commercial prompts. Product pages (26.0%), landing pages (20.6%), listicles (15.0%) and comparisons (7.9%) account for almost 70% of citations between them. If someone asks ChatGPT which product to buy, the answer comes from a short list of formats. Comparisons are the smallest of the four, but they are the one the buyer is literally asking for.
Informational prompts. The mix spreads out. Product pages still lead at 19.1%, followed by landing pages (13.9%), listicles (13.6%), how-to guides (10.7%), documentation (8.5%) and general articles (7.7%). This is where how-to content earns its place, at 10.7% against 8.7% overall.
Branded prompts. Product pages (23.3%) and landing pages (22.8%) together are almost half. Reviews reach 5.6% here, three times their overall share, and social posts hit 3.0%, almost four times theirs. Branded prompts are the moment a person has already named you, so third-party opinion and social proof count for more than usual.
What it does not tell you
It does not say that product pages cause citations, or that publishing more of them will raise your share. A page being cited and a page being chosen as the best answer are different things. It does not tell you the citation was favorable. It also covers a single week, so a launch, a news event or an engine update could tilt the numbers. Other Promptwatch reports show shares moving within a month.
And it does not say your market looks the same. A niche B2B category may cite documentation and forums far more than this average does.
What to do without any tool
- Audit your product and landing pages first. Our reading: they lead every intent in the table, so treat them as citation sources in their own right. Check that each says plainly what the product is, who it is for, what it costs and how it differs. Dated specifications and current prices beat vague taglines.
- Write the comparison and listicle pages your buyers ask for. Commercial prompts lean on listicles and comparisons. Use sourced facts, name trade-offs and say who each option is wrong for.
- Keep how-to and documentation pages for the informational layer. They matter most when the prompt starts with how, not which.
- Look after reviews and social proof for branded prompts. If your name is typed into the box, reviews carry three times their usual weight in the report.
Then run the check yourself. Pull 30 prompts split into the three intents, run them in ChatGPT, and tally the content types of every cited URL. Where your tally differs from the table, trust yours.
Measuring it continuously
Hand-tallying works for a month or two. For anything longer, classification is the task to automate. In our catalog, Promptwatch documents Citation Trends, which classify every citation by content type and source type, with citation rate over time and first cited, ramp-up days, peak rate and decay half-life per page. That is the same slicing the report uses, run on your own prompts. Its prompt tracking supports prompt types (organic, brand, competitor comparison) and intent, so you can split the results as the table above does.
A Page Tracker follows specific URLs, yours or a competitor's, and Content Refresh gives diff-style update suggestions for existing pages. Plan notes: the free Explore plan covers ChatGPT only with 10 prompts, Essential is $95 a month and Professional is $245. Crawler logs, which show whether bots can read your product pages at all, start on Professional.
Among other catalog entries, HubSpot AEO lists citation analysis and prioritized recommendations, and AthenaHQ lists an Action Center, but neither entry records content-type classification of citations. We treat that as unknown.
Who should pick which
- A team with thin product pages: fix those first. Nothing else in the table is bigger.
- A comparison or review-led publisher: your formats sit lower on the all-prompt list, but commercial and branded prompts raise them. Keep facts dated and sourced.
- A content team measuring results: use a platform with citation classification, and Promptwatch is the one in our catalog that documents it. Start on Explore to see whether your category looks like the report.
- A team with no budget: the 30-prompt tally, repeated every month, is enough.
Related reading: how often to update a listicle, whether listicles still work in 2026 and content refresh versus new articles.
Bottom line
For one week of ChatGPT citations, the pages that describe what a company sells outweighed any single editorial format, in every intent. Make those pages good before you invest elsewhere, then add the formats each intent favors. No format guarantees a citation, and the table will shift.