Which pages ChatGPT cites when someone asks about you by name
Promptwatch's October 2026 data shows what ChatGPT cites for branded prompts: product and landing pages first, then reviews and FAQ. A field guide to auditing those pages.
When a buyer types your company name into ChatGPT and asks a question, which of your pages does the answer lean on? We get this question from founders more than any other, usually phrased as "why does it say that about us?" The answer starts with knowing what kinds of pages get cited for branded questions at all.
Promptwatch's data report on content types ChatGPT cites breaks it down by prompt intent. This note covers the branded slice, what it can and can't support, and the audit we'd run on a client's pages.
The branded numbers
The report covers the window September 28 to October 4, 2026, in ChatGPT, with citations classified by content type. For branded prompts:
- Product pages: 23.3%
- Landing pages: 22.8%
- FAQ: 4.8%
- Reviews: 5.6%
- Social posts: 3.0%
- Press releases: 2.1%
Product and landing pages together come to almost half of branded citations. Reviews are cited at about three times their share across all prompts, which is 1.7%. Social posts are cited at almost four times their overall share.
Across all prompts, the report found product pages are the most cited type for commercial, informational and branded intents alike. For branded prompts, they lead by a small margin over landing pages.
What this means and what it doesn't
It means that for questions about a company by name, the page that describes what the company sells is the most common citation. That's not surprising once you say it aloud: the person asked about you, and your own pages are the most direct source on what you do.
It doesn't mean your product page is cited every time. Nearly half of branded citations land somewhere other than product or landing pages, and the report doesn't say how much of that is your domain and how much belongs to third parties. It also doesn't prove that a better product page earns more citations. This is one week of one vendor's data, a classification of page types, and a correlation at best.
The reviews figure is the one we'd watch. Reviews are a small slice, but a branded question is where a buyer is checking whether to trust you, and the report shows reviews appear in that situation far more than elsewhere. The same goes for social posts, although we'd treat that one carefully because Promptwatch's social platforms report shows ChatGPT's share of social citations falling sharply from the earlier report.
The audit, without any software
Take five questions a buyer might ask by name. "What does [brand] cost?" "Who is [brand] for?" "Does [brand] integrate with X?" "Is [brand] good?" "[Brand] vs [competitor]." Then ask yourself, for each, which page on your own site answers it in a sentence a stranger could quote.
Product pages. Do they state what the product does in plain words near the top? Do they name who it's for, and what it doesn't do? Specifications, availability and limits should be current. A product page that says "Coming soon" for something that shipped a year ago will be read as true.
Landing pages. Landing pages are often written to persuade, which makes them poor sources of fact. If a campaign page carries pricing, claims or a customer count that the product page doesn't, you now have two versions of the truth. Make them agree.
Pricing. Promptwatch's content half-life report notes that pricing pages are classified as landing or product pages, so a hidden or vague pricing page is a hole in the two most cited categories. Publish what you can.
FAQ. At 4.8% of branded citations, FAQ pages are a modest source, but they're easy to write and they match how people phrase questions. Use the real questions from your sales calls, and answer each in two or three sentences.
Reviews. Check your profiles on the review sites that matter in your category. Complete them, correct the company description, and answer reviews the way you'd answer a customer.
Your own posts and press releases. Press releases are 2.1% of branded citations. If an old release has a wrong claim, such as a retired product name, fix or retire it.
Where our diagnostic work goes next
The audit tells you what's on your pages. It doesn't tell you what ChatGPT actually cites when someone asks. For that you need tracked branded prompts and the cited URLs behind them.
We run this on Promptwatch. Its prompt types include brand prompts, so you can separate the branded set from organic and competitor-comparison prompts and look at citation analytics for that set alone. Domain-level citations show whether the sources are yours or third parties', and page-level citations show which of your URLs are used. Citation Trends classifies every citation by content type and source type, which is the same lens the data report uses, applied to your own prompts.
For the "why does it say that about us" question, Brand Narrative is the feature we reach for. It pulls the claims AI makes about you and your competitors out of tracked responses, groups the repeats and files them under themes such as pricing, features, support and reputation. Each claim is scored positive, neutral, negative or mixed. The changelog lists Brand Narrative as available on every plan. A companion feature, Fact Check, lets you mark a claim accurate or inaccurate and then reads your crawled pages for ones that state it or its opposite. That one is limited to Business, Agency Scale and Enterprise, and it uses agent credits.
If a wrong claim traces back to a stale page on your own site, the fix is a page edit. If it traces to a third-party review or listing, the fix is a correction request there. Our guide on what to check when AI says something wrong about your brand walks through that split, and the note on finding which sites ChatGPT cites about your brand covers the source side.
A caution on reading the percentages
The percentages in the report are shares of classified citations, not shares of answers. A 23.3% product page share doesn't mean 23.3% of branded answers cite a product page. Keep that in mind when you quote the figure to a colleague, and keep the window in mind too. We'd rather you run your own branded prompts for a month than repeat our numbers.
None of these steps can guarantee a citation. What they do is remove the avoidable reasons that a branded answer is wrong: stale facts, conflicting pages, and review profiles nobody has touched since launch.