Your Homepage Is Doing More Brand Work Than Your Blog
ChatGPT cites brand homepages 33.1% of the time and roundups 6.8%. Perplexity inverts it. Data from 21,508 AI citations on where brand content actually gets read.
Somebody asks ChatGPT which company to use for the thing you sell. The answer names three brands and links a handful of sources. You will never see that exchange, but the sources it picked are the closest thing there is to a ranking of where your brand content is actually being read.
We have 21,508 of those sources, labeled by page type. The results argue for moving brand content somewhere most teams have stopped investing in it.
The corpus
VerifiedDR asks tracked buyer questions of ChatGPT, Perplexity and Google AI Mode on a schedule and keeps every source each answer cites. In August 2026 the whole archive was classified by page type: 16,322 distinct pages across 21,508 citation events.
One counting note first, because it changes the headline. Measured per distinct URL, the mix reads 42% homepage, since 1,884 of those citations point at a bare hostname and each one counts once. Weighted by citation events, which is what a reader actually encounters in an answer, it looks like this:
| Page type | Share of citation events |
|---|---|
| Listicle / roundup | 17.1% |
| Homepage | 14.6% |
| Product page | 12.0% |
Then split it by engine
The average is hiding the interesting part. On the two biggest slices, the engines are close to mirror images of each other.
| Page type | ChatGPT | Perplexity |
|---|---|---|
| Homepage | 33.1% | 12.1% |
| Listicle / roundup | 6.8% | 20.0% |
ChatGPT goes to the brand's own homepage a third of the time and barely touches roundups. Perplexity does the reverse.
One caveat worth stating plainly: ChatGPT is 3,294 of those 21,508 events, because it only records sources on full paid runs. Perplexity carries most of the corpus. Treat the ChatGPT column as directional rather than precise.
What this means for brand content
For the last decade the received wisdom has been that the homepage is a signpost. Keep it short, push people into the funnel, and put the substance in blog posts and landing pages where it can rank for something.
An AI answer does not work that way. When it needs to describe what a company is, it goes to the page the company uses to describe itself. That is one third of ChatGPT's citations pointing at a page most brands treat as a menu.
Three consequences.
Your homepage has to answer the question, not tease it. What you do, who it is for, what it costs, what it connects to. If the model has to infer your category from a tagline and three testimonials, it will infer it from somebody else's page instead.
Product pages are the second surface, at 12.0%. Same rule. A page that lists features without saying what the thing is cannot be quoted.
The blog is not where this is decided. Guides and how-tos still matter for people, but in citation terms they sit well below the two pages you probably update least often.
Where the roundups still matter
None of this makes placement work pointless. It makes it engine-specific.
If your buyers use Perplexity, one citation in five is somebody else's comparison post and your homepage barely registers. Getting named in the directories and roundups of your category is the actual work there, and it looks more like digital PR than content marketing.
So the honest answer is that it depends on a fact most teams have never checked: which engine their buyers actually ask. Ten questions, asked manually in both, logged out, with the date written down, settles it in an afternoon and costs nothing.
The video and image angle
Worth naming, because it is the part that gets over-claimed. A generated clip does not get cited. Nothing in this corpus suggests that publishing more brand video moves you into an AI answer directly.
What brand media does is hold attention on the pages that do get cited, and give people a reason to name you somewhere a model will later read. Treat it as demand and recall work, not as citation work, and put the citable substance in text on the same page.
Measuring it without inventing a metric
Resist building a single AI visibility score for the board. Generation is sampled and engines personalize, so the number will move for reasons nobody can explain and you will spend a quarter defending noise.
Report three things instead. The share of unbranded buyer questions where you appear. The cited sources you do not own, which is your work list. And any factual error the model repeats about you, which is reputation exposure with no equivalent in classic reporting.
If you want the recurring version rather than a manual check, VerifiedDR tracks it across those three engines with every answer shown verbatim and the cited pages listed.
