Which Page Type Wins Which Question

Why we looked into this

Every content team has the same argument at some point. Marketing wants more blog articles. Sales wants the product pages improved, because that is where a buyer decides. Both sides argue from instinct, because until now nobody could say which kind of page an AI engine actually reaches for.

It matters more than it used to. When a buyer searched, they saw ten links and chose one, so having any presence on the page was worth something. When a buyer asks an AI engine, they see one answer built from a handful of sources. If your kind of page is not the kind the engine reaches for, you are not in the answer at all.

So we asked a simple question with an answerable shape: when someone asks a particular kind of question, what kind of page does the engine quote?

How we tested it

We sorted all 89 questions into the kinds of thing buyers actually ask. Questions asking for a recommendation. Questions comparing one option against another. Questions about price. Questions asking how to do something. Questions asking what something is.

Then we classified every one of the 9,702 quoted pages by what kind of page it is, using its address, its title, its markup and its structure. A list of options. A head to head comparison. A pricing page. A product or service page. A blog article. A support document. A review site. A forum thread. A video.

Crossing those two together across 56,089 citations gives a direct answer: for each kind of question, which kinds of pages get quoted, and how often compared with their normal share of citations overall.

What we found

Page type tracks question type sharply, and in a way that is immediately usable.

Ask an engine about price and it reaches for a pricing page almost eleven times more often than that page type appears in general. Ask it to compare two options and comparison pages nearly triple. Ask for a recommendation and lists of options double.

The mirror image is just as clear. On how to questions, lists of options collapse from a fifth of all citations to under a tenth, and blog articles rise to nearly half. The engine is not looking for the most authoritative page in the category. It is looking for the page whose format matches the shape of the question.

Your product pages are barely in the conversation

Here is the number most companies will find uncomfortable. Across every question we tracked, product and service pages account for 5.6% of all citations. Roughly one in eighteen.

Blog articles and lists of options together account for close to half.

That is a striking imbalance against where most budget goes. Product pages get the design attention, the messaging workshops and the redesign cycles. They are also the pages an AI engine is least likely to quote when a buyer asks it a question.

The exception is worth noting, because it is the actionable part. Product pages do best on how to questions, where they reach 8.1%. When a buyer wants to know how something is actually done, a page describing how your product does it becomes relevant. That is a much narrower opening than most product marketing assumes, and a much clearer one.

Video and community threads are smaller than the noise suggests

Two categories get talked about constantly in this field and turn out to be modest in practice. Video sits at 2.8% of citations and forum or community threads at 2.6%. Together they are about one citation in eighteen.

That is not nothing, and it is worth understanding what they are doing. Video citations concentrate almost entirely on one platform, and community threads on a handful of large discussion sites. Both cluster on questions where a buyer wants to hear from someone who has actually used a product rather than from the company selling it, which is exactly where your own pages have least standing.

The practical read is to treat them as evidence rather than as a channel to chase. If AI is quoting a community thread instead of your documentation for a question about how your product behaves in practice, the useful response is usually to write the page that answers it honestly, not to go and post in the thread.

One caveat on both numbers. These are HTML pages as a crawler receives them, and a video platform serves an application rather than an article, so a video citation tells you the engine surfaced a video without telling you much about what made that video citable. We report the share and stop there rather than reading structure into a page that has none.

The unclassified fifth

Just under a fifth of citations, 18.7%, did not match any of our rules cleanly. Those are mostly one off page shapes: landing pages, category listings, PDFs, association and government pages, and internal tools. We leave the bucket visible rather than forcing those pages into a category they do not belong in, so every share above should be read as a floor for its own type.

Why this happens

An AI engine is assembling an answer, not compiling a directory. The format of a page is a strong clue about whether it contains the shape of information the answer needs.

A pricing question needs numbers laid out in a comparable way, which is exactly what a pricing page is built to do. A recommendation question needs several named options assessed against each other, which is what a list article is built to do. A how to question needs sequence and explanation, which is what an article is built to do and what a list of vendors is not.

Seen that way, the pattern is not surprising at all. What is surprising is how rarely content planning is organised around it.

What this means for your business

The first move is an inventory question rather than a writing question. Take the questions your buyers genuinely ask, sort them into these kinds, and check what shape of page you own for each one. Most companies discover they have deep coverage of one or two kinds and nothing at all for the others.

The most common gap we see is pricing. Many business software companies deliberately avoid publishing prices, which is a defensible commercial decision. The consequence in this channel is concrete: pricing questions overwhelmingly quote pricing pages, so a company without one is structurally absent from every buying conversation that starts with cost. Someone else's page answers instead, and that someone is often a review site or a competitor.

The second move is to stop expecting product pages to carry this channel. They are one in eighteen citations. That does not make them unimportant, since they are still where a buyer lands and converts. It does mean they are the wrong instrument for getting quoted, and that funding them at the expense of the formats that do get quoted is a mismatch between where the money goes and where the visibility is earned.

The third move is the cheapest. Where you already have the material but the wrong format, reformatting is far quicker than writing. A capability explained in prose across three pages, rebuilt as a single comparison with named alternatives, changes what kind of question it can win.

Mapping those gaps is the first thing we do on a new engagement at Novastacks.

Methodology

89 questions agreed with chief executives and business owners as ones their companies need to monitor, put to ChatGPT, Google AI Overview and Perplexity every day through August 2026. 6,601 answers and 61,536 citations. Every quoted page was visited and stored as its server delivered it.

Question types were assigned by consistent rules applied to the wording of each question, so the classification can be reproduced exactly. Page types were assigned by rules combining the address, the title, the markup and the structure of the page, with each page receiving a single type. 18.7% of pages fall into a general category that no rule matched, so the shares above describe clear cases and the true figures for each named type are, if anything, slightly higher. All brands are anonymised.

Frequently Asked Questions

Which page types does AI quote most often?

Blog articles at 29.3% and list style articles at 20.0% together account for close to half of all citations. Comparison pages follow at 9.2%, then review and directory sites at 5.7% and product or service pages at 5.6%.

Do product pages get cited by AI?

Rarely. Product and service pages account for 5.6% of all citations, roughly one in eighteen. They perform best on how to questions, where they reach 8.1%.

What kind of page wins pricing questions?

Pricing pages, by a wide margin. They make up 29.8% of citations on pricing questions against a 2.8% baseline across all questions, which is 10.6 times over represented.

Do listicles still work for AI search?

On recommendation questions, yes. List style articles reach 42.5% of citations there against a 20.0% baseline. On how to questions they collapse to 8.5%.

Does AI cite YouTube and forum threads?

Yes, but less than the conversation around them suggests. Video accounts for 2.8% of citations and forum or community threads 2.6%, so together about one citation in eighteen. Both cluster on questions where a buyer wants to hear from someone who has used a product rather than from the company selling it.

How was this measured?

89 questions agreed with chief executives and business owners were put to ChatGPT, Google AI Overview and Perplexity daily through August 2026, producing 6,601 answers and 61,536 citations across 9,702 pages. Each page was classified into exactly one type by rules combining its address, title, markup and structure.

Which Questions Do You Have No Page For?

We map the questions your buyers ask, sort them by what kind of answer they need, and show you exactly where you have no page in the running.

Run the Free AEO AuditExplore AEO Services