The pages AI cites: listicles take two thirds, product pages a quarter, everything else is noise

A marketing lead types a buyer question into an AI tool before a meeting. The answer names several products, but the useful detail sits in the linked sources. The practical question is simple: what pages does AI cite?

CiteGraph’s weekly research on who owns AI answers reports a clear pattern. Listicles and roundups account for most recorded citations, competitor product pages take much of the rest, and the scanned product’s own site appears far less often.

The sample covers 64,667 citations

The sample contains 64,667 citations across 9,791 AI answers. Those answers came from 77 scans of 44 sites in 48 categories, covering 507 distinct buyer questions. The engines cited 21,465 distinct pages across 9,192 domains.

Listicles and roundups account for 68.3% of recorded citations, or 44,164 citations. Competitor product pages account for 24.1%, or 15,575 citations.

For a publisher, the implication is direct. A product page can explain what you sell, but it is rarely the page doing the comparative work inside an AI answer. You need pages that help an engine assemble a choice, alongside pages that describe your own offer.

Listicles carry most citation weight

Listicles and roundups are the main source type in this sample. One plausible reason is that they gather options and state differences, which may make them easier to cite. The data cannot confirm that this is why.

A useful roundup should state who each option suits, where it falls short and which buying condition changes the choice. The selection criteria need to be visible, with important claims in ordinary text rather than buried in slogans.

Given that listicles and competitor pages take 92.4% of citations here, start by checking whether your category has that same split before assuming it does. The leaderboards show which domains recur across categories, although domain presence alone does not establish page fit.

Product pages fill the second lane

Competitor product pages account for 24.1% of recorded citations. These are pages about products that compete with the product being scanned, rather than the scanned product’s own site.

A competitor’s product page serves as sales copy and provides additional information. It contains a product’s features, audience, limits or positioning in a form that can sit beside alternatives in an AI answer.

The scanned product’s own site accounts for 1.7% of citations, equal to 1,107 citations in this sample. That is a small share beside the 24.1% assigned to competitor product pages.

Keep your product pages factual and easy to quote, then publish comparison and alternative pages that answer the buyer’s surrounding question. The second layer is where the sample shows most of the citation share.

Smaller sources still matter

The lower-share page types still have jobs. Documentation can cover implementation detail, a community thread can expose first-hand experience, and news can provide current context.

The domain view adds another useful fact. The ten most-cited domains account for 8.4% of all citations, so the sample does not reduce to a handful of websites. Reddit is cited 1,885 times across 47 categories, LinkedIn 512 times across 33 categories, and GitHub 509 times across 27 categories.

That spread matters for content planning. You are not trying to imitate one winning domain. You are trying to publish pages that fit the source types appearing in your category, while making the evidence clear enough to survive comparison.

A page can also be useful without becoming a frequent citation source. Give documentation and community content a clear job, rather than producing them because a content calendar has an empty square.

Evidence beats page polish

A concrete before-and-after starts with a page whose pricing is available only after a demo, while its comparison section never says whether the product replaces a spreadsheet, another tool or an internal process. Its feature list mentions “workflow automation” and “team reporting”, but gives no buyer condition that separates one use case from another.

After the change, the page keeps its feature facts and adds “Who it replaces”, “When it is a poor fit” and “Alternatives”. It states the pricing route plainly, separates capabilities from limits and adds a comparison table organised around use cases rather than slogans.

That is a more useful page for buyers and systems extracting evidence. It gives the category, the trade-offs and the comparison in ordinary text.

The product page supplies facts. A comparison page explains trade-offs. A roundup brings several options into one frame. Each page has a job, and the jobs connect.

The methodology page explains how citations are counted. We will not restate mechanics that the ledger does not verify here.

The sample has hard limits

We cannot tell you that publishing a listicle will produce a citation. We cannot tell you that a competitor page is better than your page in every category, or that a page type with a low share has no value.

The sample measures citations from 77 scans of 44 sites in 48 categories. It covers 507 distinct buyer questions, not every question a buyer can ask and not every engine response in the wild. We measure samples, we print the margins, and we do not know what we have not measured.

We also cannot infer causation from the breakdown. Listicles may be cited because they answer comparison questions well, because they are easy to extract from, because they are common, or because several of those things happen together. The data shows the source type attached to the citation. It does not isolate the reason.

The caveat that matters most is simple: page type creates an opportunity, not a guarantee. Your result still depends on the question, the page and the engine that answers it.