"AI citation optimization" gets sold as a page-level trick — add schema, write an answer-first paragraph, wait to get quoted. The format-share data says the real lever sits one level up: which content formats you produce, and in what proportion. A 2026 study from the agency Ten Speed, written up by Nelson Brassell, tagged 7,387 citation appearances across 170 B2B vendor-evaluation prompts (monitored through Peec AI, which watches ChatGPT, Perplexity, Claude, and Gemini) by the type of page each citation pointed at. Product pages led at 24.1%, articles at 17.4%, comparison pages and listicles at roughly 13% each, how-to guides near 9%, homepages 7.8%, third-party directory profiles like G2 and Capterra 7.2%, and Reddit, YouTube, and forums combined just 4.2%. The headline everyone repeats — product pages 24%, Reddit and YouTube 4% — is real, but it is a fingerprint of one stage, not a universal ranking. That same 4% flips to a leading share on consumer and informational queries, where separate studies put Reddit and YouTube among the most-cited sources of all. This guide reads the whole finding rather than the headline. It sizes what the Ten Speed study actually measured and where it does not generalize, explains why query intent — not a leaderboard of formats — decides which content an engine reaches for, walks what each format is genuinely good at earning a citation on, and then makes the argument the page-level advice skips: AI citation optimization is a portfolio problem. You are not trying to win one format; you are trying to hold a mix of formats that covers the intents your buyers actually trigger, because the engine picks a different shape of source for every question. The teams that get cited most are not the ones with the best single page — they are the ones whose format spread happens to have the right answer already published for whatever gets asked.
Most explainers frame AI citation optimization as a checklist you apply to one page: add schema, open with a direct answer, keep passages self-contained, unblock the retrieval crawlers, keep the date current. All of that is real and all of it matters — it is what makes an individual page eligible to be lifted into a synthesized answer, and the levers that decide whether an engine quotes you are worth getting right. But the checklist has a ceiling, and the ceiling is the reason two sites can both do the on-page work well and get cited at wildly different rates. Eligibility is not selection. A perfectly structured page still loses the citation if the engine is answering a question that page is the wrong shape to answer.
The higher-leverage version of AI citation optimization works one level up, at the portfolio. An answer engine does not pick a winner from a ranked list of pages the way classic search did. It decides what kind of source a given question needs — a comparison, a how-to, a product spec, a first-hand opinion — and retrieves passages of that shape. So the question that actually governs your citation rate is not "is this page optimized" but "do I publish a source of the right shape for the questions my buyers ask." That reframes the whole practice: you are not optimizing a page, you are engineering a mix. The rest of this guide is about what the data says that mix should be, and why the honest answer is "it depends on your buyers, not a leaderboard."
The number circulating hardest in 2026 comes from a study by the agency Ten Speed, written up by Nelson Brassell. It monitored citations through Peec AI — a tool that watches how ChatGPT, Perplexity, Claude, and Gemini cite URLs — across 170 prompts built to mirror B2B vendor-evaluation queries, and tagged the resulting 7,387 citation appearances by the type of page each one pointed at. The distribution: product pages 24.1%, articles (blog, news, and PR) 17.4%, comparison pages and listicles roughly 13% each, how-to guides just under 9%, homepages 7.8%, third-party directory profiles like G2 and Capterra 7.2%, and Reddit, YouTube, and forums combined 4.2% — with Reddit the majority of that and YouTube around 1%. That is the source of the tidy "product pages 24%, Reddit and YouTube 4%" headline.
Read the methodology before you build a strategy on the number, because the study is narrower than the headline implies. The 170 prompts all sit at one stage — vendor evaluation, the point where a buyer is comparing named competitors, not learning a category. The clients behind the citations were B2B SaaS and professional-services firms in four verticals: fintech, physical security, hospitality, and IT automation. And the authors were explicit that the 24%-versus-4% split was descriptive only, not statistically tested, and that platform-level differences between the four engines were not separated out. None of that makes the study wrong. It makes it a precise measurement of one slice of the buying journey in a handful of industries — which is exactly the right way to read it, and exactly what the headline strips away.
Set it next to the larger consumer-facing studies and the picture sharpens. When Wix's AI Search Lab classified 1.06 million citations across ChatGPT, Google AI Mode, and Perplexity, product pages came third at 13.7%, behind listicles at 21.9% and articles at 16.7%. DeltaV Digital's separate 25,337-citation study also ranked product pages third, at 16.3%. So product pages are genuinely a top-three format everywhere — but their share nearly doubles in the B2B vendor-evaluation slice Ten Speed measured, because that is the exact moment a product page is the right shape of source. The 24% is not a contradiction of the other studies. It is what the format mix looks like when you zoom all the way into one high-intent stage.
The most consequential thing in all of this data is the part that looks like a footnote: the 4.2% for Reddit and YouTube. Taken as a universal claim, it is badly wrong — and knowing why is the whole key to AI citation optimization. In consumer and informational contexts, Reddit and YouTube are not marginal; they are dominant. Separate 2026 analyses of AI answers found Reddit and YouTube among the most-cited sources of any kind, with YouTube overtaking Reddit as the single most-cited social source in some datasets. Same platforms, near-opposite share, and nothing about the platforms changed. What changed was the question.
That is because citation share by format or source is not a property of the format. It is a property of the query intent the engine is serving. B2B vendor-evaluation prompts — Ten Speed's slice — are the worst possible case for community content: when someone is comparing named enterprise vendors, an engine reaches for product pages, comparison content, and analyst-style directories, and a forum thread rarely carries the structured, current facts that answer needs. Ask the same engine an experience question — "is this actually worth it," "what do real users think," "how do I fix this in practice" — and Reddit and YouTube surge, because now the answer needs first-hand opinion and demonstration, which is exactly what those platforms hold. The distribution you measure is a fingerprint of the intents in your prompt set, and a different prompt set prints a different fingerprint.
The practical consequence is that there is no format leaderboard to copy. A study's format ranking is only transferable to you insofar as your buyers ask the same distribution of questions the study measured. A B2B SaaS company selling into procurement should weight product pages and comparison content heavily, because its buyers really do live near the Ten Speed fingerprint. A consumer brand whose buyers ask experience and how-to questions should weight articles, how-to guides, listicles, and the social and video surfaces where those get answered — and treating the 24% product-page number as its north star would point it at the wrong formats entirely. The optimization is to find your own fingerprint, not to import someone else's.
If the mix is what you optimize, you need to know what each format wins, so you can match formats to the intents in your demand. The pattern that holds across the 2026 studies lines up cleanly with the buyer's journey.
Explanatory prose wins the learning stage: "how does X work," "how do I choose," "what is the difference between." Articles took 17.4% in the Ten Speed set and 16.7% in the Wix set, and how-to guides add roughly 9% more; together they own the questions asked before anyone is comparing vendors. This is the formats-for-citation layer covered in depth in the guide on content formats that get cited — the durable point is that an informational query will almost never return a product page, so if your buyers ask a lot of "how" questions and you only publish product pages, you are invisible for the top of your own funnel.
The "best X for Y" and "X vs Z" queries are won by listicles and comparison pages, and the effect is large: the Wix study found listicles captured around 40% of commercial-comparison citations, nearly double their overall share. In the Ten Speed vendor-evaluation set, comparison pages and listicles landed near 13% each — and critically, a listicle is often where a product page earns its exposure, because the roundup that wins the comparison query is what surfaces the individual product into the engine's candidate set. Treat listicles and comparison content as the connective tissue of the portfolio, not a lesser format.
This is where the 24% lives, and it is intent-specific. Product pages dominate when a buyer is close to a decision and the engine assembles a shopping-style answer from current, structured product data — the queries and mechanics are covered in AI citations for product pages. They win because at the moment of decision, the product page is the only source carrying the precise, current, structured facts the answer requires. They fade to near-zero on informational queries. So a product page is a high-value asset for a narrow, valuable band of your demand — and a poor bet for anything above it.
Third-party directory profiles (G2, Capterra, and the like) took 7.2% in the B2B set and function as independent corroboration — the outside source an engine cross-checks a vendor claim against. Reddit, YouTube, and forums are the experience and demonstration layer, marginal for enterprise evaluation but leading for consumer and how-to intents. Neither is content you fully own, which is the point: a complete portfolio includes earning presence on surfaces you influence but do not publish, because AI answers cite mostly third-party sources — the argument in which sources AI engines actually cite.
Putting it together, AI citation optimization at the portfolio level is a four-part loop. First, fix the set of real, conversational prompts your buyers ask an assistant, and tag each by intent — informational, commercial-comparison, transactional, experiential. That tagged set is your demand distribution, and it is the only leaderboard that matters. Second, map each intent to the format that wins its citations using the pattern above. Third, audit what you currently publish and what currently gets cited, tagged by the same format buckets, to expose where your production mix does not match your demand mix. Fourth, produce against the gap — and re-measure, because the mix drifts as competitors publish and as engines change what they reach for.
The failure mode this prevents is the most common one in the field: over-indexing on a single headline number. A team reads "product pages get 24% of citations" and pours effort into product pages while its buyers are mostly asking informational and comparison questions those pages cannot win — earning a bigger share of a slice of demand it was already covering, and staying invisible for the majority it was not. The discipline is to let your demand fingerprint set your format allocation, treat every published study as a prior to be checked against your own data rather than a target, and measure citation share per format so the allocation self-corrects. The step-by-step version of this loop is laid out in how to prioritize content formats for AI citations.
Be exact about the division of labor first, because it is the honest framing and it is different from the on-page work. AI citation optimization has two halves. One is the page-and-data half — your Product schema, your merchant feed, your site's technical extractability — which lives on your own site and is a job you do yourself; Kompozy does not touch it. The other is the portfolio-production half: actually publishing, and keeping current, a spread of formats wide enough to cover the intents your buyers trigger. That second half is where most portfolios break, because holding a real format mix — articles for the informational top, comparison and listicle content for the commercial middle, how-to guides, plus the social and video surfaces where experience queries get answered — is a volume-of-production problem that a small team cannot hand-build across every format at once. That is the specific problem Kompozy is built for.
It matters that Kompozy is a full generation-and-publishing engine rather than a single-format tool, because a portfolio strategy is a multi-format problem by definition. From one source and one Persona Brief, it generates the Blog Articles that win the informational top, the comparison-style and listicle posts that win the commercial middle, the how-to and text posts the middle of the funnel reads, the carousels and quote graphics that populate the social feeds engines increasingly cite, and the Persona Shorts and other persona video that answer the experience-and-demonstration queries where Reddit and YouTube-style content wins. One team can hold a wide format mix instead of picking the one format it has bandwidth to maintain — which is the difference between covering your whole demand fingerprint and covering a corner of it.
The consistency that comes free from generating the whole spread from one brief is itself a citation lever, not just an efficiency. When an engine cross-checks a claim across your article, your comparison post, your social card, and your video, the entity, the numbers, and the positioning have to line up or the engine treats your data as unreliable and moves on — the same cross-source trust test that decides product-page citations. Because a single Persona Brief and HyperFrames govern voice and visual brand across every format, that consistency holds by construction instead of fracturing across a dozen hand-built assets. Autopilot then fans the spread across eight social platforms plus blog and email on a cadence, behind a per-post review gate where a human approves every claim before it ships — the accuracy check that matters most when the entire goal is to be the source an engine quotes correctly. Kompozy will not write your schema, run your visibility tracker, or decide your format allocation; what it removes is the production ceiling that leaves most portfolios covering one format well and the rest not at all.
AI citation optimization is not a page-level trick, and the format-share studies are the proof. Yes, product pages took 24.1% of citations in Ten Speed's 2026 B2B vendor-evaluation study while Reddit and YouTube combined for 4.2% — but that is a fingerprint of one buying stage in four verticals, descriptive only, and the same social platforms lead the citation counts the moment the query turns consumer or informational. There is no format leaderboard to copy. There is a different format that wins each intent, and an engine that picks by intent every time it answers. So the optimization is a portfolio one: find the distribution of questions your buyers actually ask, map it to the formats that win those questions, and hold that mix — kept current and kept consistent — rather than betting everything on the one number that made a headline. The teams that get cited most are the ones whose format spread already has the right-shaped answer published for whatever gets asked.
AI citation optimization is the practice of shaping your content so answer engines like ChatGPT, Perplexity, Google's AI Overviews, and Gemini quote and link you when they synthesize an answer. Most advice treats it as a page-level job — schema, an answer-first paragraph, clean crawlability — and that part matters. But the higher-leverage version works at the portfolio level: deciding which content formats you produce, and in what proportion, so the engine can always find a source of the right shape for the query it is answering. Different formats win different questions, so the mix is the strategy, not any single optimized page.
In one specific study, yes. The agency Ten Speed, in a 2026 analysis by Nelson Brassell, tagged 7,387 citation appearances across 170 B2B vendor-evaluation prompts monitored through Peec AI and found product pages the single most-cited format at 24.1%, with Reddit, YouTube, and forums combined at 4.2%. But that figure describes the vendor-evaluation stage of B2B buying — the moment someone compares named competitors — in four verticals. The authors flagged it as descriptive only, not statistically tested. Larger consumer-facing studies put product pages lower (around 13.7% to 16.3%) and rank listicles and articles higher, so treat 24% as a stage fingerprint, not a universal law.
Because the 4.2% figure is specific to B2B vendor-evaluation prompts, and community sources are weakest exactly there. When someone is comparing named enterprise vendors, an engine leans on product pages, comparison content, and analyst-style directories, not forum threads. Flip the query to consumer or informational — "is X worth it," "how do I do Y" — and separate 2026 studies find Reddit and YouTube among the most-cited sources of all, with YouTube overtaking Reddit as the top social source in some datasets. Same platforms, opposite share, because the query intent changed. Citation share by source is a property of the question, not the source.
There is no single answer, and that is the point — you produce the mix that matches the questions your buyers ask. Informational questions ("how does X work," "how do I choose") are won by articles, how-to guides, and listicles. Commercial-comparison questions ("best X for Y," "X vs Z") are won by listicles and comparison pages. Transactional, buyer-close questions are won by product and category pages. Community platforms win experience-and-opinion questions. Map your real prompts to those intents, see which formats they demand, and build toward that distribution rather than betting on one format.
It is necessary and insufficient. Clean structure — valid schema, a plain answer stated up front, self-contained passages, unblocked retrieval crawlers, current dates — is what makes an individual page eligible to be quoted, and skipping it guarantees you lose. But eligibility is not selection. If the engine is answering a comparison query and you only publish product pages, no amount of on-page polish gets you cited, because you do not have a source of the shape the answer needs. The page-level work decides whether a given page can be cited; the format portfolio decides whether you have a citable page for the query at all.
Track citation share by format, not just an overall mention rate. Run a fixed set of your buyers' real prompts through the major engines on a cadence, record which of your URLs get cited, and tag each cited URL by page type. That tells you two things a single visibility number hides: which formats are actually earning your citations, and which intents you are invisible on because you have no source of the right shape. The gap between your demand mix (what buyers ask) and your citation mix (what gets quoted) is the specific thing to close.
AI citation optimization is the practice of shaping your content — at the page level and, more decisively, at the portfolio level — so AI answer engines quote and link you. The 2026 study data shows no single format wins: product pages, articles, listicles, comparison pages, and how-to guides each dominate a different query intent, and the winning mix shifts by buying stage and vertical. The lever most advice misses is the format mix itself. You optimize by holding a spread of formats that covers the intents your buyers actually trigger, because the engine reaches for a different shape of source for every question it answers.
Get started → · ← All guides · Compare Kompozy vs other tools