Which Content Formats Get Cited Most Often by AI?

What does the research cover?

Two large studies in 2026 looked at formats directly, and others at related factors:

Which content formats perform best in AI search results?

The Wix and Peec study's breakdown by intent is the clearest answer available:

Three lessons follow. Informational questions are won by articles and guides. Commercial questions, such as "best CRM for small teams", are won by listicles, most of them on third-party sites. And questions close to purchase are won by product and category pages. A brand needs different page types for different stages, not one favourite format.

What patterns in 9.5 million AI citations reveal about preferred content?

Meltwater's analysis of 9.5 million citations in B2B categories looked at which LinkedIn content AI cites. It found:

• Structure: the top-cited LinkedIn articles all used bullets, and 92% used clear headings. 75% named specific entities, such as tools, companies or categories, and 67% included numbers.
• Decision support: "best X" lists appeared in 54% of the top-cited content, side-by-side comparisons in 50%, and how-to-choose guides in 33%. Pure opinion was less likely to be cited on its own.
• Text over media: plain text posts and articles made up 83% of cited content.
• Freshness and originality: 72% of cited content was original rather than reshared, and 48% had been published in the previous three months.
• Beyond your own site: user-generated platforms accounted for 47.5% of all citations, peer review sites for 15.0%, and company websites for 18.7%.

The last point matters as much as format. Most citations in these B2B answers pointed somewhere other than the brand's own site.

What types of websites and content get cited by ChatGPT?

A mix, and the mix shifts over time. Across studies, ChatGPT cites encyclopaedic sources such as Wikipedia for facts, review and comparison sites for product questions, professional platforms such as LinkedIn for B2B topics, video platforms, news, and brand websites for specific product details. Semrush's three-month tracking in 2025 found that Wikipedia's share of ChatGPT responses fell from about 55% to under 20% within weeks, a reminder that any single source's share can move quickly. For content types, ChatGPT follows the same intent pattern as other assistants: articles for learning, listicles and reviews for comparing, product pages for buying. Why Wikipedia features so often is covered in why Wikipedia is so often cited for factual answers, and page types in which types of pages LLMs cite.

Do LLMs and AI Overviews prefer lists, tables or schema?

• Lists: they are common in cited content. In Meltwater's sample of LinkedIn content, every top-cited article used bullets. Lists make items easy to extract and quote.
• Tables: no large study isolates tables. They suit comparisons and specifications, where they make the data easy to read, but a table on the wrong kind of page adds little.
• Schema: Ahrefs' controlled study found that pages adding JSON-LD saw no measurable change in ChatGPT or AI Mode citations compared with similar pages, and a small decline in AI Overviews that it could not attribute to the schema. Structured data still supports rich results, but it is not a citation lever on its own.

In short, visible structure that helps a reader, such as lists, headings and tables where they fit, is associated with citations. Invisible markup alone is not.

Do tables get cited more often for comparison queries, or does the format need to match the intent?

The format needs to match the intent. The Wix and Peec data shows that dedicated comparison pages earned only 4.1% of citations even for commercial questions, and alternatives pages under 1%. What won commercial questions were listicles that compare several options, usually on independent sites. A table helps when the question is itself a comparison of attributes, such as prices, features or specifications, because it lays out exactly what the answer needs. It does not turn an informational article into a commercial winner. Build the page type that fits the question first, then use a table inside it where comparison data is central. When to publish comparison pages at all is covered in whether to publish comparison and alternative pages for AI search.

Do bullets and headings help AI cite content more than long paragraphs?

The evidence points that way, with caveats. Top-cited content in large samples is heavily structured, and a lab study of generative engine optimisation found that adding statistics, quotations and cited sources raised visibility by 30 to 40% in its benchmark. Structure helps because an assistant can find the relevant part quickly and quote it cleanly. But Google says there is no need to break content into artificial chunks for its AI features, and a well-written paragraph that answers the question directly is perfectly quotable. The goal is clarity: a heading that states the question, an answer in the first sentence or two, and lists where there really is a list.

How to structure blog posts with FAQs, lists, and short sections for AI citations

A structure that matches what cited content tends to share:

1. A direct answer at the top. Two to four sentences that answer the title question, with the key number or recommendation.
2. Question-style headings. Each section answers one question a reader would ask next.
3. Short sections. Open each with the answer, then give the detail and the evidence.
4. Lists for real lists. Steps, options, criteria and pros and cons.
5. Named entities and numbers. Name products, places and sources, and give figures with their origin.
6. A table where comparison data is central.
7. An FAQ at the end with three to five short questions not already covered.
8. A visible date and named author, and a habit of updating the post when facts change.

How to write the opening answer is covered in how to write direct answer-first content for AI search.

Are category buying guides the best content for AI citations?

For early research, often yes; for "which should I buy", usually not your own. Buying guides behave like articles and how-to guides, which lead for informational questions such as "what to look for in a standing desk". For commercial questions such as "best standing desk under $500", the Wix and Peec data shows listicles winning. Among the top 1,000 cited URLs in professional services, third-party listicles outnumbered self-promotional ones about four to one, although Ahrefs found self-ranked "best X" lists are also among ChatGPT's most prominent sources. So a brand's own buying guide can win the learning stage, while the recommendation stage is mostly won by being included in independent lists and reviews.

How do I create a buyer guide that ChatGPT recommends for my category?

Write it to be useful to a buyer who may not choose you:

1. Answer the category question first, such as what matters most when choosing, in two or three sentences.
2. Set out criteria with numbers: price ranges, capacities, specifications, typical costs.
3. Name real options, including competitors, with who each suits and where each falls short. A guide that only recommends its publisher reads as advertising.
4. Make trade-offs explicit, such as cheaper but slower, or more features but harder to set up.
5. Add a how-to-choose section and a short FAQ.
6. Cite sources and show your testing where you have it; say plainly where you have not tested.
7. Update it quarterly, since nearly half of the LinkedIn content cited in the Meltwater sample was under three months old.
8. Pair it with outreach, so independent sites that publish lists in your category know about your product.

What about length and freshness?

Length matters little: Ahrefs found almost no link between word count and AI Overview citations. Freshness matters more: Ahrefs found that URLs cited by AI were on average about a quarter newer than those in organic results, and ChatGPT favoured the newest. Keep important pages current rather than long. More detail is in whether word count affects AI citations.

How should you audit your existing content against this?

A one-afternoon review that turns the research into a to-do list:

1. List your 20 most important buyer questions, and label each by intent: learning, comparing, buying or finding a specific business.
2. Match each question to a page type. Learning questions need an article or guide; comparing questions need a list or review, often on another site; buying questions need a strong product or category page; questions about finding a specific business need a clear homepage and accurate directory profiles.
3. Find the gaps. Questions with no matching page on your site, or only a page of the wrong type, are where you are least likely to be cited.
4. Check structure on existing pages. Does each open with a direct answer? Do headings state questions? Are there named options, numbers and sources?
5. Check the date. Pages not updated in a year are at a disadvantage against fresher competitors.
6. Look outside your site. For each comparing question, note which independent lists and review sites assistants cite, and whether you are on them.

Then run the same questions in two or three assistants, record what is cited, and repeat after your changes. That tells you whether format changes, or outside coverage, made the difference.

What might this look like in practice?

Consider an illustrative standing-desk brand, Keststead. Its blog had long opinion essays, and it was rarely cited. It rebuilt three pages: a buying guide with criteria, price bands and named options including competitors; a specifications table on each product page; and an FAQ for delivery and assembly. It also sent review units to four independent sites that publish "best standing desk" lists. Three months later, its buying guide was cited in answers to "how to choose a standing desk", its product pages in answers about specific models, and two of the independent lists that now included Keststead were cited for "best standing desk under $500".

What do people ask about content formats and AI citations?

Which format gets cited most by AI?

It depends on intent: articles for informational questions, listicles for commercial ones, and product or category pages for buying and navigational ones.

Does FAQ schema help AI citations?

A controlled Ahrefs study found no measurable lift from adding schema, including FAQ markup. Visible FAQs that answer real questions are still useful to readers.

Are listicles still worth writing?

Independent listicles are cited heavily for commercial questions. Self-promotional lists are cited too: Ahrefs found "best X" lists, including ones where the publisher ranks itself first, were the most prominent page type in ChatGPT sources, although among the top 1,000 cited URLs in Wix and Peec's professional-services data, third-party lists outnumbered them about four to one.

Does video get cited?

Video platforms are cited often: YouTube received 1.52% of all citations in Meltwater's 9.5-million-citation study, nearly three times LinkedIn's share. Within the LinkedIn content that study analysed, plain text posts and articles made up 83% of what was cited.