Which content formats does AI actually cite? 2021–2030

• 34%Listicles' share of article citations in 2026. Still #1.
• 12→20%Original research, 2026 to 2030. The fastest climber.
• 45%Share of all AI citations going to articles. The rest go to product pages, Reddit, YouTube and more.

That viral “AI citation” table? Nobody can trace it.

A table claiming listicles get 32% of AI citations has been going around. We looked for its source and couldn't find one. No published study uses its 10 categories, its rows add up to 98%, not 100%, and its “2.7x” figure comes from a different study about a different format. It does get one thing right: listicles are #1 in every study we found.

• 21.9%Listicles' share of all citations in the largest study: Wix and Peec AI, 1.06M citations across ChatGPT, AI Mode and Perplexity.
• 40–63%Listicles' share on “best X” and recommendation prompts. The query decides the format.
• 0Studies with a separate “original research” category big enough to test the viral claim.

Ten years of what AI cites

Share among the 10 formats; each year adds to 100%. 2021–22 use Google featured snippets as a stand-in (AI citations didn't exist yet), 2023–24 are interpolated, 2025–26 are anchored to measured studies, and 2027–30 are a forecast.

• FAQs collapsedFrom 32% in 2021 to 5% in 2026. AI now answers simple questions itself, without citing anyone.
• Listicles peakedUp from 12% to 35% in 2025, now easing as Google and ChatGPT cool on self-promotional “best of” lists.
• Original data climbsFrom 2% to 20% by 2030, becoming #2. AI favours sources that own a fact.

Every format, every year

Share of AI citations among the 10 formats, in %. Shaded columns are forecasts. “Viral” shows the circulating figures for comparison.

What each period rests on

• 2021–22: proxy — Google featured snippets (low confidence)
• 2023–24: early AI (SGE, Bing Chat, Perplexity); no format data (low)
• 2025–26: anchored to measured studies (medium-low)
• 2027–30: forecast (low, then very low)

Confidence falls the further out the forecast goes, because shares have swung by double digits within months after a single AI model update.

Will original research take 30–50% of AI citations?

Original research's share of the 10 formats under four scenarios. The shaded band is the 30–50% thesis.

• Strictly: noA page that is just a data studyReaches about 20% of article citations by 2030 and becomes #2, not #1. Even the bull case tops out near 32%. As a share of all AI citations, it's only about 8.4%.
• Broadly: yesYour own data as the ingredientCount benchmarks, tested comparisons, case studies and listicles built on first-hand data, and that group reaches about 42% by 2030. The lists that keep winning will be the data-backed ones.

Articles get less than half of what AI cites

Estimated share of all AI citations by source type. Product and category pages keep gaining as AI agents start shopping for people, while Wikipedia's share shrinks.

• 5.4%Original research's share of all AI citations today: 12% of articles × 45% article share.
• 8.4%The same figure in 2030, base case. Big growth from a small base.
• 22%Product, category and homepages by 2030, the biggest non-article source.

What would make original research win big

1. AI engines extend their shift toward primary sources (PubMed jumped from #64 to #4 in ChatGPT citations in 2026) to data that brands publish.
2. They cite the original study, not the “100+ statistics” roundup quoting it.
3. ChatGPT and other AI engines, not just Google, keep penalising self-promotional listicles.
4. More prompts become data-seeking as AI agents do research for people.

Six moves for creators and businesses

1. Run small, recurring studies. Benchmark 50 local businesses' reviews, track which queries show AI Overviews, time how fast Google indexes new pages. Data no AI can make up.
2. Publish data as a named benchmark. Give it a name, a method box and tables. Benchmarks with named options get cited; raw data dumps mostly don't. How to build one: the Citation Hub playbook.
3. Get into other people's lists. Most listicles AI cites are written by third parties. Don't rank yourself #1 on your own “best of” page; that tactic lost visibility in 2026.
4. Refresh every quarter. Pages left un-updated lose citations fast; about 70% of pages cited in AI Overviews change within 2–3 months.
5. Keep listings and reviews clean. For local searches, business listings and reviews take a large share of citations. Your Google Business Profile is content too — and fake 1-star reviews can be removed.
6. Post your video on YouTube too. YouTube is one of the most-cited sources in AI Overviews. One study can feed a reel, a thread, a LinkedIn post and a video.

Sources and method

Every year-by-year figure is an estimate we built from the studies below. None of them measures all 10 formats directly. All are vendor or agency studies except the GEO paper, which is academic.

Assumptions

• The 10 formats are treated as covering all article citations. Real pages mix formats; a “best X” list with original tests counts as a listicle.
• Reddit, Wikipedia, YouTube, product pages, news and listings are removed before each year is scaled to 100%.
• The mix of informational, shopping and local prompts stays roughly stable to 2030.
• Numbers are global and English-language. No India-specific citation data exists yet.