What Content Do AI Models Cite Most: Listicles, Comparisons, or Brand Content?

The Short Version, By Platform

No single AI engine behaves like another. The three major citation datasets available all agree on one thing: brand pages lose to third-party content. But which third-party content wins differs by platform, and it shifts over time.

Source: Profound, AI Platform Citation Patterns (June 2025)

A separate, more recent dataset covering July–October 2025 confirms Reddit and Wikipedia as top-five domains across ChatGPT, Google AI Mode, and Perplexity, but with one important wrinkle: ChatGPT's citation rate for Reddit and Wikipedia collapsed from roughly 55–60% of prompts to 10–20% within about a month in September 2025, while the same domains stayed comparatively stable on Google AI Mode and Perplexity. Semrush's own analyst is not fully certain why, though a change to Google's search results parameters around the same time is one candidate explanation. The point isn't the exact percentage. It's that any citation-share number you read is a snapshot, not a law. (Semrush, The Most-Cited Domains in AI, November 2025)

Why Third-Party Content Beats Brand Pages

The instinct is to assume AI models simply distrust brand pages because they're promotional. That's roughly right, but the evidence points to something more specific: it's not commercial content that's disadvantaged, it's first-party content about itself.

Commercial (.com) domains actually make up over 80% of ChatGPT's citations, with .org sites a distant second at around 11%. That sounds like it contradicts the "brands lose" story, until you look at which .com domains are doing the winning: G2, NerdWallet, TechRadar, PCMag, Forbes, Business Insider, Reuters. These are commercial sites, but they're independent commercial sites writing about other companies, not companies writing about themselves. (Profound, AI Platform Citation Patterns, June 2025)

That distinction matters for strategy. A brand's own comparison page ("Us vs. Competitor X") competes for citation against every independent review site covering the same comparison, and independent sites generally win because engines can cite them without appearing to amplify a company's self-promotion. The practical implication: pitching for inclusion in existing third-party roundups, review sites, and comparison content is often a faster path to citation than trying to out-rank those sites with your own version of the same page.

Why Comparison and Listicle Content Looks Favored

If third-party trust explains most of the pattern, there's still a second, format-related effect worth understanding: Google's AI Mode doesn't just answer the query you typed. It uses a technique called query fan-out — issuing several related sub-queries across subtopics simultaneously, then synthesizing the results together. A page never has to directly target a sub-query to get swept into an answer built from it. (Google, AI Overviews and AI Mode in Search)

This gives naturally multi-faceted content, like comparisons and listicles that cover several products, criteria, or angles in one page, more surface area to get pulled into multiple fan-out sub-queries than a narrow, single-topic brand page has. Neither Google nor the independent studies quantify exactly how much citation volume comes from fan-out versus the primary query, so treat this as a plausible mechanism rather than a measured effect. But it's a more precise explanation than "AI loves listicles" as an unqualified rule. What's actually happening is that comparison-shaped content matches how these engines search, not that the list format itself carries some special weight.

The Real Gate: Can You Rank in Ordinary Search At All?

Before any of the above matters, there's a more basic requirement for Google's AI surfaces specifically: about 70% of the URLs cited in Google AI Overviews come from the existing top-10 organic results for the query or its fan-out sub-queries. (Surfer SEO, AI Search Study: Sources in Google AI Overviews, July 2025)

Google confirms this by design, not by accident. Its own documentation states that AI Overviews are built to surface information "backed up by top web results" and that core web ranking systems are integrated directly into the feature. (Google, AI Overviews and AI Mode in Search)

The practical implication is one that a lot of "AI optimization" advice glosses over: for Google's AI surfaces, you can't skip traditional SEO and go straight to "AI-friendly" content. If a page can't rank in ordinary search, it's very unlikely to be cited in an AI Overview. The remaining 30% of citations that come from outside the top 10 shows ranking isn't the only factor, but it's close to a prerequisite, not a separate discipline. ChatGPT and Perplexity use different retrieval systems and aren't bound by Google's index the same way, but there's no evidence they reward pages that can't earn attention through ordinary discovery either.

Within a Page, Density Beats Length

Assuming a page clears the ranking bar and lives on a trusted-enough domain, one more mechanism shapes what actually gets cited: how much of the page the engine pulls into its answer.

One analysis of Google's AI Overviews found that each query draws from a roughly fixed grounding budget of about 2,000 words, split across all cited sources by relevance rank. The median individual source contributes only around 377 words, with about 77% of pages contributing somewhere between 200 and 600 words regardless of how long the source page actually is. Content beyond roughly 1,500–2,000 words shows diminishing returns: it dilutes what fraction of the page gets used without increasing the amount that gets selected. (dejan.ai, How big are Google's grounding chunks?)

This finding comes from one proprietary dataset without confounding-variable controls or statistical testing, a limitation the author acknowledged directly in public discussion of the piece. Treat it as a mechanistically plausible explanation, not a settled rule. But it lines up with everything else here: engines extract small, quotable chunks rather than rewarding comprehensive length. The practical takeaway is to put your clearest, most specific, most quotable facts early in a page, and to stop assuming that a longer, more exhaustive article automatically earns more citation. A tight 800-word comparison with a clean table can out-cite a rambling 4,000-word guide covering the same ground.

Don't Confuse Being Mentioned, Cited, and Crawled

A common measurement mistake is treating "AI visibility" as one signal. It's really at least three, and they mean different things:

• Mentioned — your brand name appears somewhere in the AI's generated answer text, with no source link attached.
• Cited — the answer includes a link back to your specific page as the source of a claim.
• Crawled — an AI system's bot has visited your page at all, which says nothing about whether that visit led to a mention or citation.

A brand can be mentioned constantly without ever being cited. A page can be crawled repeatedly by an AI bot and still never make it into an answer. Marketing claims about "AI visibility" frequently blur these together, which makes it hard to tell whether a strategy is actually working or just generating crawler traffic that never converts into a citation. If you're going to track this yourself, tools that distinguish AI bot activity from actual citation and mention data give you a much clearer read than a single visibility score.

Citation Shares Move Fast. Build for Resilience, Not a Snapshot

The Reddit/Wikipedia collapse on ChatGPT in September 2025 is the clearest warning here: a domain that accounted for 55–60% of citation share dropped to 10–20% within roughly a month, while the same domains barely moved on Google AI Mode and Perplexity over the same period. Even Semrush's own researcher, who documented the shift, wasn't confident about the exact cause, suggesting it may reflect an attempt by ChatGPT to reduce over-reliance on a small number of sources. (Semrush, The Most-Cited Domains in AI, November 2025)

The lesson isn't "target Wikipedia" or "target Reddit." It's that any specific citation-share number is a snapshot of one platform at one moment. A strategy built around chasing last quarter's winning domain is fragile. A strategy built around durable fundamentals, ranking well, earning third-party trust, and structuring content for extraction, holds up regardless of which platform is having a volatile month.

What This Means for What You Build Next

Ordered roughly by how foundational each step is:

1. Get your rankings in order first. If you're not appearing in the top 10 organic results for the queries that matter, Google's AI Overview citations are largely out of reach regardless of content quality elsewhere.
2. Pursue third-party inclusion as a primary lever, not an afterthought. Pitching for placement in existing review sites, comparison roundups, and relevant community discussions is often more efficient than trying to out-cite them with your own version of the same content.
3. Structure first-party content as dense, comparison-shaped answers. Put the clearest, most specific claim early, use tables or lists for comparable facts, and don't pad length past the point where it adds new information.
4. Track citations by platform, not as one aggregate score. Since ChatGPT, Perplexity, and Google's AI surfaces behave differently and shift independently, a single "AI visibility" number can hide which platform is actually working.
5. Recheck your own niche's citation patterns periodically. The cross-industry averages in this article will look different in specific verticals and will keep shifting; treat any number here as directional, not permanent.

How to Tell If It's Working

Watch three separate signals over several weeks, not days, given how volatile citation share can be: whether your pages are being crawled by AI bots at all, whether your brand is being mentioned in relevant AI answers, and whether any of those mentions come with an actual source citation back to a page you control or influence. Progress on the first two without movement on the third usually means content is being discovered but isn't extractable or trusted enough yet to cite. That's a structure and third-party-trust problem, not a discovery problem, and the fix is revisiting density and placement rather than producing more content.

Tideflow AI's MCP-based publishing and analytics tracking are built around this exact distinction: monitoring where a brand is mentioned, cited, and crawled across platforms, then turning the specific gaps into content and third-party inclusion targets rather than guessing from industry-wide averages. If you're evaluating platforms for this kind of monitoring, see how the approach compares in Tideflow AI vs. Profound.

Sources

• Google, AI Overviews and AI Mode in Search
• Sergei Rogulin, Semrush, The Most-Cited Domains in AI: A 3-Month Study (November 2025)
• Profound, AI Platform Citation Patterns: How ChatGPT, Google AI Overviews, and Perplexity Source Information (June 2025)
• Jakub Sadowski, Surfer SEO, AI Search Study: Sources in Google AI Overviews (July 2025)
• Dan Petrovic, dejan.ai, How big are Google's grounding chunks?