If you've watched which pages AI engines actually cite, one format dominates: the listicle. The "best CRMs for startups," "top 10 expense cards," "X vs Y" roundups. And it's not close.
The numbers
Multiple 2026 analyses. Including Search Engine Land's AI-citations study and AirOps' 2026 State of AI Search. Converge on the same pattern (figures vary by study and shift over time):
- Third-party sources are cited roughly 6.5× more often than brand-owned pages.
- Listicles are the single most-cited format: around 22-25% of all citations, and roughly 40% of commercial-intent citations (the "best X for Y" questions where a buyer is deciding). Brand product pages sit far lower, around 14%.
- Nearly 90% of third-party brand mentions originate from listicles, comparison pages, and review roundups.
The mechanism, not just the stat
When a user asks "what's the best expense card for startups," the model wants a neutral-looking, multi-option answer. A page that already lists several brands against criteria matches that shape perfectly. It reads as a survey of the market.
A page that lists only your products reads as a sales page, which the model is trained to treat as biased and therefore weaker as a citation. The roundup also has higher entity density (multiple named brands co-occurring with the category) and usually more verifiable specifics. Both things retrieval favours.
Where each type wins
This doesn't mean owned content is pointless. It means it has a different job:
- Listicles win discovery & commercial intent. "best / top / vs / alternatives" queries. If you're not in those third-party lists, you're largely invisible for the highest-intent buyer questions. No matter how good your own blog is.
- Your own pages win brand & feature queries. "how does X work," "X pricing," "does X integrate with Y." Here the model wants the canonical first-party source, and that's you.
The trap is writing more self-listing posts to compete with roundups. You'll lose. A brand blog that lists only its own products can't out-neutral an independent comparison.
Has this changed on ChatGPT? (the GPT-5.6 shift)
The baseline above is cross-engine and about which format wins a "best X" answer — it still holds. But on ChatGPT specifically, the listicle premium narrowed after the GPT-5.6 model became the default in mid-to-late 2026. A Peec AI before-and-after analysis (same ~1M prompts, week-before vs week-after the rollout, 2026) found the share of listicles cited by ChatGPT fell about 50.5%, and comparison pages about 32.1%. The cause was not a demotion of roundups but a change in how the model searches: it issued fewer "top"/"vs" fan-out queries, generated about 154% more searches per chat, cited roughly twice as many sources per chat (~25.55 vs ~12.48), and increasingly used the site: operator to pull facts straight from first-party pages. So listicles became a smaller slice of a larger, more first-party citation set — ChatGPT is citing fewer listicles, in detail here.
This does not overturn the strategy — it sharpens it. It is single-vendor, directional, and ChatGPT-specific (Google's AI Overviews and Perplexity still lean hard on third-party roundups). The read: keep earning listicle placement for discovery and other engines, and make your own pages a clean, citable first-party source so ChatGPT can lift them when it goes straight to your domain.
The strategy that follows
- Get placed into the independent roundups. Earned media. Being included in "best [category] tools" lists, review sites, and comparison content. Is higher leverage than trying to out-rank them with your own page. This is the single biggest move for commercial-intent visibility.
- Reserve owned content for what owned content wins: brand-specific and feature-specific questions, plus genuinely educational explainers (like this one) that build entity strength.
- Mind the platform split. Perplexity and Google's AI Overviews lean hardest on third-party roundups; ChatGPT leaned that way too but has been shifting toward first-party sources since GPT-5.6 (above); results also vary by region (US answers cite Reddit and listicles heavily). The same question can favour the roundup on one engine and your page on another, which is why you measure across engines.
The net for most brands: for "which is best" buyer questions, the multi-brand listicle wins decisively, so the work isn't more self-promotion, it's getting into the comparison content that already gets cited, and then tracking whether it's working across every engine. That tracking is what Buffy Intel is for.