Third-party sources win evaluative queries

Your own comparison page matters less than you think for the queries that actually decide a purchase. For evaluative queries (“best X”, “X vs Y”, “top tools for Z”), engines are about 6.5× more likely to cite a third-party source than your own domain, and roughly 85% of brand mentions in those answers come from third-party pages.

The finding

From converging crawl data across ChatGPT, Perplexity, and Gemini on evaluative queries:

• Engines cite third-party sources ~6.5× more often than brand-owned domains.
• About 85% of brand mentions in those answers come from pages you don’t control: G2, listicles, editorial reviews, community threads.

Confidence here is medium: it’s vendor crawl data, not a controlled study, and it varies by engine. But the shape is consistent and it matches our own probe results: on the highest-intent questions, the assistant is quoting the rest of the web about you, not you.

Why it works

For a “best tool for X” question, a model has a credibility problem with your own page: of course you say you’re the best. So it leans on sources that read as independent. Your comparison page is necessary but not sufficient: it gives the model a clean answer to ground on, but on its own it rarely clears the bar for an evaluative citation.

This is also the honest case for comparison pages, just reframed: build them so third-party writers and the models have something accurate to quote, then go earn the third-party presence that does the citing.

How to apply it

• Get on the lists. G2, Capterra, category roundups, “top N” listicles: presence there is what gets cited on evaluative queries. One accurate roundup placement can outweigh a page of your own copy.
• Still build your /vs and comparison pages: they’re the necessary half, and they’re the source third parties (and models) crib from. Make them honest enough to quote.
• Seed accurate third-party mentions rather than more owned content. The leverage is off your domain.
• Target evaluative queries specifically: they’re the highest-intent class, and the one where third-party citation matters most.

Then re-scan and watch the evaluative-query citation rate, with its interval. If it moved past the ±CI after a listicle placement, that’s signal; inside the interval, it’s noise.

---

From Collimer’s research library. We label medium-confidence findings as such; see the methodology. Want your own evaluative-query reading? Run a free scan.