Why third-party sources outrank your own marketing pages

An assistant asked to recommend a tool rarely quotes the vendor. It quotes the people who wrote about the vendor. That asymmetry is structural, not accidental — and once you see how concentrated it is, most content strategies start looking misdirected.

How concentrated

An index synthesising six independent studies across more than 680 million citations between August 2024 and April 2026 found that a handful of domains dominate. Reddit alone accounts for roughly 40% of aggregate citation frequency across engines. Wikipedia is second, appearing in 26–48% of ChatGPT’s top-ten answers. YouTube is third.

Thirteen sources together capture around 68% of every citation produced by ChatGPT, Claude, Gemini, Perplexity and Google AI Overviews.

Your own domain is not on that list, and it is not going to be.

Why the asymmetry exists

A model combines what it absorbed in training with what it retrieves at answer time. Both favour text other people wrote about you over text you wrote about yourself, for a reason that has nothing to do with algorithms: a description written by someone with nothing to sell is more informative than one written by someone with everything to sell. Reviews, forum answers, analyst notes and user-written documentation all carry a signal your landing page structurally cannot.

The practitioner literature has converged on the same point from a different direction. Wikipedia’s entry on generative engine optimization describes the discipline as managing an online presence rather than a single site, and Stackmatix’s comparison of SEO, AEO and GEO treats third-party surfaces as a distinct optimisation target from owned pages.

What this does not mean

It does not mean go and post about yourself on Reddit. That is the obvious inference and it is wrong twice over: communities detect and punish it, and a thread that reads as marketing carries none of the disinterested signal that made the platform citable in the first place. The value of those sources is precisely that they are not you.

It also does not mean your own pages are worthless. They are the fact source that everyone else copies from. When a reviewer writes about your pricing, they read your pricing page. If that page is wrong, ambiguous, or unreadable to a machine, the error propagates into the sources that do get cited — and it persists long after you fix the original.

What follows

The work splits in two, and most teams only do the first half.

Make your own pages copyable. Facts stated plainly, in text, with numbers and dates that a person writing about you can lift without ambiguity. Not prose worth admiring — prose worth quoting.

Make it easy for other people to describe you correctly. Documentation good enough that users answer each other’s questions with it. Data worth citing. Positions specific enough that an analyst can attribute them to you. This is slower, harder to measure, and it is where the citations actually come from.

The uncomfortable version: most of your visibility in AI answers is decided on pages you do not control, by people you cannot brief, using facts you published months earlier. The only leverage you have is the accuracy of what they picked up.