Finding 1 — Winner-take-most is the default
Across six industries, the three most-named brands captured a median 62% of all brand mentions in category prompts. Unlike a results page with ten slots, an answer names a handful — and synthesis prefers to repeat brands it has already resolved confidently. Early visibility compounds; late entrants fight a feedback loop.
Finding 2 — Engines crown different winners
The top brand on one engine was also top on a second in only about half of industry runs. The mechanism is structural: different indexes, different crawler policies, different trusted-source weightings. A "we're winning on ChatGPT" screenshot can be simultaneously true and irrelevant to your Google AI Overviews position. Per-engine strategy is not pedantry — it's the game.
Finding 3 — The citation layer is third-party
In B2B categories, the majority of cited URLs were directories, review platforms and trade media. Vendor sites were cited mainly for definitional and specification content. Practical translation: the shortest path into AI answers for most brands runs through other people's domains — claim, complete and correct them before commissioning more owned content.
Full tables, prompt-set design and limitations: Q4 2026 AI-visibility benchmarks.