ChatGPT Still Picks Brands It Already ‘Knows’ — Recommendation Equity Is the Hard Problem

GPO’s September 2026 search-and-AI brief summarises a blunt finding: for category-level “what’s the best brand for X” questions, ChatGPT leans heavily on associations formed in training. Brands without that web of prior mentions had roughly a 2% chance of being recommended in the cited testing.

Citation ≠ recommendation

You can earn a citation with crawlable, well-structured pages. Earning a recommendation is closer to brand-building across the corpora models trained on — reviews, press, comparison sites, forums, documentation.

TollBit-style analyses of AI crawler traffic also remind publishers that a non-trivial share of bot hits now come from AI systems assembling answers, not humans clicking blue links.

What to do

• Invest in third-party presence, not only owned blog posts.
• Separate dashboards for “were we cited?” vs “were we recommended?”
• Accept that AEO without PR and community proof is half a strategy.

The soup is thick here: visibility inside answers is becoming as competitive as visibility inside ten blue links — with slower feedback loops.