Citations Are a Bad GEO KPI: ChatGPT-5.6 Luna Uses site: Operator in 76% of Fan-out Queries
gaganghotra_ · x · 2026-09-26
Seer Interactive compared GPT-5.6 Luna (now ChatGPT's default) against GPT-5.5 across 346 matched prompts in six industries. Wil Reynolds argues citations alone are a bad KPI for Generative Engine Optimization.
Key findings:
- Total fan-out queries barely changed, but Luna searches far more precisely: it used the site: operator in 76% of fan-out queries, up from 22% with GPT-5.5 — a 3x+ jump in domain-targeted searching
- Consulted sources and inline citations dropped, and open searches shrank, raising the bar for sites not already in Luna's direct search range
- Many domains are retrieved without being cited — retrieval itself matters (a chance to influence the answer) but current AI visibility trackers don't report it
Recommendation: use a fan-out Chrome plugin to inspect the retrieval process and measure full visibility rather than citation trends.
More from Venture
- ChatGPT co-creator's startup launches Jev, a non-LLM model reportedly eyeing $10B valuation — aakashgupta · 2026-09-26
- ChatGPT RLHF Co-Author's Startup Jev in Funding Talks at $10B Valuation — aakashgupta · 2026-09-26
- Warp Raises $85M to Launch Warp 2.0, Pitched as the First AI Head of HR — CurieuxExplorer · 2026-09-26
- Investor: boutique data training may beat strip miners amid model collapse risk — StewartalsopIII · 2026-09-26
- Commentator goes 'full doomer': GPUs depreciate in 2-3 years, AI bubble may end in depression — petetrainor · 2026-09-26
- Legal AI adoption accelerates: Legora's law firm customers are growing and hiring, not shrinking — chetanp · 2026-09-26