Welcome to episode 40 of The GEO Show, the briefing on Generative Engine Optimization and AI visibility. Brought to you by GEOforge.
In this episode, Paris Childress, founder of Hop AI and co-founder of GEOforge, covers seven stories: agentic web search that varies by model, Server.ai's expansion into referrals and crawler logs, Google AI Overviews vs AI Mode as separate citation ecosystems, first-party logs that undercut raw crawler counts, Share of Model as a competing KPI, an open-source Python package for citation measurement, and a Catalyst recommendation benchmark where Peak outpaces Profound and AirOps.
In this episode:
- 🔬 Agentic search varies radically by model
- In a controlled 1,000-prompt experiment, Claude Sonnet 4.6 invoked search 825 times versus 140 for ChatGPT 5.3, roughly 6X. More search did not reliably mean better answers, and used evidence did not always match citations. Measure the retrieval life cycle, not only the final answer.
- 📡 Server.ai expands into referrals, crawler logs, and API
Updated September 20: Server.ai documents AI referral tracking across 13 platforms, crawler log import via user-agent patterns, citation and competitor monitoring, plus a REST API for visibility data. APIs are becoming commodity even among smaller vendors. Verified crawler identity and confidence intervals matter more.
- 🔀 Google AI Overviews and AI Mode barely share citations
SE Ranking found 10.7% URL overlap and 16% domain overlap between AI Mode and AI Overviews. Ahrefs measured 13.7% citation overlap on a larger paired set. Even inside Google, AI visibility is not one surface. SignalForge inside GEOforge already separates them.
- 📉 Crawler counts are a weak proxy for AI discovery
E-commerce Fast Lane analyzed 14 days of logs: 49,559 crawler visits, roughly 100 from OpenAI, Anthropic, and Perplexity bots. ChatGPT referred 613 human sessions while crawling the site only 55 times. Correlate bots to indexation, citations, answers, and referrals before assigning meaning.
📊 Share of Model challenges blended Share of Voice
Monroyia's dataset covers 20,996 buyer questions, 143,298 cited sources, 41,397 domains, and four models. Share of Model is the percentage of sampled answers naming a brand, broken down by model and buyer stage. Vendor-owned pages claimed well under 1% of sources (vendor research; needs replication).
🐍 AI citation measurement ships as open-source Python
PyPI released version 0.29.0 of an early-release open-source package for measuring and improving citations across ChatGPT, Google AI Overviews / AI Mode, Perplexity, and Claude. Basic AI search measurement is moving from SaaS-only features toward developer building blocks. We still call the category GEO.
🏆 Peak beats Profound and AirOps on one recommendation board
Catalyst's leaderboard (618 answers, 12 buyer questions, 5 platforms) showed 7-day averages of Peak 58.1%, Profound 15.9%, AirOps 9%. One-day noise is roughly 15 percentage points. Commercial traction and AI recommendation rates are not the same thing.
Subscribe to The GEO Show wherever you get your podcasts, and watch the full episode on YouTube: https://www.youtube.com/watch?v=pp6AzglZgFs
💬 Question: Are you still blending Google AI Overviews and AI Mode into one AI visibility score, or measuring them as separate citation ecosystems?