Welcome to episode 27 of The GEO Show, the briefing on generative engine optimization, LLM citations, and AI search visibility. Brought to you by GEOforge.
In this episode, Paris Childress, founder of Hop AI and co-founder of GEOforge, runs ten stories. He covers AirOps native CMS and project-management integrations, OpenAI licensed financial datasets inside ChatGPT, the ChatGPT Work data agent across major warehouses, reported ChatGPT ad refusals for competing image and audio products, Similarweb ecommerce 2026 AI referral findings, August LLM web-share shifts with Claude at 9.3%, Anthropic crawl-to-refer ratio spreads of about 31x, Chattopic persona and journey simulation, Hexagon's 50K shopping-recommendation study, and Good Firms' directory of 1,178 GEO agencies.
In this episode:
🔌 AirOps: publish without opening your CMS
AirOps launched native integrations with seven CMSs (Webflow, WordPress, Contentful, Sanity, Contentstack, Ghost, Strapi) and four PM tools (Asana, ClickUp, Monday.com, Airtable). The pitch is insight to published content without manual handoffs. Like Profound, AirOps is pushing a marketing system-of-record play that abstracts CMS and project management.
📚 OpenAI answers move onto licensed data
ChatGPT for financial services now includes premium datasets from Daloopa, PitchBook, and LSEG News. OpenAI says it indexes and hosts partner data itself with granular citations. Commercially important evidence may live in licensed datasets GEO cannot influence through conventional public-web optimization.
💬 ChatGPT Work data agent: dashboards become conversation
OpenAI's data agent connects BigQuery, Databricks, Snowflake, Redshift, and ClickHouse. Users ask why a metric moved and get investigation in natural language. Standalone analytics UIs get less defensible as work shifts into the LLM via MCP connectors.
🚫 OpenAI ads vs competing gen products
PYMNTS, citing The Information (Sep 10), reported OpenAI told some partners it would no longer accept ChatGPT ads for image and audio generation products that compete with OpenAI. Adobe Firefly was reportedly affected. Separate paid AI visibility from organic GEO.
🛒 Similarweb ecommerce 2026: AI referrals +200% YoY
Direct referrals from AI platforms grew more than 200% year over year but remain small in volume. An AI recommendation gave up to a 2:1 purchase advantage in some cases. 89% of AI shoppers also use conventional search, often ending in branded bottom-funnel queries.
📈 Claude is no longer a rounding error
Similarweb August web traffic: ChatGPT 55.5% (was 73%), Gemini 25.6% (was 13%), Claude 9.3% (was 1.9%), DeepSeek 3.4%, Grok 2.4%, Copilot 1.6%, Perplexity 0.9%. Add Claude to AI visibility tracking.
📉 Crawl-to-refer ratios ~31x apart
Search Engine Journal documented Anthropic crawl-to-refer ratios via Cloudflare from 71,000:1 down to 2,200:1. Crawl volume rises while human referrals drop. Measurement windows and missing referrers explain some spread, not the whole trend.
🧭 Chattopic: persona-first visibility
Chattopic positions AI visibility as starting with the user, not a static prompt: buyer simulation, persona-level visibility, persona chat, and AI focus groups. Gumshoe-style workflow: build personas first, then align prompts, for more runs and better statistical relevance.
📊 Hexagon 50K shopping recommendations
Hexagon analyzed 50,000 shopping recommendations across ChatGPT, Perplexity, and Claude. Authoritative content, third-party validation, and structured data accessibility allegedly explain 73% of citations. DA60+ roundups and structured data correlated with recommendations, with debate versus earlier episodes.
🏢 Good Firms: 1,178 GEO agencies
Good Firms (Sep 10) lists 1,178 GEO agencies across 80 countries, median $37/hr, 9% with 50+ people. Directory classification, not a pure-play census, but the SEO-to-GEO rebrand wave is real and early-mover advantage is largely gone.
If this briefing is useful, subscribe to The GEO Show so we can keep walking through the platform changes and citation mechanics that actually move AI visibility.
💬 Question: If OpenAI answers increasingly cite licensed partner datasets, which evidence sources in your category can you still influence, and which sit outside conventional GEO?