One ChatGPT Model Update Moved AI Visibility 56 Percent

What happened

Profound, an AI search analytics company, launched the Profound Index at its Zero Click New York conference on June 29, 2026, according to a GlobeNewswire release. The index tracks 1.5 billion real-user prompts across more than 50 industries and updates daily, giving marketers a standing benchmark for how often AI platforms recommend their brand. Its first data release showed how unstable that visibility is: Travelocity, Orbitz, and Priceline dropped 35 to 56 percent on ChatGPT after a single model update, while Ambetter and Humana fell more than 50 percent in health insurance visibility. IKEA moved in opposite directions on the same day, up 19 percent on ChatGPT but down 11 percent across AI platforms overall. Days earlier, at Google I/O, Google announced "information agents" inside AI Mode, background processes that monitor a topic around the clock and push updates directly to a user's phone rather than waiting for a search query.

Why AI visibility matters now

A Google organic ranking used to be a reasonable proxy for whether a buyer could find you. Profound's numbers show that proxy has broken: a single model update can cut a brand's AI recommendation share in half without touching its Google rank at all. Add Google's information agents, which will surface recommendations to users who never open a results page, and the entire discovery layer for B2B buyers is moving to a surface most founders have never measured, let alone optimized.

1. 01
   Run the same four prompts across ChatGPT, Gemini, Claude, and Perplexity monthly

   Ask each model a version of the question a prospect would ask about your category, screenshot the answer and cited sources, and track whether you appear. Do it on a fixed monthly schedule so you can catch a swing like the one Profound documented.
2. 02
   Treat AI answer share as a separate metric from search rank

   Your Google position and your AI recommendation share now move independently. Report them separately to your team so a drop in one doesn't get masked by stability in the other.
3. 03
   Publish the structured, dated content models cite

   Model answers lean on pages with clear question-shaped headers, current dates, and direct answers in the first two sentences. Audit your highest-intent pages against that format before assuming a visibility drop is about competitors, not formatting.

By the numbers: The Profound Index is built on 1.5 billion real-user prompts across more than 50 industries, and its first release found ChatGPT 5.5 now recommends AI competitors Anthropic, Mistral, OpenRouter, and Groq more often inside its own answers, a sign that model updates are reshuffling entire categories, not just individual brands.

"Marketing teams spent years asking where they ranked on Google. Increasingly, that is becoming the wrong question. People are asking ChatGPT, Gemini, Claude, and other AI systems for recommendations, advice, and answers. The brands that matter are the brands AI recommends."

— James Cadwallader, co-founder and CEO, Profound

What to do this week

Pick your single highest-intent buyer question, the one a prospect asks right before they shortlist vendors, and run it through ChatGPT, Gemini, Claude, and Perplexity today. Screenshot the results and cited sources. That's your baseline. Repeat it on the same day each month so a 35-plus percent swing shows up as data, not a surprise.