A shot at the funnel: What millions of GLP‑1 convos reveal about AI search

We analyzed millions of AI conversations about GLP‑1s at the start of 2026. The headline: every stage of the funnel now plays out inside AI chat interfaces—often in non‑linear, surprising ways. If you market in the GLP‑1 space (or any fast‑moving category), this creates new opportunities to measure what matters, spot gaps, and influence decisions earlier.

Below you’ll find what we learned about GLP‑1 behavior in AI, which signals to track, and how to act on them using Scrunch.

The short version: AI isn’t just replacing search—it’s reconceptualizing the journey. It’s a companion from first question to ongoing care.

AI funnel archetypes for GLP‑1s

We clustered conversations into five audience archetypes based on jobs‑to‑be‑done. Four of the five are similarly sized; “Access Seeker” is smaller but highly commercial.

What this means for marketers

The AI search metrics that matter (and how to use them)

Traditional SEO measured rankings and traffic. AI search (often called AEO) measures visibility and influence. Four signals now tell you if you’re winning:

Analytics leaders: expect your funnel metrics to look different. Insights we shared with Lenovo at eTail London 2025 hold up here: bounce rates can rise on AI‑referred sessions (visitors arrive with high intent and specific expectations), and conversion rates can spike as users drop into your site mid‑journey.

How Scrunch helps you monitor and improve AI visibility

Credibility in the community

Scrunch shared the main stage with Lenovo’s Global SEO & Operations Lead at eTail London 2025, unpacking how LLMs are reshaping discovery and measurement. Key takeaways from that session—AI redefining the funnel, “bounce rates are back,” and SEO teams evolving into behavioral research hubs—are embedded in the platform’s approach. Our team regularly publishes practical guidance and data‑backed takes on AI search trends and tactics across the Scrunch blog.

A GLP‑1 playbook you can run this week

Get started

Methodology (summary)

We scored conversations with a behavior‑first taxonomy (user state, task, assistance mode, personal context, decision proximity) using structured LLM outputs, then clustered one‑hot encoded labels (k‑means, k=5) validated by silhouette and ARI on held‑out splits. Demographics were joined and weighted to AI‑active populations for projection. Access Seeker estimates are directional due to a smaller base.

Further reading