Investigation Report // 08
BOFU Belongs to Brands: Empirical Proof That Bottom-of-Funnel AI Search Retains Entity Attribution
We tested whether high-intent "Bottom of Funnel" (BOFU) queries trigger explicit brand and provider entity mentions in AI search overviews compared to Top-of-Funnel informational queries.
Hypothesis Conclusion
HYPOTHESIS VALIDATED / CONFIRMED BY DATA
Empirical data validates the hypothesis that Bottom-of-Funnel (BOFU) queries generate significantly more brand and provider entity mentions than Top-of-Funnel (TOFU) or Middle-of-Funnel (MOFU) queries. While TOFU queries suffer 60% brand erasure, BOFU queries trigger explicit brand mentions in 65.5% of Google AI Overviews and 90.0% of Gemini responses.
Key Takeaways
- 90.0% Entity Attribution in Gemini: When users issue BOFU queries seeking specific solutions, providers, or costs, Gemini included named health systems, manufacturers, or surgeons in 27 out of 30 cases.
- 65.5% Brand Retention in Google AIO: Google AI Overviews generated explicit brand citations in nearly two-thirds of BOFU search cases, confirming that intent drives entity surfacing.
- The Funnel Inversion: As user intent transitions from Educational (TOFU) to Transactional/Actionable (BOFU), AI models switch from brand-blind synthesis to active entity recommendation.
- IP-Based Local Entity Resolution: For non-localized "near me" BOFU queries, Google AI Overviews dynamically resolve the user's IP geolocation to recommend nearby accredited specialists.
The Data Behind this Study
We analyzed 30 live AI search queries across Google AI Overviews and Gemini, specifically targeting high-intent Bottom-of-Funnel (BOFU) keywords across orthopedics, cardiology, and surgical devices (e.g., Stryker Mako, Watchman device, TAVR, ACL revision).
Definitions
- Bottom-of-Funnel (BOFU): High-intent queries where the user is seeking a specific provider, facility, pricing structure, or surgeon recommendation.
- Entity Attribution: The explicit inclusion of branded institutions, specialized medical devices, or named physicians within the AI response.
- IP Geolocation Resolution: The mechanism by which an AI search engine uses the user's network origin to filter local provider entities for un-geolocated "near me" prompts.
Experimental Setup
The study evaluated 30 randomized BOFU surgical and clinical queries across two primary model engines:
- Query Set: 30 high-intent transactional prompts (e.g., "Best hospital in Dallas for heart valve repair," "Stryker Mako cost with Medicare," "ACL reconstruction surgeons NYC").
- Models Tested: Google AI Overviews (via SerpAPI) and Gemini 3 Flash.
- Validation Criteria: Marked as "Validated" if the AI response contained explicit brand, manufacturer, hospital system, or physician entity mentions.
Findings
The table below shows the validation metrics proving that BOFU intent protects brand visibility in AI search.
| Model | Response / Trigger Rate | Hypothesis Validation Rate (Brand Mentions Present) | Primary Entity Types Surfaced |
|---|---|---|---|
| Google AI Overviews | 96.7% (29/30) | 65.5% (19/29) | Hospital Systems, Local Clinics, Device Brands |
| Gemini 3 Flash | 100.0% (30/30) | 90.0% (27/30) | Academic Medical Centers, Named Specialists, Manufacturers |
Raw Evidence & Case Studies
Case Study 1: High-Intent Regional Provider Resolution
Query: "Best hospital in Dallas for minimally invasive heart valve repair"
AI Response Snippet: "Leading hospital systems in Dallas for minimally invasive heart valve repair and replacement include UT Southwestern Medical Center, Baylor Scott & White – The Heart Hospital (Plano & Dallas), and Texas Health Presbyterian."
Analysis: Hypothesis Validated. For BOFU queries requesting regional recommendations, AI search engines actively resolve local healthcare entities, explicitly naming top academic medical centers and health systems in the metro area.
Case Study 2: Commercial Device & Insurance Attribution
Query: "Stryker Mako knee replacement surgery cost out of pocket with Medicare"
AI Response Snippet: "Determining your out-of-pocket cost for a Stryker Mako robotic knee replacement under Medicare involves understanding how Original Medicare (Parts A & B) applies to outpatient surgery centers vs hospital inpatient stays..."
Analysis: Hypothesis Validated. When product brands (Stryker Mako) and financial terms (Medicare) are combined, AI models preserve full brand attribution and explain specific billing parameters without anonymization.
Case Study 3: IP Geolocation for Un-geolocated "Near Me" Searches
Query: "Orthopedic surgeons near me specializing in partial knee replacement"
AI Response Snippet: "Orthopedic surgeons near Mahopac, New York, who specialize in partial knee procedures include local specialists in Carmel and regional experts in New York City. Local Orthopedic Specialists (Carmel, NY) include Dr. Scott Levin, MD and Dr. Daniel Smith, MD..."
Analysis: Hypothesis Validated. Rather than anonymizing the answer, Google AI Overview resolved the network's IP geolocation (Mahopac/Carmel, NY) and cited specific local orthopedic surgeons by name.
Deep Research & Industry Context
The contrast between TOFU and BOFU AI search behavior highlights a fundamental principle of Retrieval-Augmented Generation (RAG): Intent dictates entity filtering.
According to research from Ahrefs and BrightEdge, while informational broad queries (TOFU) trigger brand-blind RAG synthesis, transactional and commercial queries (BOFU) force LLMs to pull from structured entity knowledge graphs. When a user asks "who" or "where," the model cannot synthesize an abstract answer—it must reference verified real-world entities.
Furthermore, studies from Pew Research indicate that users searching with BOFU intent actively expect entity recommendations. Search engines like Google AI Overviews leverage local NPI registries, health system directories, and business profile graphs to populate these high-intent answer blocks.
What this means for your brand
While Top-of-Funnel content is losing organic traffic to zero-click AI summaries, Bottom-of-Funnel optimization remains high-ROI:
- Capture Actionable Intent: Focus content strategies on BOFU queries (e.g., provider comparisons, device specs, regional availability, pricing breakdown) where AI models actively feature brand entities.
- Solidify Knowledge Graph Data: Ensure NPI registries, Google Business Profiles, and institutional directory data are 100% consistent across all medical databases to maximize AI entity resolution.
- Win the Co-Citation Battle: As seen in the study, AI search engines frequently group top regional providers together (e.g., UT Southwestern and Baylor Scott & White). Ensuring your brand is co-cited alongside category leaders establishes entity authority in LLM training corpora.
Primary Sources & Citations
- Ahrefs. "AI Overviews Tracker & Brand Citation Correlations." Source
- BrightEdge. "Weekly AI Search Insights: Branded vs. Non-Branded Prompts." Source
- Pew Research Center. "Google Users and AI Search Summaries." Source
- Think with Google. "Decoding Decisions: The Messy Middle of Purchase Behavior." Source