Share-of-Answer Measurement

Pharmaceutical brands have traditionally measured visibility through share-of-voice: advertising impressions, search rankings, social mentions, and media coverage. But a fundamental shift has occurred: patients and HCPs now ask AI first.

Over 40 million people ask ChatGPT health-related questions every day^{[2, 3]}. Three in five U.S. adults have used AI tools for healthcare queries in the past three months. This isn't a future trend - it's current behavior.

The problem: traditional measurement doesn't capture this layer. SEO tracks search rankings, but AI answers aren't search results. Social listening tracks mentions, but AI conversations are private. Advertising measurement tracks impressions, but AI doesn't show ads.

Share-of-answer is the new metric: how often your brand appears in AI responses compared to competitors, across the questions that patients and HCPs actually ask.

Share-of-answer breaks down into four key dimensions:

• Mention Rate: The percentage of relevant AI responses where your brand appears. If your drug treats condition X and you run 100 questions about condition X treatments, mention rate tells you what percentage included your brand.
• Ranking Position: When AI lists multiple options, where does your brand appear? First, second, or buried at the end? Ranking affects perception of leadership and preference.
• Citation Frequency: How often do AI systems cite your sources (website, clinical trials, prescribing information) vs competitor sources? Citations indicate which content AI considers authoritative.
• Drift Over Time: Share-of-answer isn't static. AI models update, sources change, and competitors publish new content. We measure week-over-week changes to detect trends early.

Each dimension is tracked by theme (efficacy questions, safety questions, access questions), by provider (ChatGPT, Claude, Gemini, Perplexity), and over time.

Measurement follows a structured process designed for consistency and reliability.

We design query sets based on real-world question patterns, not arbitrary prompts. Query design considers:

• Patient intent patterns: Treatment options, side effects, cost/access, lifestyle impact, comparison questions ("Which is better, X or Y?")
• HCP intent patterns: Dosing, drug interactions, clinical evidence, guidelines, mechanism of action, formulary status
• Journey stage: Awareness ("What causes X?"), consideration ("What are treatments for X?"), decision ("Should I take Y?"), ongoing management ("How do I manage side effects of Y?")
• Question phrasing variants: The same intent can be asked many ways. We include natural language variations to capture how real users phrase questions.

A typical baseline query set includes 50-200 question variations, covering your therapeutic area and key competitors. Query sets are customized per brand and indication.

Each LLM provider has different training data, source preferences, and answer patterns. We run query sets across:

• ChatGPT (OpenAI) - Largest consumer base, 40M+ daily health queries^[2]
• Claude (Anthropic) - Growing enterprise and professional adoption
• Gemini (Google) - Integrated with Google Search and Android
• Perplexity - AI-native search with citation-heavy answers

Provider-level tracking reveals where you're strong (e.g., good visibility in Claude) and where you're weak (e.g., missing from Gemini). This informs content strategy: different providers may weight different source types.

We control for model versions, temperature settings, and other parameters to ensure consistent, reproducible measurements.

Share-of-answer measurement produces the following outputs:

• Visibility Score: A normalized 0-100 score reflecting overall share-of-answer across all providers and question types. Benchmark against competitors to understand relative position.
• Mention Rate by Theme: Breakdown of visibility by question type (efficacy, safety, access, etc.). Identifies specific gaps - you might be strong on efficacy but weak on access questions.
• Ranking Distribution: Histogram of where your brand ranks when mentioned. High ranking (1st, 2nd) indicates perceived leadership; low ranking indicates also-ran positioning.
• Citation Map: Visual representation of which sources AI cites for your brand vs competitors. Reveals whether AI trusts your evidence or relies on third-party content.
• Drift Snapshots: Week-over-week comparison showing changes in mention rate, ranking, and citation patterns. Flags emerging trends or sudden changes.

All outputs include raw data and evidence links so findings can be verified and investigated further.

Share-of-answer data informs action across functional teams:

• Brand Marketing: Identify competitive gaps and prioritize content investments. If competitor X is winning on "treatment comparison" queries, create content that addresses that intent directly.
• Omnichannel Marketing: Ensure AI visibility aligns with broader channel strategy^[7]. If patients hit your website after asking AI, the messaging should be consistent.
• Medical Affairs: Monitor whether AI is accurately representing clinical evidence. Low citation of peer-reviewed sources may indicate a need for more accessible medical education content.
• Communications: Track narrative trends and prepare for questions about what AI says about your brand. Early warning of drift enables proactive messaging.

Share-of-answer measurement requires careful methodology. Common pitfalls include:

• Single prompt bias: Relying on one or two questions produces unreliable data. AI responses vary by phrasing, context, and even time of day. Robust measurement requires diverse query sets with sufficient volume.
• Personalization effects: AI systems may personalize responses based on user context, location, or prior conversations. We use fresh sessions with controlled parameters to minimize personalization noise.
• Time variance: AI answers change as models update. A single snapshot tells you today's state but not trends. Continuous measurement with weekly retests reveals patterns.
• Provider differences: Aggregating across providers obscures important variation. Provider-level views are essential because each AI has different source preferences and training data.
• Confusing visibility with accuracy: High mention rate doesn't mean accurate mentions. Share-of-answer measures visibility; PI-backed claim defensibility measures accuracy. Both are needed.

Share-of-answer measurement is not a rebranding of existing approaches. The fundamental differences:

AI answers are synthesized, not linked. When a patient asks ChatGPT about treatment options, they get a paragraph - not a list of pages to click. That paragraph shapes perception in ways that search results never did.

Share-of-answer measures this new layer. It's complementary to SEO and social listening, not a replacement - but it's also not optional. If you're not measuring share-of-answer, you have a blind spot where 40+ million daily health questions are being answered^{[2, 3]}.