Why answer rate and share of voice tell different stories.

Most GEO dashboards report a single brand-level number: the share of prompts in which a brand appears. That metric has a name — answer rate (AR) — and it is genuinely useful. But it is only one coordinate on a two-dimensional map. The second coordinate is share of voice (SOV): the fraction of total brand-token surface area that a given brand occupies across all answers in a category. AR tells you whether you are in the room. SOV tells you how much of the room you own. The two diverge in ways that matter strategically, and reporting one without the other produces systematically misleading conclusions about GEO performance.

The gap between the two metrics is not a quirk of methodology. It is baked into how large language models construct answers. When an HCP asks which biologics are appropriate for moderate-to-severe atopic dermatitis, a well-formed answer names several options, describes mechanism, cites trial data, and compares tolerability profiles. Every brand mentioned in that answer earns one AR count. But the brand that receives three sentences of trial detail, a safety summary, and a guideline endorsement earns far more token surface than the one that gets a single parenthetical mention. AR treats both identically. SOV does not.

The 4.2× compression: what the data shows

The May 2026 PharmaGEO public index captured answer rate and share of voice simultaneously across four therapeutic areas and three LLM engines. Two data points from that index illustrate the AR/SOV gap most cleanly.

In the atopic dermatitis category on OpenAI, Dupixent carried an answer rate of 65.5% and a share of voice of 16.3% — an AR/SOV ratio of approximately 4.0×. In the obesity category on the same engine, Wegovy's answer rate reached 93.2% against a share of voice of 22.2%, producing a ratio of 4.2×. A brand that is named in virtually every obesity answer — nine out of ten queries — still accounts for fewer than one in four brand tokens across those answers. The remaining 77.8% of brand conversation belongs to Zepbound, Qsymia, and the rest of the competitive field.

Source: May 2026 PharmaGEO public index, OpenAI engine. AR = percentage of prompts in which the brand is named at least once. SOV = brand's share of total brand-token surface area across all answers in the therapeutic area.

The structural ceiling: why SOV compresses differently by category

The AR/SOV compression ratio is not uniform across therapeutic areas. It is shaped by the competitive structure of each category — specifically, how many brands an LLM considers relevant when constructing a complete answer. The May 2026 PharmaGEO public index measured top-3 combined SOV as a proxy for category concentration, and the differences are structural, not incidental.

Source: May 2026 PharmaGEO public index, OpenAI engine. Top-3 brands by SOV in each TA.

Oncology and biologic-rich categories distribute SOV across 20 or more brands because a thorough answer in those categories requires naming many agents across distinct mechanisms, biomarker profiles, and lines of therapy. Metabolic and GLP-1 categories concentrate SOV into two or three brands because the competitive set is structurally narrow. This is the fragmentation rule: the SOV ceiling in lung cancer is not something a brand can market its way through. It is a category-level structural property of how LLMs interpret the treatment landscape.

The practical implication is that an oncology brand holding 8% SOV may be performing at or near the structural ceiling for its category, while the same 8% in obesity represents a competitive failure. Benchmarking SOV without reference to category structure produces incoherent targets.

What a high AR can obscure

Consider a brand with a 90% answer rate and an 8% share of voice. By most GEO reporting frameworks, that brand looks strong: it is present in nine of ten answers. But the SOV number reveals the actual situation. The brand is co-mentioned in the majority of answers in the category, but it is dwarfed by competitors that carry 20, 30, or 40% of the brand-token surface in those same answers. The brand is a supporting cast member, consistently present but consistently brief. Clinical evidence, mechanism detail, guideline context — the content that drives prescribing decisions — accrues to the brands with high SOV, not merely high AR.

An AR-only lens leads teams to optimise for presence rather than depth. They may celebrate a jump from 60% to 75% AR without noticing that their SOV has remained flat or declined, because the additional mentions are single-sentence acknowledgements rather than substantive treatment descriptions. High AR with flat SOV is a warning sign, not a success metric.

What a high SOV can obscure

The opposite distortion is equally common. Imagine two brands in the same category:

• Brand A: 12% SOV across 30% of answers
• Brand B: 8% SOV across 90% of answers

A pure SOV ranking puts Brand A ahead. But these are fundamentally different strategic situations. Brand A's high SOV is concentrated in a minority of prompts where it appears at all — likely the queries where the brand has a clear mechanism advantage or a specific guideline endorsement. Brand B is present across nearly the entire query space but contributes a smaller share of conversation in each answer. Brand A has depth without breadth. Brand B has breadth without depth.

Which position is more valuable depends on the query mix. If the high-SOV prompts for Brand A are the ones that HCPs actually ask in clinical practice, Brand A's concentration may be commercially decisive even at lower breadth. If those prompts are edge cases, Brand B's broader presence is worth more. Neither metric answers this question in isolation. Both are required.

The Ahrefs 75,000-brand study (December 2025) found a correlation of r = 0.664 between web brand mentions and AI Overviews citation — a meaningful relationship, but far from deterministic. Brand mention frequency does not translate linearly into citation surface area. A brand can accumulate mentions across a wide range of low-depth web properties and still carry minimal SOV in structured AI answers. Mention frequency and surface area are different constructs that require different measurement approaches.

A combined metric: Answer-Weighted SOV

AR and SOV are complementary, not competing. The most informative single GEO metric is one that captures both dimensions simultaneously. We propose Answer-Weighted SOV, defined as:

Answer-Weighted SOV = SOV × AR

Interpreted as the expected brand-token surface per random query in the category. If a brand has 22.2% SOV and 93.2% AR, its Answer-Weighted SOV is approximately 20.7% — meaning that for any randomly selected query in the obesity category, this brand is expected to account for roughly one in five brand tokens in the resulting answer. If a brand has 12% SOV but only 30% AR, its Answer-Weighted SOV is 3.6%, even though its per-answer depth looks competitive.

Source: May 2026 PharmaGEO public index (rows 1–2). Rows 3–4 are illustrative scenarios using the same metric definition.

Answer-Weighted SOV collapses a two-dimensional position into a single number without destroying the diagnostic signal. When broken down into its component AR and SOV values, it tells the brand team not just where they stand but which lever to pull: if AR is low relative to SOV, the content distribution problem is reach; if SOV is low relative to AR, the problem is depth — the brand is consistently mentioned but not substantively described.

Category structure determines which lever matters more

The relevance of each lever depends on the structural ceiling of the category. In a highly fragmented field like lung cancer, where the top three brands combine for only 20.9% of SOV, the structural ceiling on any single brand's Answer-Weighted SOV is approximately 10–12% at most. Chasing SOV beyond that ceiling requires content investment that the category's competitive dynamics will absorb without producing proportionate returns. The more tractable lever in fragmented oncology categories is AR: ensuring the brand is present across the full range of relevant query types, including biomarker-specific, line-of-therapy-specific, and combination-regimen queries where the brand has a documented clinical role.

In concentrated metabolic categories like obesity, where the top-2 brands already hold more than 40% of combined SOV, the AR ceiling is high — as the Wegovy data shows, a category-leading brand can reach 93% AR. The competitive opportunity is in SOV depth: the brand that converts its near-universal presence into richer, more detailed, more evidence-heavy token surface will widen its Answer-Weighted SOV advantage even without adding to AR.

The content architecture that drives SOV depth

SOV is not a function of brand mentions in the conventional sense. An Ahrefs analysis of 75,000 brands found that web brand mentions correlate with AI citation at r = 0.664 — meaningful, but not causal. A brand can accumulate a large mention footprint across low-authority properties and still carry minimal SOV if those properties are not the sources LLMs retrieve when constructing authoritative answers.

SOV depth is driven by the content that ends up in the answers — and in pharma, that content comes overwhelmingly from a narrow set of source archetypes. In lung cancer, NCCN guidelines dominate citation to such a degree that NCCN accounts for the majority of Perplexity's source uses in that TA. In obesity, FDA label text and manufacturer prescribing information are the top-cited sources. In atopic dermatitis, society guidelines, peer-reviewed literature, and specialist hubs share the citation pool.

A brand with robust representation in the source types that an engine actually retrieves for its category will earn more token surface per answer than a brand with equivalent mention volume spread across lower-authority properties. The path to higher SOV runs through the citation architecture of the specific engine and category, not through generic content volume.

Tracking AR/SOV over time

The AR/SOV ratio — how many answer appearances it takes to generate a unit of share of voice — is itself a diagnostic metric worth tracking over time. A ratio that is increasing means the brand is gaining presence but losing depth: it appears in more answers but with thinner representation in each. A decreasing ratio means depth is improving relative to breadth, which is generally a positive signal for brands that are already well distributed across the query space.

For a brand launching into an established category, the expected trajectory is high AR/SOV ratio early — the brand gets mentioned in categorical answers as a new entrant without the depth of clinical detail that accrues to established agents — followed by a declining ratio as trial publications, guideline inclusions, and authoritative disease-state content build up in the indexed ecosystem. Tracking this trajectory against a competitive benchmark makes visible whether GEO investments are compressing the ratio faster or slower than competitors.

Operationalising both metrics together

Running AR and SOV simultaneously is straightforward in a structured audit. The prompt set is the same; what differs is the measurement layer applied to responses. AR requires only brand presence detection — a binary flag per answer. SOV requires tokenisation or word-count approximation of brand-specific content across all answers in the set, normalised to the total brand-token surface. Both can be run programmatically on LLM API outputs, making them suitable for regular monthly pulse audits rather than quarterly-only deep reviews.

The reporting format that makes both metrics legible to a brand team is a 2×2 positioning chart: AR on the x-axis, SOV on the y-axis, plotted for each brand in the competitive set. The chart immediately identifies the four strategic archetypes: high AR / high SOV (category leader); high AR / low SOV (co-mentioned but dwarfed); low AR / high SOV (niche authority); low AR / low SOV (structurally absent). Each quadrant has a different content intervention and a different expected payback timeline.

Answer rate alone maps a brand's presence. Share of voice maps its weight. Neither is a proxy for the other, and the 4.2× compression visible in the May 2026 PharmaGEO public index data is a conservative estimate of how far apart those two readings can diverge. Brands that report only one will systematically misread their competitive position in the AI answer layer.

Want a real audit on your brand? Request a sample report or get the full PharmaGEO Playbook.