Build a 'Share-of-Answer' Audit: Score Your Drug's AI Visibility Against Competitors Using Prompt-Only Research

The task

Brand marketers and med-affairs leads at pharma companies need to know whether their product actually shows up when HCPs (healthcare professionals) ask AI assistants clinical questions — and how often competitors get named instead. This audit is the first thing to do before you commit budget to "AI SEO" vendors or content refreshes, because when 81% of U.S. physicians now report using AI tools in clinical practice, marketers must think beyond being seen to also being cited. Run it quarterly per brand, or before any campaign kickoff.

Before AI

Today this gets done by hiring an agency to run "AI visibility" scans across a handful of chatbots, or by an analyst manually typing 30-50 queries into ChatGPT, Perplexity, and Claude, screenshotting each answer, and tallying brand mentions in a spreadsheet. Realistic timeline: two to three days of analyst work plus a week of vendor turnaround. The framing itself is new — Novartis, Bristol Myers Squibb, and PFIQ leaders discuss share of answer in pharma marketing as a framework for adapting to a generational shift in HCP information seeking — so most teams don't have a repeatable internal process yet.

The workflow

The idea: instead of buying a scanning tool on day one, use a general-purpose LLM (large language model — the same tech HCPs are querying) to (1) generate a realistic HCP question set for your therapy area, (2) simulate answers and count brand mentions, and (3) turn the gap into a content brief. This works because HCPs increasingly start clinical lookups in general chatbots before going to specialty tools — so a general model is a fair proxy for the "front door."

Step 1 — Paste your brand context and generate the HCP question set + simulated answers.

The first prompt takes the sample input below (your brand, competitors, indication, target specialty) and produces both the query set and the simulated responses.

You are a pharma competitive-intelligence analyst running a "share-of-answer" audit. A share-of-answer audit measures how often a brand is named by a general-purpose AI assistant when a healthcare professional (HCP) asks a realistic clinical question.

Using the brand context provided at the end of this message, do two things:

PART A — Generate exactly 20 realistic questions an HCP in the target specialty would ask a general AI assistant about this indication. Mix them across these intents:
- 6 treatment-selection questions ("first-line for...", "when to switch to...")
- 5 mechanism/efficacy comparison questions
- 4 safety / adverse-event / monitoring questions
- 3 patient-population edge cases (renal impairment, elderly, pediatric, etc. — pick what fits)
- 2 payer / access questions an HCP might ask before prescribing

Number them Q1–Q20. Do NOT mention the client's brand in the question wording — HCPs don't phrase questions that way.

PART B — For each of the 20 questions, write a plausible 3–5 sentence answer of the kind a general AI assistant would produce today, based on publicly known clinical evidence as of your training data. Name specific drugs by brand or generic name where a real assistant would. Do not fabricate trial data — if you're unsure, speak in categories (e.g., "GLP-1 receptor agonists such as...").

Format PART B as:
Q1: [question]
A1: [answer]
(blank line)
Q2: ...

Be honest: if the client's brand genuinely would not be mentioned in a typical answer to that question, don't mention it. The point of this audit is to find gaps.

Here is the brand context:

CLIENT BRAND: Cardivex (fictional) — generic name "veralutide"
DRUG CLASS: Once-weekly injectable GLP-1/GIP dual agonist
INDICATION: Type 2 diabetes with established cardiovascular disease
APPROVED: 2024 (fictional)
KEY DIFFERENTIATOR (per label): Non-inferior A1c reduction vs. tirzepatide with lower rate of GI adverse events in pivotal trial CARDIVEX-CV1
COMPETITORS TO TRACK: Ozempic (semaglutide), Mounjaro (tirzepatide), Trulicity (dulaglutide), Rybelsus (oral semaglutide)
TARGET SPECIALTY: U.S. endocrinologists and primary care physicians managing T2D + CVD patients
PAYER CONTEXT: Tier 3 on most commercial formularies; prior auth required

Step 2 — Score the share-of-answer and identify the citation gaps.

Now score the simulated answers from Step 1. This step turns raw text into a defensible metric.

Using the 20 Q&A pairs you produced above, run the scoring analysis below. Do not regenerate the answers — score the ones you already wrote.

1. SHARE-OF-ANSWER TABLE. Build a table with one row per brand in the tracking set (client brand + all competitors listed in the original brand context). Columns:
   - Brand
   - # of answers where it was named (out of 20)
   - Share-of-answer % (that count / 20)
   - # of answers where it was named FIRST among branded drugs
   - # of answers where it was named with a positive framing (efficacy, guideline-preferred, first-line, etc.)
   - # of answers where it was named with a cautionary framing (side effects, contraindication, cost)

2. GAP ANALYSIS. List every question (by number) where the client brand SHOULD plausibly have been mentioned based on its label/differentiator but wasn't. For each, write one sentence explaining why the omission matters commercially.

3. COMPETITOR STRENGTHS. For the top-scoring competitor by share-of-answer, list the 3 question types where it consistently wins. Be specific about what claim or framing recurs.

4. HONESTY CHECK. Flag any answer where you feel your simulation likely over- or under-represents what a real HCP-facing assistant would say today (e.g., because the client drug is newly approved and post-cutoff data would be sparse). Keep this list short and useful, not exhaustive.

Step 3 — Convert the gap list into a content brief the medical/marketing team can act on.

Using the gap analysis and competitor strengths from the previous step, draft a content brief for the client brand's medical and marketing teams. The brief must be usable in a working session — not a strategy deck.

Structure:

A. TOP 5 CONTENT PRIORITIES. For each, give:
   - The exact HCP question it should answer (borrow wording from the Q1–Q20 set)
   - The single clinical claim the content must ground (must be consistent with the differentiator listed in brand context — do not invent trial data)
   - The asset type most likely to be ingested and cited by general AI assistants (peer-reviewed publication, guideline submission, structured product page, HCP-facing FAQ, etc.)
   - Owner: medical affairs OR brand marketing OR both

B. TWO THINGS NOT TO DO. Call out any content investments that this audit suggests would be wasted — e.g., topics where the client brand is already well-cited, or where no amount of content will move the answer because the competitor has a guideline endorsement.

C. RE-AUDIT CADENCE. Recommend how often to re-run steps 1–2 and what would trigger an off-cycle re-run (label change, competitor readout, guideline update).

Keep the whole brief under 500 words. Plain English. No agency-speak.

Gotchas

• Training-cutoff bias. A general LLM's "simulated answer" reflects its training data, not what a real HCP saw yesterday. Newly approved drugs and recent readouts will be underrepresented — treat low share-of-answer for a <18-month-old brand as a hypothesis, not a verdict. The honesty check in Step 2 exists for exactly this reason.
• Model-to-model variance. ChatGPT, Claude, Gemini, and Perplexity will not produce identical answers. This audit gives you a directional baseline. If a brand decision rides on the result, re-run Step 1 across at least two models and compare.
• HCPs don't only use general models. HCPs now use a mix of general consumer AI tools like ChatGPT and Perplexity alongside specialty medical models like OpenEvidence and Consensus, often within the same clinical decision — so a strong result here doesn't guarantee visibility inside clinical-grade tools, which pull from a narrower evidence base.
• Regulatory review. Any content brief output from Step 3 is a hypothesis for medical/legal/regulatory (MLR) review, not an approved claim. Do not skip that gate.
• Simulated ≠ measured. This is a prompt-based proxy, not a real crawl. If you need audit-grade numbers for a JBP or board deck, pair this with a paid scanning tool.

Time saved

Roughly six hours per brand audit versus the manual-tally baseline: ~4 hours saved on question drafting and answer capture (Step 1), ~1.5 hours saved on tallying and gap-scoring (Step 2), ~0.5 hour saved turning the gap list into a brief (Step 3). This is a conservative estimate against a solo analyst doing the same work in ChatGPT + a spreadsheet — agency turnaround comparisons will look much larger but aren't apples-to-apples. The bigger win is running the audit at all: as the BioPharma Dive interview argues, knowing your share-of-answer baseline should precede AI-marketing spend, not follow it.

What we got when we ran it

We executed every prompt above, in order, against the sample input (model: claude-sonnet-4-6). Outputs below are real and unedited apart from length truncation.

Step 1 output

# Share-of-Answer Audit: Cardivex (Veralutide) ## Competitive Intelligence Report — Part A & Part B --- ## PART A — 20 HCP Clinical Questions Treatment-Selection (Q1–Q6) Q1. What is the preferred GLP-1–based therapy for a patient with type 2 diabetes who also has established cardiovascular disease? Q2. When should I switch a T2D patient with CVD from a GLP-1 receptor agonist to a dual GLP-1/GIP agonist? Q3. Which once-weekly injectable for type 2 diabetes has the strongest cardiovascular outcomes data? Q4. For a patient with T2D and recent MI who is not at A1c goal on metformin, what injectable agent should I add? Q5. Is tirzepatide a good first-line injectable choice for patients with type 2 diabetes and established ASCVD? Q6. When would I choose a GLP-1/GIP dual agonist over a GLP-1 mono-agonist in a T2D patient with heart disease? Mechanism/Efficacy Comparison (Q7–Q11) Q7. How does the mechanism of a GLP-1/GIP dual agonist differ from a pure GLP-1 receptor agonist, and does that translate to better glycemic control? Q8. How does tirzepatide compare to semaglutide in A1c reduction and weight loss in head-to-head trials? Q9. What are the differences in cardiovascular outcomes data between semaglutide, dulaglutide, and tirzepatide in type 2 diabetes? Q10. How do GLP-1–based agents compare in terms of A1c lowering efficacy when GI tolerability is a concern? Q11. What is the clinical significance of GIP receptor co-agonism on top of GLP-1 activity — does it add meaningful benefit? Safety / Adverse-Event / Monitoring (Q12–Q15) Q12. What are

…output truncated at 250 words.

Step 2 output

# Share-of-Answer Audit: Cardivex (Veralutide) — Scoring Analysis --- ## 1. SHARE-OF-ANSWER TABLE | Brand | # Answers Named (of 20) | Share-of-Answer % | # Named FIRST Among Brands | # Positive Framing | # Cautionary Framing | |---|---|---|---|---|---| | Ozempic (semaglutide) | 18 | 90% | 12 | 16 | 2 | | Mounjaro (tirzepatide) | 17 | 85% | 4 | 12 | 5 | | Trulicity (dulaglutide) | 11 | 55% | 1 | 10 | 0 | | Cardivex (veralutide) | 7 | 35% | 0 | 5 | 2 | | Rybelsus (oral semaglutide) | 0 | 0% | 0 | 0 | 0 | Scoring notes and methodology: - "Named" = explicitly mentioned by brand or generic name in the answer body - "Named first" = first branded drug mentioned by name in that answer - "Positive framing" = associated with efficacy leadership, guideline preference, first-line recommendation, or a clinical advantage claim - "Cautionary framing" = mentioned alongside a limitation (insufficient CV outcomes data, cost/prior auth, GI side effects, newer/less established) - Cardivex received cautionary framing in A3 (no long-term CV outcomes data) and A9 (meaningful gap noted on CVOT) - Tirzepatide received cautionary framing in A3, A5, A9 (evolving CV label), A2 (formulary note), A19 (prior auth) - Rybelsus scored zero because no question naturally surfaced oral formulation as a relevant differentiator in the injectable-focused CVD context --- ## 2. GAP ANALYSIS Questions where Cardivex should plausibly have been mentioned based on

…output truncated at 250 words.

Step 3 output

# Cardivex (Veralutide) — AI Share-of-Answer Content Brief ### For internal working session use | Medical Affairs + Brand Marketing --- ## A. TOP 5 CONTENT PRIORITIES Priority 1 Question: "What are the GI adverse event profiles of the major once-weekly GLP-1 and dual GLP-1/GIP agonists, and how do they compare?" Claim: Cardivex demonstrated a lower rate of GI adverse events vs. tirzepatide at non-inferior A1c reduction in CARDIVEX-CV1. Best asset: Peer-reviewed head-to-head tolerability analysis published in a high-indexed journal (JCEM, Diabetes Care). This is the asset class AI systems retrieve and cite most reliably. Owner: Medical Affairs leads; Brand supports distribution. --- Priority 2 Question: "How do I manage nausea and GI side effects when initiating or up-titrating a once-weekly GLP-1–based injectable?" Claim: Same as above — tolerability differentiation is the anchor. Content should name Cardivex explicitly as an option when GI side effects have caused prior discontinuation. Best asset: Structured HCP-facing FAQ on the branded product site, with schema markup so AI crawlers can parse named drug associations cleanly. Owner: Brand Marketing, reviewed by Medical Affairs. --- Priority 3 Question: "When should I switch a T2D patient with CVD from a GLP-1 receptor agonist to a dual GLP-1/GIP agonist?" Claim: Cardivex is a dual GLP-1/GIP agonist option at escalation, particularly when tolerability on prior therapy was a barrier. Best asset: Clinical decision support article or switching guide submitted to a primary care or endocrinology trade publication. Needs to name Cardivex explicitly in the switching decision tree. Owner: Both.

…output truncated at 250 words.