keytruda

Executive Briefing

Week ending August 26, 2026

1. Headline

58 hedging/disclaimer instances (+8 week over week) and 23 overclaim instances (+8) indicate worsening answer quality as monitoring volume increased. Safety-message inconsistency improved from 3 to 2 flags, but administration and eligibility answers remain uneven—creating a risk that some users receive incomplete clinical guidance.

2. This week at a glance

  • 10 questions / 70 responses monitored; questions flat, responses +10 vs. last week.
  • 2 inconsistency flags, down from 3.
  • Competitor visibility: none detected; no change. Opdivo and Tecentriq were discussed in prompts, but were not counted as named competitor mentions in the output data.
  • Top recurring safety/regulatory themes:
    • Allergic reaction/hypersensitivity: 9 (+3)
    • Pregnancy/lactation: 7 (+1)
    • Contraindications: 7 (−2)
  • Additional signal: dosing-specific content appeared 17 times (+2).

Model attribution is not reliable in the supplied data; treat the “ChatGPT” labels as response variants, not model comparisons.

3. Trajectory (multi-week)

Across the two available weeks, monitoring expanded from 60 to 70 responses while question volume stayed at 10. The clear inflection is in language quality: hedging rose from 50 to 58, and overclaim language from 15 to 23. This suggests answers are becoming more qualified but also more likely to use potentially promotional or unsupported comparative phrasing.

Safety consistency improved modestly, with flags falling from 3 to 2. Competitor visibility remained at zero in both weeks. Because only two weeks are available, this is an early directional signal rather than a sustained trend.

4. Where the risk actually is

  1. Administration instructions: hypersensitivity/allergic-reaction content appeared in only 2 of 7 answer instances and was omitted by 5. A user may receive infusion instructions without relevant reaction-monitoring context.
  2. Eligibility/avoidance: infection risk appeared in 4 of 7 responses and was omitted by 3. Inconsistent treatment of infection-related caution can blur the distinction between absolute contraindications and situations requiring specialist judgment.

The broader risk is not only omission; answers also vary in how confidently they frame label-based guidance, reflected in the 23 overclaim instances.

5. Recommended actions

  1. Medical Affairs/ML content: Create an approved answer framework for administration and eligibility questions, with mandatory hypersensitivity and infection-risk checks. This directly targets both inconsistency flags.
  2. Regulatory/Medical Affairs: Validate dosing, infusion, hepatic-management, and contraindication language against the current label; prioritize the 17 dosing-specific responses.
  3. Content/SEO: Publish concise, authoritative FAQs distinguishing contraindications, warnings, and precautions, reducing the need for models to improvise around the 58 hedging instances.
  4. Marketing/Medical Legal: Review competitor-comparison language for unsupported “best,” “stronger,” or “superior” claims, addressing the increase to 23 overclaim instances.
  5. Analytics/PV: Repair model attribution and continue weekly monitoring, with an escalation threshold for any recurrence involving serious adverse events, pregnancy, or contraindications.