MedSelect.ai

Methodology

How we measure, produce, and verify physician content in the AI era.

MedSelect runs in three layers: measurement (AIDO), production (voice card + editorial chain), compliance (KVKK/GDPR). This page explains the inner workings of each.

AIDO — AI Citation Share measurement

Our north-star metric: how often physician content gets cited by large language models. A daily cron sends every tenant's probe set against ChatGPT, Claude, Perplexity, and Gemini; we parse the response text for URL and physician-name matches.

  • 1. Each tenant gets 5–25 intent-tagged probe queries.
  • 2. Cron runs at 04:30 UTC and queries the four systems sequentially (1.5s pacing).
  • 3. Response text is parsed: we look for tenant canonical_domain matches + physician full_name matches and count them as citations.
  • 4. The daily (tenant × system) aggregate is upserted into ai_citation_daily; the admin panel surfaces the trend.

Voice card — capturing the physician's voice

Each physician gets a two-layer voice card: long-form guidance (vocabulary, clinical philosophy, opinion / decline topics) and structured guard rails (never-discuss list, always-disclaim list, preferred CTA variants). The card is open in the side panel whenever the editor writes a draft.

  • Vocabulary preferences: words the physician uses / avoids
  • Clinical philosophy: patient selection, expectation management, follow-up approach
  • Opinion topics: the split between strong-opinion topics and decline topics
  • Never-discuss list: hard-stop topics (off-label, pricing, etc.)
  • Always-disclaim list: lines that must appear on every piece
  • Preferred CTA list: call-to-action variants the physician has approved

Editorial chain

Every draft passes through a five-state machine. High-risk topics (off-label, high-dose, paediatric, etc.) add a mandatory senior-medical-advisor review. Every transition is written to the append-only audit log.

  1. Brief: operator + physician agree on the target query
  2. Draft: physician writes the draft (voice card in side panel)
  3. Medical review: a second physician signs off on medical accuracy (author ≠ reviewer)
  4. Senior advisor review: mandatory for high-risk; senior_medical_advisor role approves
  5. SEO review: medselect_seo role does the final technical SEO pass
  6. Published: pushed to the tenant's publish API; GSC/GA4 tracking begins

KVKK / GDPR compliance layer

Data hosted in the EU. No third-party identity providers. Append-only audit log with daily SHA-256 integrity chain. Bouncing / complaining addresses are auto-suppressed. Magic-link tokens are 15-min, single-use, stored only as a SHA-256 hash.

Let's dig into the details.

Reach out to discuss what the methodology looks like for your physician — which probe set, which voice card depth, which editorial cadence fits your operation.

Get in touch
    MedSelect — Methodology · MedSelect