What is AI Citation Share? — practitioner guide

For 25 years patients found their doctor through Google. They typed a question, scrolled blue links, clicked one, judged the site, and either booked or kept looking. "Being found" meant ranking in those blue links.

Between 2024 and 2026 a new layer slipped in front of the blue links: the AI answer layer. Google's AI Overviews, ChatGPT's web-search mode, Claude's own web search, Perplexity, and Gemini — all of these produce a direct answer to the user's question, with 3-5 cited sources. Most users now stop at that answer layer; they never click through to the blue links.

Definition

AI Citation Share is the rate at which a practitioner (physician, dentist, attorney, financial advisor) is cited in AI engine answers:

AI Citation Share = (cited probes / total probes) × 100

A probe is a single test query sent to an AI engine (e.g. "best aesthetic surgeon Istanbul"). If the response contains a link to your canonical domain, the probe counts as "cited". Range: 0%–100%.

Why Google ranking isn't enough

A #1 Google ranking means your link can be clicked. In the AI answer layer the user gets the answer without clicking. Being cited as the source is now what "being found" means. The click is a by-product; the real prize is being named in the answer.

AI engines weight different signals than Google: structured data richness, entity reconciliation clarity, content cadence, verifiability of authorship, audit defensibility. The SEO agency's "backlinks + keyword density" recipe loses here.

How is it measured?

MedSelect's AIDO runner executes a standard probe set every night at 04:30 UTC for every active tenant:

1. 5-25 standard queries per vertical (e.g. for aesthetic: "botox prices 2026", "thread lift vs liquid facelift").
2. Each query fires to all four AI engines in parallel.
3. Engine responses are parsed; citation URLs are extracted.
4. If the tenant's canonical domain appears in any citation, the probe counts as cited.
5. The aggregate ai_citation_daily row is upserted.

Full methodology: /ai-citation-methodology.

What number is "good"?

Varies by vertical and content cadence. General baseline:

• Day 0 (new tenant): 0%-5%. AI engines don't know you yet.
• Day 30: 5%-15%. Voice card + first 3-5 articles + Schema.org markup deployed.
• Day 90: 15%-30%. Steady cadence + entity reconciliation + positive sentiment trajectory.
• Day 180+: 25%-45%. Stabilises. Higher than that may indicate uncontested category.

100% isn't a target — AI engines typically cite 3-5 different sources per answer; 30%-40% cited rate is strong.

What practitioners should do

1. Voice card session: 90-minute capture of clinical philosophy + vocabulary + opinion topics + decline-list. AI engines learn "what does this practitioner talk about" from the voice card.
2. Steady publishing: 4-6 deep articles per month. Each passes the reviewer chain.
3. Schema.org markup: profile page + every article emits Physician / Article / MedicalWebPage JSON-LD.
4. Entity reconciliation: sameAs links on LinkedIn, Wikidata, professional society pages.
5. Outcome data sourcing: numerical claims (success rate, case count) link to registry / journal / your own case series.

You can't improve what you can't measure

The strongest feature of AI Citation Share is its measurability. In SEO "Google's algorithm" is a black box; AI Citation Share has the probe + parser + audit log as a verifiable pipeline. The practitioner sees exactly which query they appear on, alongside which competitors, and with what sentiment.

Public leaderboard: /transparency. Instant snapshot for your own URL: /probe.