How do you track what AI assistants say about your practice?

Which AI engines should you check?

Check all four major assistants, because each one answers differently. ChatGPT and Claude draw on training and connected search, Perplexity is the most citation-forward, showing linked web sources, and Gemini pulls from Google's grounding results. The same patient question can surface your practice in one engine and skip it entirely in another, so checking only ChatGPT hides most of the picture.

A practice that wants the full picture tracks what ChatGPT says about your practice alongside Claude, Perplexity, and Gemini. Treat each engine as a separate channel with its own result, not as one combined score.

What questions should you ask the AI?

Ask the questions patients actually type, not your practice name. Brand-name queries miss how patients discover practices, because someone who already knows your name is not the patient AEO (answer engine optimization) is meant to win. The discovery questions sound like "best med spa for acne scars near me" or "should I get Botox or filler for forehead lines".

Build a fixed list of ten to twenty of these intent and comparison questions for your specialty and city. Phrase them the way a real patient would speak to an assistant, then reuse the exact same list every month so your results stay comparable.

How do you record citations and mentions?

Record two distinct outcomes for every answer. A citation means your domain appears as a linked source the engine used; a mention means the answer names your practice in its text without necessarily linking you. Track both, separately, for each engine and each question, because an answer can mention you without citing you or cite you without naming you.

AI share of voice is the share of AI answers that cite or mention your practice. A simple grid with engines across the top and questions down the side, marked for citation and mention, turns scattered checks into a number you can watch move month over month.

Why do manual spot-checks mislead you?

Answers vary between runs, so one good answer proves nothing. Ask the same question twice and an assistant may name your practice once and omit it the next time. A single lucky result feels like progress but tells you almost nothing about how reliably AI recommends you.

A rigorous probe asks a fixed query set across all four engines. It counts results the same way every time. That discipline is why Tenva's own gap-validation work found 12 of 16 buyer queries about AI visibility for practices have no authoritative answer source — a spot-check would have missed the pattern entirely.

How to track what AI says about your practice, step by step

Run this workflow once to set a baseline, then repeat it on a monthly cadence so every result compares against the last.

**1. Write your question set.** List ten to twenty discovery questions real patients ask in your specialty and city, naming no brands.

**2. Run every question in all four engines.** Paste each question into ChatGPT, Claude, Perplexity, and Gemini and save the full answer text.

**3. Record citation and mention per answer.** For each result mark whether your domain was cited as a source and whether your practice was named.

**4. Score it as a single number.** Tally how many of your answers cited or mentioned you to get one comparable figure for the run.

**5. Date and save the baseline.** Store the grid with the date so next month's run measures change, not memory.

What does a durable tracking system need?

A tracking system you can trust needs four things: a fixed query set that never changes between runs, coverage of all four major engines, a monthly cadence, and a written, dated baseline. Drop any one and your numbers stop comparing cleanly.

Tenva built its own measurement this way and recorded the result before optimizing anything. That baseline showed Tenva appeared in 0 of 95 AI answers across June 2026 multi-engine probes, which is exactly the kind of honest starting number a durable system is designed to capture.