AI visibility is measurable — just not in a single dashboard with green arrows. There is no Search Console for ChatGPT. No reliable impression count for Claude. What you can measure is whether your clinic gets named, for which intents, in what position, against which competitors, and whether those citations convert into real intake — using a disciplined weekly process.
We treat this like a clinical trial for your brand: fixed prompts, fixed platforms, fixed logging method, week over week. Change one variable on-site or off-site, then observe whether mention rate moves.
Build your prompt library
Group prompts by intent. Don't only test branded searches — those always under-report your problem. Include:
- Condition-led: sciatica, whiplash, desk neck pain, pregnancy back pain, headache
- Audience-led: athletes, office workers, seniors, new patients
- Logistics-led: same-week appointment, Saturday hours, walk-ins, insurance
- Comparison-led: chiropractor vs physiotherapist for [condition]
- Branded: '[clinic name] reviews' and '[doctor name] chiropractor'
Aim for thirty to fifty prompts that reflect your market. Store them in a spreadsheet tab. Run the same set every Monday morning before you change anything on the site — consistency matters more than volume.
What to log each week
- Named yes/no — is your clinic in the answer at all?
- Position — first recommendation vs 'also consider' vs footnote only
- Intent match — cited for sports when you only do general wellness?
- Competitor set — who appears every week and with what descriptor
- Source URLs — domains Perplexity or ChatGPT cite in footnotes
- One-line summary — how the model describes you ('family-focused', 'sports specialist', etc.)
Screenshot answers monthly. Models drift with updates; you'll want a visual record when a competitor suddenly appears or disappears.
Connect prompts to front desk reality
Add 'ChatGPT / AI search / Perplexity' to your new-patient intake form — even as an optional field. Citation lift in testing should eventually show up as real-world mentions. If models name you weekly but nobody books, the bottleneck is usually offer clarity: insurance list outdated, booking link buried, wrong neighbourhood in copy, or online scheduling broken on mobile.
Benchmark cadence and when to change course
Give schema and on-page changes four to eight weeks before declaring failure. Perplexity often moves first; ChatGPT lags. If mention rate is flat after a rebuild plus seeding, revisit positioning — you may be semantically consistent but positioned for a niche you don't actually want.
The spreadsheet becomes your GEO record: what you changed, when, and what moved. That's how you justify the work to partners and avoid chasing vanity metrics that don't connect to booked appointments.
Tools and habits that keep you honest
Use the same browser profile or incognito window each week so personalization doesn't skew results. Test ChatGPT, Claude, and Perplexity separately — they cite different sources. When Perplexity footnotes a directory you ignored, that's your next seeding target. Share the spreadsheet with your front desk so intake data and prompt logs tell one story: visibility without bookings usually means fixable friction on the site, not a failed GEO campaign.