Visibility
Did the practice surface?
For a dated patient question on a specific AI/search surface, did the practice appear at all—and in what context?
AI visibility & accuracy
Being found by AI and being represented accurately are different problems.
A practice can appear in an AI answer and still be described incorrectly. It can also have a correct website and still be omitted, summarized from stale information, or represented differently across systems.
The practical standard: verify what is true about the practice, make the first-party website clear and current, correct controllable source problems, allow time for changes to propagate, and then retest the AI behavior itself. Those are separate evidence layers.
Two questions, not one
Visibility asks whether the practice appears. Accuracy asks whether the answer is right.
Both matter to a patient deciding whom to call, whether a plan is accepted, which services are offered, or what to expect before booking.
Accuracy
Were decision-critical facts correct?
Doctors, location, services, hours, vision plans, insurance, policies and patient actions can be more important than simple mention.
Consistency
Does the story stay stable?
One favorable answer is not permanence. Repeated or matched tests can reveal stale, conflicting or unstable representations.
Start with the practice
You need a reference point before you can call an AI answer wrong.
GSA uses verified practice facts as the reference point for testing. Conceptually, that means maintaining the current facts a patient could reasonably rely on—not creating a secret AI profile.
- Identity: practice name, providers and current roles.
- Location: address, hours, contact details and accessibility.
- Care: services, capabilities, technology and important clinical boundaries.
- Coverage: relevant vision plans, insurance relationships and patient-facing caveats.
- Actions: booking, forms, reorders, directions and other real patient workflows.
When something is wrong
Correcting the source is not the same as correcting the AI answer.
A clean remediation process separates what the practice controls from what happens downstream.
1 — Source correction
Fix the controllable information.
Correct inaccurate, stale or conflicting information on the practice website or another source the practice can legitimately update.
2 — Propagation
Verify the correction is visible downstream.
A change being published does not prove every search or AI system has refreshed it. Propagation is observed separately where possible.
3 — AI behavior
Retest the same question.
Use the same surface and matched question after the correction has had time to propagate. The new answer is evidence of model behavior—not proof of permanence.
Evidence before folklore
“AI-ready” should mean more than adding whatever tactic is fashionable this month.
Different platforms document different controls, crawlers and behaviors. Where a platform publishes first-party guidance, that guidance should come before industry folklore. Where important behavior is undocumented, the honest answer is to test it—or say it is unknown.
This is why GSA separates ordinary search/indexing controls, user-directed retrieval, model-training choices, source correction, propagation and observed AI output rather than treating them as one universal “AI SEO” mechanism.
No single file, schema block, crawler setting or technical change is presented as a guarantee that an AI system will rank, cite, recommend or describe a practice in a particular way.
Questions for any vendor
If someone says your website is “AI-ready,” ask what they can actually prove.
Useful answers should be specific about the system being tested, the facts being checked and the evidence that separates implementation from outcome.
What exactly are you testing?
Which AI/search products, which patient questions, which dates, and under what test conditions?
What is your source of truth?
How do you know the provider, service, insurance, hours or policy fact you are judging is actually current?
What happens when an answer is wrong?
Do you identify the likely source problem, correct what is controllable, verify the correction and then run a matched retest?
Do you separate correction from propagation?
Can you distinguish “we changed the website” from “the changed information propagated” and from “the AI answer changed”?
Which claims come from platform documentation?
Ask what the AI/search company itself says—and what is instead an experiment, an inference or a marketing convention.
What do you refuse to promise?
A credible methodology should have boundaries around rankings, citations, recommendations, permanence, appointments, revenue and ROI.
A real example
Park Slope Eye showed why visibility alone is not enough.
The practice already had substantial search visibility and was being surfaced by AI. In separate patient-intent testing, Gemini still described both VSP and EyeMed network facts incorrectly.
The lesson is not that one AI system is always wrong. It is that surfacing, factual accuracy, source correction and later AI behavior have to be measured as different things.
Your practice
Start by observing what patients are actually being told.
GSA’s complimentary AI Discoverability Report tests a governed set of major AI/search surfaces, preserves the evidence and identifies material factual discrepancies without turning one testing window into a universal grade.