A Patient Found Your Clinic on ChatGPT and Then Asked "Are You Even Real?" Here's Why That Keeps Happening.
It starts as a good sign — a patient mentions they found you through an AI recommendation. Then comes the follow-up question: "But it said you were on Oak Street, and I can't find that address." Or "ChatGPT said you do hair transplants — do you?" Or the patient who confirmed an appointment based on hours the AI quoted that you haven't kept in two years. Inaccurate AI citations are not just embarrassing. They cost you patients who give up before they call. Here's exactly what's causing it.
How AI Systems Actually Build a Picture of Your Clinic
There is a common misconception that AI systems read your website and cite what they find there. This is partially true but significantly incomplete. AI systems use a process called entity resolution — they cross-reference multiple data sources to construct a unified picture of what your clinic is, where it is, what it does, and when it is open. The sources they draw from include your website, your Google Business Profile, Healthgrades, Yelp, Zocdoc, Vitals, any local business directories you've ever been listed on, and any web pages that mention your clinic's name and address.
When all of these sources agree — same name, same address, same phone number, same service descriptions — the AI constructs an accurate, confident entity profile and cites it reliably. When they disagree — as they do for the majority of independent medical clinics that have been operating for more than three years and have accumulated inconsistent data across dozens of directories — the AI faces a resolution problem. It must decide which source to trust. The resolution logic varies by platform but generally weights recency, source authority, and frequency of agreement across sources. What it produces is a citation that seems confident but may be assembling information from different points in your clinic's history.
This is how a patient ends up with an old address. This is how a clinic gets cited as offering services it discontinued. This is how hours from a directory that was set up in 2019 and never updated become the hours an AI confidently quotes to a patient deciding whether to call.
The Specific Data Conflicts That Produce Inaccurate Citations
Stale directory listings. Most independent medical clinics were listed on dozens of directories over the years — some by the clinic owner, many automatically populated from other data sources. A clinic that moved locations, changed its phone number, rebranded, or expanded its service menu in the last five years almost certainly has inconsistent data across these listings. Yelp might have the old address. A local chamber of commerce directory might have the original phone number. A medical directory aggregator might have the pre-rebrand name. Each of these inconsistencies is a data point AI systems cross-reference when constructing the clinic's entity profile. The more conflicting data points exist, the less reliable the AI's synthesised citation becomes.
No formal entity declaration on the website. The solution to entity resolution problems is to give AI systems a formal, authoritative declaration of the clinic's entity data — one they can weight above all other sources as the canonical record. Organization and MedicalBusiness schema markup on the clinic's website serves this function. It formally declares: this is the clinic's legal name, this is its current address, this is its current phone number, these are its current services, and these are the credentials of its practitioners. When an AI system encounters a clinic website with complete @graph schema containing consistent @id references, it treats that as the authoritative entity source and resolves conflicts with other data sources in favour of the schema data.
No ai-index.json file. Beyond schema markup, the ai-index.json file at the website root serves as a machine-readable business card specifically for AI retrieval systems. It provides structured entity data — clinic name, address, services, practitioners, credentials, hours — in a JSON format optimised for AI parsing. When AI systems encounter an ai-index.json file during crawling, they use it as a high-confidence entity reference that supplements and reinforces the schema data on the website. Together, schema and ai-index.json create two overlapping authoritative entity declarations that AI systems use to anchor their entity resolution — making accurate citation significantly more reliable.
The Citations That Hurt Most — And Why
Not all inaccurate AI citations create the same damage. A citation with a slightly incorrect address might confuse a patient but they will likely call to confirm. A citation that lists services the clinic no longer offers generates inquiries the clinic cannot convert — wasted calls from patients who need something else. A citation with incorrect hours leads patients to show up when the clinic is closed, or to not call at all because the AI-quoted hours don't fit their schedule.
The most damaging type of inaccurate citation is entity conflation — where an AI system partially merges two similarly named clinics into one recommendation. This happens most often with common clinic naming patterns: "Advanced Dermatology," "Premier Eye Care," "Total Orthopaedics" — names that are shared by multiple unrelated clinics across different cities. AI systems that cannot reliably distinguish between entity instances of similarly named businesses sometimes cite one clinic with the address of another. For the clinic whose address is incorrectly attributed, this is an invisible problem — patients looking for the other clinic show up at their location, creating confusion with no obvious connection to AI search.
How to Establish a Reliable Entity Signal That AI Systems Cite Correctly
The fix operates on three levels. First, audit and correct NAP consistency across all directories. This means identifying every online listing that contains the clinic's name and address — Google Business Profile, Yelp, Healthgrades, Vitals, Zocdoc, local business directories, hospital referral pages, insurance provider directories — and verifying that the name, address, and phone number are identical across all of them. Any discrepancies are corrected to match the clinic's current canonical information. This is the foundational step without which all other entity-building work is less effective.
Second, implement Organization and MedicalBusiness schema in @graph format on the clinic's homepage and location pages. This establishes the formal authoritative entity declaration that AI systems weight above conflicting directory data. The schema includes the clinic's current name, address (with full geo-coordinates), phone, hours, services, practitioners, and sameAs social links. The @graph pattern with consistent @id references across all pages ensures that every page on the clinic's website contributes to the same entity graph rather than creating disconnected schema blocks that AI systems cannot reliably associate with each other.
Third, deploy ai-index.json at the website root. This file formalises the entity data in a format specifically designed for AI retrieval systems — providing the structured machine-readable record that Perplexity, ChatGPT, and other AI systems use as a high-confidence entity reference. Together these three steps replace the fragmented, conflicting data landscape with a consistent, authoritative entity signal that AI systems cite accurately.
If your clinic has ever moved, rebranded, changed its phone number, or significantly changed its service menu — and if you have been in operation for more than three years — the probability of entity data inconsistencies causing AI citation inaccuracies is very high. The free Iris agentic readiness audit assesses citation consistency as one of its six scored dimensions and tells you exactly what it finds in 60 seconds. If the citation consistency score is low, it is almost always the underlying cause of inaccurate AI recommendations — and it is fixable in a matter of weeks.