Somewhere in your area tonight, a person with a failing tooth will open ChatGPT instead of Google. They will type something like: “I need dental implants and I’m scared of choosing the wrong dentist. Who should I trust near me?”
The assistant will answer. It will name two or three practices, often with a sentence of reasoning for each: strong reviews for implant work, an experienced clinical team, clear information about the procedure. The patient will read those names, feel their anxiety drop, and start their shortlist with them.
Here is the uncomfortable part. If your practice was not in that answer, you did not lose that patient. You were never in the running, and no report you receive from any marketing agency will ever show you the moment it happened.
I am a practicing clinician who also runs a marketing agency, so I sit on both sides of this. In this piece I want to explain, in plain language, how these recommendations are actually assembled, why most practices are invisible to them, and what to do about it, in order. No invented statistics, no scare tactics. The mechanics are unsettling enough on their own.
First, Understand What Kind of Answer This Is
A traditional search result page is a list of options. It shows ten links, the patient scans, compares, clicks around. Being third or fifth still gets you seen.
An AI answer is not a list of options. It is a recommendation. The assistant compresses everything it can find into two or three names and, crucially, it gives reasons. That framing does something psychologically powerful: it arrives with the tone of a knowledgeable friend, not an advertisement. Patients treat it accordingly.
There is no page two of an AI answer. There is no position five. You are either one of the names, or you are absent, and absence is invisible to you because these conversations happen inside private chat windows that no analytics tool can see.
How the Answer Is Actually Assembled
AI assistants are not consulting a secret ranking of dentists. When a patient asks for a recommendation, two things happen, usually together.
The model draws on what it already knows. Large language models are trained on enormous amounts of public text: websites, directories, articles, reviews. If your practice has a substantial, consistent footprint in that text, the model may already “know” you: your name, your location, what you are known for. If your footprint is thin or contradictory, you are a blur, and models do not confidently recommend blurs.
The model looks things up in real time. Modern assistants also search the live web while composing an answer. ChatGPT can browse. Perplexity is built around retrieval. Google’s AI Overviews are generated on top of Google’s own index, with the map and business profile data attached. This means the answer is heavily shaped by what the assistant can read at that moment: your website as machines see it, your business profile, your reviews, directories, and anything written about you.
The assistant then does something very human. Faced with a high-stakes health question, it looks for reasons to trust before it names anyone. Can I verify this practice exists and does what it claims? Are the credentials checkable? Do the reviews support the recommendation I am about to make? Health is a domain where AI systems are deliberately cautious, which raises the bar for who gets named at all.
The Five Signals That Decide Who Gets Named
Across the systems that matter, the same categories of evidence keep deciding the answer. None of this is secret. It is simply work that almost no practice has done.
1. Entity Clarity: Can the Machine Tell Who You Are?
Your practice, to an AI system, is an entity: a name connected to a place, people, services, and a history. If your name, address, and details are consistent everywhere they appear, the entity is solid. If your website says one thing, an old directory says another, and your business profile a third, the machine sees contradiction. Contradiction reads as risk, and risk gets skipped, silently.
2. Machine-Readable Structure: Does Your Website Speak to Machines at All?
Patients see your website’s design. Machines see its structure. Schema markup is the layer of code that tells a machine, unambiguously: this is a dental practice, here is where it is, these are the clinicians and their qualifications, these are the treatments, these are answers to common questions. Most practice websites have none of it. A machine reading such a site has to guess, and cautious systems do not build recommendations on guesses. I have written a separate plain-language guide to schema for practice owners if you want the specifics.
3. Review Substance: What Does the Text Say, Not the Stars?
This is the one that surprises owners most. A wall of five-star ratings that all say “great practice, lovely staff” teaches a machine almost nothing. A review that says a nervous patient had two implants placed, the surgeon explained every step, and the final cost matched the quote teaches it exactly what you are good at, in the patient’s own words. When an assistant explains why it recommends a practice for implants, that reasoning is very often traceable to review text. Star count gets you considered. Review substance gets you named.
4. Crawlability: Can AI Systems Read Your Site at All?
A quietly technical one. Some practice websites block or break for the crawlers that feed AI systems, sometimes through security settings, sometimes because the site only renders properly for human browsers. The practice believes it has a website. The machine sees a locked door. This is checkable in an afternoon and fixable in most cases, but you have to know to look.
5. Corroboration: Does Anyone Else Vouch for You?
Machines weigh independent evidence more than self-description, exactly as a careful patient does. Professional register entries, directory listings, local press, association memberships, anything that confirms your claims from outside your own website strengthens the recommendation. A practice that only describes itself is asking to be taken on faith, and these systems are built not to.
Why This Is Probably Not You (Yet)
None of the above requires genius. It requires knowing the list exists and doing the unglamorous work, and here the industry has failed practice owners. Most dental marketing agencies are still selling the previous era: a templated website, some rankings, a monthly traffic report. Ask your current agency what ChatGPT says when a patient asks for the best implant dentist in your area. In my experience, most cannot answer, because they have never once checked.
Meanwhile the signals above sit unmanaged. The entity is inconsistent because nobody has audited the old listings. The schema is absent because nobody scoped it. The reviews are generic because nobody built a system for asking the right way. And so the AI answer in your area is being won, right now, by whichever competitor happens to be accidentally strongest, not deliberately best.
That accident is also your opportunity. In most local markets, no practice has done this work on purpose. The first one that does tends to become the default recommendation, and defaults, once machines settle on them, are stubborn.
What to Do, in Order
- Look at your own answer first. Open ChatGPT, Perplexity, and Google. Ask each one for the best practice for your highest-value treatment in your area, phrased the way a worried patient would phrase it. Ask each one about your practice by name. Twenty minutes, and you will know more about your real competitive position than any rankings report has ever told you.
- Fix the entity. Make your name, address, phone number, and details identical everywhere they appear, and correct or kill the old listings that contradict you.
- Give machines the structure. Implement schema for your practice, your clinicians, your treatments, and your FAQs, and validate it. For most sites this is a small, bounded technical job with outsized effect.
- Rebuild your review ask. Stop asking for stars and start inviting stories. Patients asked a specific, warm question at the right moment write reviews that mention treatments and outcomes, and that text becomes your case for recommendation.
- Publish content only you could publish. Honest, clinically accurate pages on your key treatments, attributed to a named, credentialed clinician. This is what health-cautious systems are built to reward, and it cannot be faked by a content mill.
An Honest Caveat Before You Spend Anything
No one can guarantee what an AI system will say, including me. These systems change, their answers vary between sessions, and anyone promising you a guaranteed spot in ChatGPT’s recommendations is selling something they do not control. What is real is the direction of the work: every signal above also strengthens your traditional search visibility and your credibility with human patients. Done honestly, this work has no losing outcome. It simply also happens to be how AI recommendations are won.
The Question to Sit With
Patients trusted directories, then they trusted search rankings, and practices learned to be visible in both. The recommendation layer is simply the next place trust has moved, except this time the answer is singular, the process is invisible, and the winner in each market is being decided now, mostly by accident.
You can find out where you stand tonight, for free, in twenty minutes, using the checks above. If you would rather have it done properly, our AI Audit checks five AI systems (ChatGPT, Claude, Gemini, Perplexity and Google AI Overviews) against your market’s real patient queries and walks you through a prioritized fix list in a recorded video. Either way, do the check. The patients you never hear from are already asking.