How to get your dental practice cited by ChatGPT and AI search

Why do ChatGPT recommendations matter for a dental practice?

"Best dentist for implants near me" used to be a Google search. It is now also a ChatGPT prompt, a question to Perplexity, and an AI Overview at the top of Google itself. The answer format changed with the habit: ten blue links became one paragraph that names one, two, maybe three practices. Everyone else is not ranked lower. Everyone else is absent.

The field is still nearly empty. SOCi's 2026 Local Visibility Index tracked more than 350,000 locations across 2,751 multi-location brands and found ChatGPT recommends 1.2% of them, while those same brands appear in Google's local 3-pack 35.9% of the time (vendor-reported, so treat it as directional). That sample is chains, which is worth saying out loud because the follow-up matters more to you than the headline: SOCi reports that "LLMs often favor single-location businesses", since a practice with one address, one phone number and one review history gives a model a much cleaner signal than a 200-branch group. Read the gap the right way: the map pack rewards thousands of businesses per city, the AI answer rewards a handful, almost nobody has claimed those seats yet, and your single location is an advantage in that race rather than a handicap.

The AI layer is not a fringe surface either. In our own tracking (July 2026), we ran a panel of 15 buying prompts across dental, gym and med spa searches through Google, three runs each. An AI Overview fired in 42 of 45 runs. The paragraph above the classic results is now the default first impression.

One honest caveat before the steps: for most practices, patient volume from AI answers is still small next to classic search and the map pack. The reason to move now is not today's volume. It is that these recommendations are scarce and winner-take-most, and the practice that becomes "the answer" in a city first is hard to displace later.

How does ChatGPT decide which dentists to recommend?

Two stages happen behind every recommendation, and they respond to different work. Call them retrieval and selection.

Retrieval is the search stage. When a question needs local or current facts, ChatGPT runs a real web search before it writes. OpenAI operates a dedicated crawler for this, OAI-SearchBot, which exists to "surface websites in search results in ChatGPT's search features". Retrieval decides which pages even reach the model's desk, and it behaves like classic search: if your site is not indexed, readable and relevant, you are not in the candidate pool. This is why SEO did not die; it became the entry ticket. ChatGPT is also not the only engine with its own reading habits: Microsoft's Copilot reads Bing's index, and Google's AI features read Google's, so index hygiene everywhere still buys you entry.

Selection is the writing stage. From the retrieved candidates, the model decides which names make the sentence. Language models put forward what they can corroborate. When three independent sources describe the same practice the same way (same name, same city, same treatments, solid reviews), naming it is easy. When your identity is fuzzy or contradicted across the web, the model quietly picks a competitor it can verify. Selection is a clarity and reputation problem, and it is the part "GEO" (generative engine optimization) actually covers.

The selection side responds to deliberate work. The academic paper that coined GEO (Aggarwal et al., KDD 2024) measured visibility gains of up to 40% in generative answers after adding quotable statements, citations and statistics to source pages. It is an academic benchmark on earlier GPT-3.5-class engines, so treat the number as directional, not a promise. The direction is the point: what the page says, and how quotable it is, changes whether the machine repeats it.

On the Google side, the company is refreshingly boring about it: there are "no additional requirements" and no special markup to appear in AI Overviews or AI Mode. Normal indexing is the foundation. The differentiator is everything stacked on top of it. Here is where each engine reads, and what that means for a practice:

Third column reflects our own run logs, July 2026. Engines change retrieval details without notice, which is one more reason to measure instead of assume (step 5).

Step 1: Make your practice machine-readable

Machines cross-reference before they recommend. The first job is making your practice one unambiguous entity everywhere it appears, because every variant of your name splits your evidence across two weaker identities.

• One exact NAP string. Name, address, phone, character for character, on your site footer, Google Business Profile, Yelp, Healthgrades, Zocdoc and insurance directories. "Smith Family Dental" on one and "Smith Dental Care" on another is two entities, each with half your reputation.
• Dentist schema on the site. JSON-LD with your name, address, geo coordinates, phone, opening hours, and sameAs links pointing to your profiles. Mark up each treatment page for the service it covers.
• A Business Profile filled to 100%. Primary category, every service listed, current hours including holidays, real photos, and the Q&A section seeded with the questions patients actually call about.
• Directories claimed and cleaned. The big dental and health directories, your ADA and state association listings, the chamber of commerce. Boring, load-bearing.

Honesty note on schema: it is not an AI cheat code, and Google says outright that no special structured data is required for its AI features. Its value is disambiguation. Every engine that reads you resolves to the same entity, which is exactly what selection rewards. This is week-one work with a high floor and zero downside.

Step 2: Publish treatment pages an AI can quote

When an engine cites a dental website, it lifts passages, not vibes. Build one page per money treatment: dental implants, clear aligners (Invisalign), emergency dentistry. Each page answers the money question in its first 60 to 80 words, then earns its depth. A quotable treatment page has:

• The direct answer up top: what it costs in your city as an honest range, how long it takes, who is a candidate.
• Question-shaped headings that match how patients ask: "Does getting an implant hurt?", "How much is Invisalign per month?", "Can I be seen today?"
• Short, self-contained paragraphs a machine can lift whole without losing meaning.
• A visible date, a named author (the dentist, with credentials), and sources for clinical claims.
• A short FAQ that mirrors what patients ask at the front desk, word for word.

This maps directly to the GEO study finding above: quotable statements, citations and statistics measurably raise generative visibility. It also converts human readers better, which is not a coincidence. And delete the padding while you are in there: 2,000-word "why choose us" essays and thin service blurbs give an engine nothing to quote.

Step 3: Get mentioned in the places AI reads

Selection leans on corroboration, and corroboration lives off your website. Three buckets, in rising order of effort:

• Reviews. Volume and recency matter, but so do words. A review that says "implant" and names your neighborhood gives an engine parsing text far more to work with than a bare five stars. Reply to every review; it is more consistent text tying your name to treatments and place.
• Directories and local press. Stable third-party pages are what engines lean on hardest. Reference-grade sites like Wikipedia turn up constantly in AI answers, and your practice will never get a Wikipedia page. That is not the lesson. The lesson is that reference-grade third-party pages outweigh your own homepage, so collect the ones available to you: directories, local news, community sponsorships that get written up.
• Community, above all Reddit. In our own tracking (July 2026), Reddit threads were cited in 27 of 45 Google AI Overview runs on our buying-prompt panel. Threads like "dentist recommendations?" on your city's subreddit feed AI answers directly.

On Reddit, do not fake it. Planted posts and sockpuppet praise are easy to spot, against the platform's rules, and a terrible look for a medical practice. What works and stays clean: know the recurring threads where your city asks for dentists, correct factual errors politely and transparently as the practice, and give patients experiences worth writing about. This bucket compounds slowly and cannot be rushed, which is exactly why it is the strongest signal of the three.

One more pattern from our runs: Perplexity's citations on buying prompts are dominated by roundup listicles. In our own niche, agency "best of" lists fill nearly every answer. The dentist equivalent is the "best dentist in [city]" roundup from a local magazine or city blog. Getting onto the legitimate ones is unglamorous outreach, and it pays across every engine at once.

Step 4: Open the door to AI crawlers

None of the above matters if the machines cannot read you. Thirty minutes with whoever manages your website covers the three checks:

• robots.txt. Allow OAI-SearchBot. OpenAI is explicit that sites opted out of it "will not be shown in ChatGPT search answers". GPTBot is a different bot with a different job (training data for future models); blocking it does not remove you from ChatGPT search, so decide each one deliberately instead of blanket-blocking anything with "GPT" in the name.
• No JavaScript walls. If treatment content only appears after client-side rendering, or sits behind popups and interstitials, a crawler may see an empty shell. Some AI crawlers execute little or no JavaScript, so server-rendered HTML is the safe default for anything you want quoted. Quick test: load your implant page with JavaScript disabled and see what survives.
• Check you have not opted out by accident. Google's snippet controls (nosnippet, max-snippet, noindex) also limit what its AI features can show. Old SEO plugins set these silently.

There is also llms.txt, a proposed standard (Jeremy Howard, September 2024) for a markdown map of your site written for language models. Honest status: no major engine has confirmed reading it. It takes 20 minutes, so we ship one on our own site anyway. File it under cheap hedge, not lever.

Step 5: Measure it honestly

The failure mode of this whole field is measuring nothing, or measuring once. AI answers are probabilistic: the same prompt on the same day can name different practices from run to run, and answers drift as the engines update. Citation drift is real, so the measurement has to absorb it:

1. Freeze a prompt panel. Ten to fifteen prompts real patients use: "best dentist for implants in [city]", "emergency dentist open now in [city]", "how much is Invisalign in [city]". Write them down on day one and never edit them mid-flight. Editing the panel to flatter the results is the oldest trick in reporting.
2. Run every prompt multiple times. At least three runs per prompt, per engine, per checkpoint. Our own July 2026 panel is 90 runs for exactly this reason. Single runs lie: one prompt in our panel never triggered a Google AI Overview in any of its three runs, while the panel overall triggered in 42 of 45.
3. Track share of voice, not screenshots. In what fraction of runs are you named? Who else is named? Which sources get cited? That last list is next month's outreach plan.
4. Re-run monthly against the frozen baseline, and report the misses alongside the wins.

If an agency shows you one screenshot of ChatGPT naming their client, ask how many runs it took to get it. The same skepticism drives our comparison of dental SEO companies: every claim verified and dated, self-reported numbers labeled.

How long does it take, and what does it cost?

The honest timeline: foundations (schema, NAP, Business Profile, treatment pages, crawler access) go live inside the first month. Mentions and reviews accumulate over 30 to 90 days, and engines refresh what they read on their own schedule. Day 90, measured against the baseline you froze on day one, is the honest first read point. Anyone promising a specific AI answer by a specific date is selling you weather.

Can you do it yourself? Steps 1, 2 and 4, yes: a diligent owner or office manager working with a web person can run them from this page. Step 3 is a grind that rewards consistency, and step 5 is tedious without tooling, but none of it is secret. This guide is the playbook. Go run it.

If you would rather delegate it, this is what we do for dental practices. The deep audit is $1,500 and credits 100% toward your first month. Retainers run Foundation at $1,500/month, Growth at $2,500/month and Dominate at $5,000/month, all detailed on the pricing page. Those numbers sit mid-market on purpose; the full market ranges, with sources you can check, are in our guide to what dental SEO costs in 2026. On Growth and up, the measurement carries the risk: cited in 90 days against your frozen prompt panel, or we keep working free. And we take one practice per city, because an AI answer only names one or two. Still weighing a specialist against the SEO firm you already pay? GEO agency vs SEO agency breaks down what actually differs.