When Patients Ask an AI Instead of Google, Will It Recommend You?
A growing share of patients now ask ChatGPT or a similar AI assistant “who is a good dermatologist near me” instead of typing it into Google.
That changes how a practice shows up online, because these assistants don’t rank pages the way a search engine does.
They fetch, read, and cite sources in a specific, mechanical way, and once you understand that process you can tell a wasted marketing budget from a real result.
How ChatGPT actually finds and uses sources
Independent research went past the answers and watched ChatGPT’s own network traffic while it worked. There’s a consistent structural pattern behind how it sources information from the web.
It fetches through scrapers, and they aren’t all equal. Every web result ChatGPT pulls in carries a tag saying where it came from.
One tier is a licensed allowlist of major publishers, big newspapers, wire services, encyclopedic sites, and that tier is closed to a local medical practice.
You’re not getting into it.
The tier that matters for you is the open commercial scraping layer, which handles local, shopping, and general business content. That’s the one you’re competing in, and it behaves a lot like ordinary web search.
Plenty of questions never trigger a web search. Before searching, the assistant sorts a question into a bucket. A quick lookup, a shopping question, a local question, and so on.
Some buckets never touch the live web at all and get answered purely from what the model already learned during training, so no page gets fetched and there’s nothing to win.
General how-to and definition questions usually land there.
The local, cost, insurance, and comparison questions patients ask right before they choose a provider are far likelier to trigger a real fetch, and that’s where your effort belongs.
Fetched, cited, and mentioned are three different outcomes. A page can get pulled into the assistant’s context (fetched) without ever becoming the clickable source behind a sentence (cited), and your brand name can turn up in an answer (mentioned) with neither of those happening.
You win or lose each one on its own. And citations attach at the sentence level, one sentence at a time.
Local results are capped tight. Ask a “best [specialty] near me” question and the assistant typically comes back with about two local businesses.
You could sit third in a five-listing Google Map Pack, the little block of local results on the map, and still be completely invisible in the AI answer.
That makes your review and profile work more important than it already was. If local visibility everywhere runs through roughly the same short list, the short list is worth fighting for.
You cannot cite yourself
This is the single most important thing to understand about AI search for a practice.
Your own website can be the source of your own facts, your hours, your services, your location, your accepted insurance.
But when the assistant makes an actual recommendation (“the best dermatologist in this city is…”), it typically cites a third party for that judgment. Nobody treats a business’s self-description as proof the business is good.
So the recommendation layer runs through the same places that already matter for the rest of your marketing: your Google reviews, listings on sites like Healthgrades, local press coverage, and mentions on forums like Reddit where real people talk about what happened to them.
Building those third-party signals is the same work as your review and local SEO effort, doing double duty.
What your site actually needs
- Plain, readable HTML text for your core facts. Services, pricing or cost information, accepted insurance, hours, and location all need to be readable straight from the page’s HTML, which rules out a JavaScript widget, a PDF, or a picture of a price list. An AI assistant reads the text of the page. If your pricing only appears after a script loads a table, the assistant may never see it, and when it can’t find your facts it’ll source them somewhere else, or leave you out.
- One clear, complete page per topic. Several thin pages saying the same thing work against you. When results collapse, near-duplicate pages on the same domain get treated as a single thin page. One well-written page beats five short ones.
- Text-based content over video. A Reddit thread gets cited because the actual text of the discussion is sitting right there on the page. A YouTube video usually returns only metadata to a fetch, so the transcript never comes along, and the video rarely gets cited even when it’s genuinely useful. If you have valuable video content, put a full written version of it on your site too.
- Content written for the question a person would actually ask, including a plausible version with your city or specialty in it, because the assistant often rewrites somebody’s messy phrasing into a cleaner search before it fetches anything.
What is a waste of money
Be skeptical of anyone selling a “ChatGPT ranking algorithm” service. There’s no public scoring system to game, and anyone promising to manipulate one is selling something that doesn’t exist in the way they describe.
Be careful too with large one-time investments in “AI-optimized content” that restate what your site already says in slightly different words.
If the underlying facts are already clean, readable HTML, the reworded version adds nothing, and it can actively hurt you if it spawns multiple thin pages that dilute your one good page on a topic.
The thing that actually works is dull. It’s accurate, plainly written facts on your own site, and genuine third-party proof (reviews, listings, press, forum mentions) that a real person thinks you’re good at what you do.
View your own page source
- Open your services and pricing pages and see whether the information is visible in the raw HTML or whether a script draws it. If you’re not sure, view the page source.
- Confirm your hours, location, and accepted insurance are stated in plain text somewhere on your site. A PDF doesn’t count.
- Search your own name plus “reviews” and see which third-party sites come up. Those are your current AI-recommendation sources, so make sure they’re accurate and current.
- Pick one high-intent question a patient would actually type (“[specialty] near [your city],” “does [your practice] take [insurance]”) and make sure your site answers it in one clear place.
Questions practices ask about this
How do I check whether an AI assistant can actually read the pages on my site?
Right-click the page, choose View Page Source, and search that raw code for a service name or your hours. If it's there, an assistant can read it. If it shows up only in the rendered page, a script is drawing it after load, and a fetch can come back with an empty shell. A PDF or a photo of a service list fails the same test.
Do I need to hire someone for this, or can my office handle it?
Much of it is in-house work. Writing your hours, services, and accepted insurance as plain text is a writing task, and asking satisfied patients for reviews is front-desk work. Bring in a specialist when the site draws content through scripts or keeps facts locked inside PDFs, because that means changing how the site is built, and it isn't something the front desk can do.
How long before any of this changes what an AI says about my practice?
There's no timeline you can count on, so plan in months rather than weeks. A page edit becomes readable once the page is crawled again, while reviews, listings, and press mentions build up slowly. Assistants also pull a fresh set of sources every time somebody asks, so the same question asked twice can come back with different names. Check monthly, watch the trend, and don't judge it on one answer.
Is this worth the effort for a solo or single-location practice?
Usually yes, because the local answer is short. When an assistant names only about two practices for a best-near-me question, a small practice with accurate pages and steady real reviews can hold its own against a larger group that buries its facts in scripts and PDFs. It's also the same work that feeds your ordinary Google visibility.
Photograph: Tim Samuel / Pexels