What AI Can't Add to Your Practice Pages
Every physician with a practice website is getting pitched some version of the same offer right now. Unlimited content, generated by AI, for a fraction of what a writer costs.
Fifty service pages by Friday. A blog post every week, forever, and you don’t lift a finger.
And the pitch isn’t entirely wrong, which is exactly what makes it worth thinking through.
AI drafting tools are genuinely good at structure, at plain-language explanation, at volume, and used well they’re a legitimate part of how modern websites get built, including ours.
The real question is narrower than whether AI should be involved at all. It’s who the author is.
On a medical practice website the author has to be you, and the practices getting that wrong are building a specific, measurable problem for themselves.
What actually happens when a practice publishes generated content at volume
Google has said plainly that it doesn’t penalize AI-assisted content as such. What it targets is scaled content produced mainly to fill search results rather than to help a reader, and it’s gotten aggressive about that.
Recent updates have pulled large amounts of exactly the fifty-pages-by-Friday material out of search results.
The pattern that gets caught is easy to spot once you’ve seen it a few times. Every page reads the same. Every condition gets the same eight-paragraph structure, the same careful generalities, the same absence of anything a real clinician would say out loud.
The content is rarely wrong. It’s also interchangeable, and interchangeable is what the ranking systems are being tuned to discard, because interchangeable is useless to the patient reading it.
Google’s quality standards for health content specifically look for evidence of first-hand experience, the small details that show the author does this work.
That’s the ingredient a language model can’t supply. It has never sat in a room with one of your patients.
A generated page is a floor-model answer, and your competitors can all buy the same one.
The signature problem
There’s a second issue here, and it lands harder in medicine than in any other business. Your service pages carry your name, or they should, since unsigned clinical content is its own problem.
A patient reads those pages as you speaking. So does Google. So do the AI assistants that increasingly read practice websites and then repeat what they find to a patient asking for a recommendation.
If boilerplate is what stands behind your name, that’s the impression every one of those readers walks away with, and it’s what gets repeated back when an assistant summarizes who you are.
You already carry this instinct from the rest of your professional life. You don’t sign notes you haven’t read.
The same discipline is enough here.
AI can produce the draft, you decide what goes out under your signature, and you add the things only you can add.
The version of this that actually works
So, the division of labor.
Let the drafting tools do what they’re good at: organizing a page, keeping the reading level right for patients, covering the questions people actually search for, producing a competent first pass at volume.
Then you do the part that can’t be delegated, and it’s smaller than you fear. A review pass for accuracy, plus the two or three additions per page that make the page yours.
What patients consistently misunderstand about this procedure. What you tell people in the room when they ask the question the page is answering. Why your approach differs from what they might have read elsewhere.
Twenty minutes per important page, once.
Those additions aren’t decoration. They’re the material Google’s experience standard rewards, the material patients recognize as a real person talking, and the material no competitor can generate, because it comes out of your Tuesday afternoons.
Start with your three busiest pages
Pull up the three service pages that bring you the most patients and read each one the way a patient would.
Then add the parts only you can add. What people consistently misunderstand about the procedure. What you tell them in the room when they ask. Why your approach differs from what they’ve read somewhere else.
Three pages in one sitting is enough to tell you whether the rest of the site is worth the same treatment.
Questions practices ask about this
How can I tell if my service pages are the generic kind?
Read three service pages and ask whether a competitor across town could swap in their name and publish them unchanged. If they could, the pages are interchangeable. Two other checks help. Look in Search Console for service pages whose clicks slid after a core update while the rest of the site held steady, and look for any sentence that names something specific to how you practice.
Can my office manager or a writer add this, or does it have to be me?
The specifics have to come from you, but the typing doesn't. One workable pattern is a five minute voice memo per page, answering what patients get wrong about the procedure and what you tell them in the room, handed to staff or a writer to shape. You read the result before it goes up, the same way you read a note before signing it.
What should I ask a content vendor before signing anything?
Ask whether they interview you, and who does the interviewing. The scope should name a call with the physician, a draft you sign off on before anything publishes, and a page count per month rather than a headline volume number. Ask to see two pages they wrote for a practice like yours, and read them for anything only that doctor could have said. If they can't show that, keep looking.
What happens if I just leave the generated pages up?
Usually nothing dramatic on day one. The slower cost is that pages a patient can't tell apart from a competitor's tend to lose ground as ranking systems keep tuning against interchangeable content, and they are also what an AI assistant repeats when someone asks about you. If you do rewrite, recrawling and re-evaluation run on Google's schedule, and any movement is possible rather than promised.
Photograph: cottonbro studio / Pexels