“We’ve been using ChatGPT at the front desk for a few months. It’s fine for emails. Where do we start if we want to use it for marketing, and what rules do we set first?”
Before anyone on your staff uses an artificial intelligence chatbot such as ChatGPT for marketing, the practice needs written rules about what staff can enter into the chatbot, where the facts in each draft come from, and who checks each draft before it is published. Without those rules, errors can reach the website before anyone notices them. A service page may list a treatment the practice no longer offers, a blog post may cite a study that does not exist, or a reply to an online review may mention a patient’s visit.
This article explains how to set up a process for producing marketing content with AI at your practice. It covers what AI models do well and poorly, the rules under the Health Insurance Portability and Accountability Act (HIPAA) to put in place first, the reference documents the AI model should work from, where to find topics, how to review drafts, and how to measure results. Most sections end with one task you can complete this week.
What AI does well and poorly in practice marketing
AI chatbots such as ChatGPT, Claude, and Gemini run on large language models (LLMs), which are AI models trained to generate written text. These models are useful for drafting and research, but they should not be treated as a source of facts about your practice or as the final reviewer of anything you publish.
An AI model saves the most time when it writes the first draft of a blog post or service page. It can also reformat approved content, for example by turning one article into a set of frequently asked questions, several social media posts, and an email. When you provide a document such as a clinical guideline, the model can summarize it in plain language.
AI models also have limits that matter for a medical practice. A model does not know the facts about your practice, such as your hours, locations, providers, or accepted insurance plans, and it fills in missing details with confident guesses. Unless the model searches the web, it may not reflect regulations that changed after its training data was collected. A model can also describe clinical details incorrectly in ways that a reader without medical training will not notice, and it can cite studies that do not exist.
A 2026 study published in The Annals of the Royal College of Surgeons of England measured how reliable chatbot citations are. Researchers reviewed 1,249 references from 108 answers to common surgical questions, produced by chatbots including ChatGPT, Google Gemini, Grok, and DeepSeek. In the worst-performing models, 34 percent of the references were fabricated or could not be verified. Several other models produced no fabricated references, and the advanced reasoning versions produced better references than the standard versions. Because accuracy varies this much between models and versions, a person at your practice has to check every citation in an AI draft. Use the AI model only to produce drafts, and have staff supply the facts and review every draft before it is published.
This week: write down the tasks your practice will use AI for and the information AI is not allowed to be the source of. Post the list where the front desk can see it.
Set your HIPAA rules before anyone opens an AI chatbot
Consumer versions of the major AI chatbots, including free and paid individual plans, do not include a business associate agreement (BAA). A BAA is the contract HIPAA requires before a vendor can handle protected health information (PHI) on behalf of a medical practice. If a staff member enters information that identifies a patient, together with details about that patient’s care, into a chatbot without a BAA, the practice is responsible for the disclosure. In marketing work, this most often happens when a staff member drafts a reply to a patient’s online review, writes a patient story for social media, or asks the chatbot to summarize call notes. The second article in this series explains what staff should not enter into ChatGPT and how to write a one-page AI use policy. Read it before your staff use AI for any task that involves a patient.
This week: tell your staff, in writing, that no patient name, visit detail, photo, or message goes into an AI chatbot until the policy is done.
Build your knowledgebase
A knowledgebase is a collection of verified documents about your practice that the AI model uses as its source material. You can keep the documents in a shared folder and upload them when you start a draft, or, if your chatbot supports projects, store them inside a project. When the AI model drafts from these documents, it uses your practice’s actual services, providers, and wording. When the model has nothing to work from, it produces generic healthcare copy and fills in details it does not know.
The knowledgebase should include the following documents:
- Your current website copy, page by page, so drafts match what your site already says.
- Provider biographies and credentials, written exactly as they should appear in print.
- A list of the services offered at each location.
- The insurance plans you accept, listed by location if they differ.
- Your brand voice rules, including the words you use, the words you avoid, and how formal your writing should be.
- Your compliance rules, such as no guaranteed outcomes, no before-and-after claims, and any restrictions from your specialty board or your attorney.
- The questions patients ask your front desk most often, since these are the topics most worth writing about.
Patient information, including names, visit details, photos, and messages, does not belong in the knowledgebase, because the knowledgebase describes the practice and not its patients. Assign one staff member to maintain the knowledgebase and review it every quarter. The AI model cannot tell when a document is out of date, so an old provider list or a discontinued service will appear in every draft that uses it.
For example, a multi-location practice might build its knowledgebase from its website after a site rebuild. If an old service page survived the rebuild and lists a procedure at a location that no longer offers it, every draft built from that page will repeat the error until someone corrects the file. A quarterly review catches errors like this before they reach a published page. This same set of verified documents is the foundation of our strategy and foundation work with every practice, because every later piece of content depends on it.
This week: create the folder and drop in the first two items: your current website copy and your provider bios. The rest can come next week.
Find the topics your patients care about
The best topics for practice marketing are the questions patients already ask. Your front desk staff hear these questions every day, so ask each person to write down the five questions they answer most often by phone and in person. Your online reviews, along with the reviews of similar practices, show what patients praise and what they complain about. Google Search Console, Google’s free tool for website owners, shows the search queries your site already appears for. Together, these sources give you a topic list for the next several months.
For clinical facts, use primary sources such as the relevant medical society, the Centers for Disease Control and Prevention, the U.S. Food and Drug Administration, or the original peer-reviewed study, and record the link and the date you read it. An AI model can summarize a source you provide, but it should not be the source itself, because you cannot see where its information came from or whether that information has since been revised.
This week: collect the front desk’s five questions each. That’s your first topic list.
Review every draft before it is published
Every AI-assisted draft needs two reviews before it is published. A licensed provider checks every clinical statement, and a staff member checks every fact about the practice and every compliance rule. The published page then includes a line that names the reviewer, lists their credentials, and gives the review date. The third article in this series explains the two-reviewer process, Google’s guidance on AI-generated content, and a checklist to use before publishing.
This week: name the provider who’ll do clinical review and the staff member who’ll do the practice review. Ask both if they have an hour a month.
Set up a content development workflow
The workflow below lists each step and the staff member responsible for it. It is the same sequence we follow in our medical content engine, and later articles in this series add steps for social media and patient reviews.
- Choose a topic and search term. Assigned to: marketing lead or office manager. Choose one topic from your research list and write one page for it. Do not create separate pages for small variations of the same question. Google’s spam policy on scaled content abuse covers large numbers of pages created without adding value for readers, whether a person or an AI model wrote them.
- Write a brief. Assigned to: marketing lead. The brief is a half-page plan that states the patient’s question, answers it in two sentences, and lists the sources to use, the services and locations to mention, and the call to action. Most decisions about the content are made in the brief, which keeps the drafting step simple.
- Draft with AI from the knowledgebase. Assigned to: marketing lead. Give the AI model the brief, the knowledgebase, and the sources, and instruct it to mark any claim it cannot trace to a source you provided.
- Edit the draft. Assigned to: marketing lead or office manager. Remove wording that sounds machine-written, check the voice against your brand rules, and cut anything the brief did not ask for.
- Clinical review. Assigned to: a licensed provider. The provider checks every clinical statement, procedure description, and outcome claim.
- Compliance check. Assigned to: office manager or practice administrator. This person confirms practice facts, services by location, insurance plans, and provider credentials, and checks that the page contains no outcome guarantees and no patient information.
- Publish with a reviewed-by line. Assigned to: marketing lead. The page lists the reviewer’s name, credentials, and the review date.
- Log the approval. Assigned to: office manager. Record each page, its review date, the reviewer, and any changes in a shared spreadsheet, so the practice has a record if the page is questioned later.
Eight steps may seem like a lot for one blog post, but most of them are short. Steps 1 through 4 take about an afternoon, steps 5 and 6 take about twenty minutes each, and steps 7 and 8 take about five minutes. Assigning each step to a named person ensures that no page is published on the assumption that someone else checked it.
This week: put the eight steps and the person assigned to each in a document and send it to the people named.
How this process helps your practice appear in AI search results
Google now answers some searches with an AI Overview, which is an AI-generated summary at the top of the search results, or in AI Mode, which is Google’s conversational search experience. Both link to a small number of source pages. Answer engine optimization is the work of making your pages the sources these AI answers use, and it builds on the same SEO and local search visibility work that gets your pages indexed in the first place.
Google’s guide to its AI search features states that a page must be indexed by Google and eligible to appear in search results with a text snippet before it can be shown in these features. The guide also recommends what it calls non-commodity content, which it describes as content that “provides unique expert or experienced takes that go beyond common knowledge and the ordinary.” By following the process in this article, your practice can produce this kind of content. The knowledgebase keeps your practice facts accurate, the provider review adds medical expertise, and the front desk’s question list makes sure each page answers what your patients actually ask. The sixth article in this series explains how patients find practices when they do not click through to a website.
Measure the results
Measure results by the actions patients take rather than by page views. Each month, track phone calls, appointment requests, and actions on your Google Business Profile, which include calls, requests for directions, and clicks to your website from the profile.
To see how your content performs in Google’s AI features, use the Generative AI performance report in Google Search Console. The report shows how often your pages appear in AI Overviews and AI Mode, where patients may read your information without visiting your website. Google began releasing this report to a subset of websites in June 2026, so it may not be available in your account yet. Google’s AI optimization guide, updated July 10, 2026, directs site owners to this report.
Record your baseline numbers before you publish the first AI-assisted page. Without a baseline, you cannot tell whether the new content changed anything. Results will depend on the topics you choose, the competition in your area, and how consistently your staff follows the workflow. Our article on measuring visibility by patient actions instead of clicks explains the full setup.
This week: write down this month’s call count, appointment requests, and Google Business Profile actions. That’s your baseline.
The mistake to avoid
The most common mistake we see is using AI to publish more content without changing how that content is researched, written, and reviewed. For example, a practice that published one blog post per quarter starts publishing four posts per month, each drafted from a one-line prompt, with no provider review and no knowledgebase. These posts repeat what other practices’ websites already say, which Google’s guide calls commodity content, so traffic is unlikely to improve. The posts can also contain errors, and the practice may not learn about an error until a patient asks the front desk about a service that a post described incorrectly.
Instead of increasing the sheer volume of pages, the goal is to output pages that are more useful and relevant to patients. One reviewed page that answers a question your patients ask does more for patients and for your search visibility than many generic pages. Start with the first “This week” task in this article, which takes about ten minutes, and complete the rest in order.
Net One Click works exclusively with medical practices. If you’d like to talk about how we can help build and grow your practice, book a free consultation.
