Healthcare Marketing

AI Patient Education Rules in Canada: What CPSO, CPSBC, and CPSA Actually Say About AI-Generated Content

15+ years in local marketing; Google Ads certified; Shopify Partner.

Most clinic owners I talk to have one of two reactions when I bring up AI-generated patient education content. Either they're already using ChatGPT to draft blog posts and FAQ pages and haven't given the compliance question a second thought, or they're terrified to touch it because they've heard "the College will come after you." Both reactions are wrong. The real ai patient education rules sit somewhere in the middle, and they're more workable than most doctors think, as long as you know what each provincial college actually prohibits versus what they just want you to document.

Here's the thing. No Canadian college has written a policy called "AI-Generated Content Rules" yet. What they have is existing advertising and professional conduct policy that applies to anything you publish, including AI-drafted blog posts, chatbot responses, and video scripts. The content doesn't get a free pass because a machine wrote the first draft.

This article is the compliance rulebook for practice owners using (or thinking about using) AI to draft patient education material. I'll walk you through what CPSO (Ontario), CPSBC (British Columbia), and CPSA (Alberta) actually say, what the practical rules look like when you sit down to publish something, and how to set up a physician review process that keeps you out of trouble. If you want the broader strategy of how content fits into your overall visibility strategy, our guide to medical SEO for Canadian practices covers the wider picture. This piece stays narrow on the compliance question.

One note before we start. I'm a marketer, not a lawyer or your regulator. The rules below come from the colleges' published policies. Your specific situation might have wrinkles I can't see. When in doubt, email your college advisor. Most will give you written guidance within a few weeks.

Quick math on the stakes. A single college complaint that triggers a formal investigation typically costs a Canadian physician 10-20 hours of legal-counsel time at $400/hr (so $4,000-$8,000), plus a public-record finding that, in our experience, depresses new-patient inquiries by 15-25% for 6-12 months. If you run 30 new patients/month at an average annual revenue of $1,800 per patient, a 20% drop for 9 months works out to roughly $81,000 in lost revenue, on top of the legal bill. That's the scale of risk you're managing when you decide whether to publish AI-drafted content unreviewed.

The Three Rules That Apply No Matter What Province You're In

Before you get into provincial differences, there are three rules every medical college in Canada enforces, and they apply whether a human wrote the content, an AI wrote it, or a committee of both wrote it together.

Rule one: no outcome guarantees. If your AI drafts a blog post that says "our treatment cures X" or "you'll see results in four weeks," you cannot publish it. CPSO's Advertising policy is explicit. All claims must be "verifiable and supported by available evidence and science." An AI model does not know what evidence supports what. It generates plausible-sounding sentences based on training data, and a lot of that training data was written by US marketers who play by different rules. I've seen AI-drafted physiotherapy blog posts that casually promise "full recovery" for conditions where full recovery isn't the clinical norm. That one sentence is a college complaint waiting to happen.

Rule two: no testimonials of specific outcomes. CPSO Policy Statement 2-17 prohibits testimonials on physician advertising. CPSBC has similar restrictions. If your AI tool is pulling from Google reviews or patient comments and weaving them into "patient stories" for your website, stop. The fact that the reviewer wrote it originally doesn't matter. You published it. You own the liability.

Rule three: no comparative claims against other practitioners. "Better than other clinics in Saskatoon" is a nonstarter. "The leading chiropractor in Regina" is a nonstarter. AI models love comparative superlatives because the training data is full of them. This is the single most common violation I catch when I audit AI-drafted content.

Those three rules are the floor. Get them wrong and any college in the country will have something to say.

What CPSO (Ontario) Says About AI-Drafted Content

CPSO has not issued a standalone AI policy as of 2026. What they have is the Advertising policy, which defines advertising broadly to include "print ads, newsletters, business cards, logos, signage, TV or radio ads, websites, blogs and social media posts, posters and billboards." AI-drafted blog posts fall squarely inside that definition.

The key CPSO requirements for any patient-facing content you publish:

All statements must be "accurate and factual"
Any statistical, scientific, or clinical claim must be "verifiable and supported by available evidence and science"
You cannot reference specific drugs, appliances, or equipment by brand name
You cannot make comparative claims against other physicians or health professionals
Before-and-after photos require the disclaimer "the outcome or results depicted are not guaranteed, and may vary between patients"

The practical translation for AI content: the physician named on the practice's licence is responsible for every word published under that practice's name. If the AI drafts something factually wrong, and you publish it, CPSO treats that the same as if you wrote it yourself. There is no "the AI did it" defence.

The operational fix is a documented review process. I'll walk through what that looks like below.

What CPSBC (British Columbia) Says

CPSBC Professional Standards require specific disclaimer language for paid advertising, and the bar is slightly higher than Ontario's on disclosing the nature of promotional content. If AI is drafting ad copy or paid social content for your practice, the disclaimers need to appear in the final published version, not just in a reviewer's notes.

Two practical points for BC practice owners:

First, CPSBC wants to see that you can distinguish clearly between clinical information and promotional claims. AI tools tend to blur these. A blog post titled "Three Signs You Need Physiotherapy" that ends with "book your appointment today" is a promotional piece dressed up as education. That's allowed, but it needs to be labelled and written to the promotional standard, not the educational one.

Second, BC has been active in enforcement around scope-of-practice language. If your AI drafts physiotherapy content that accidentally describes medical diagnosis or treatment language that sits outside a physiotherapist's scope, you've got a problem. The AI doesn't know where scope lines sit in BC. You have to catch it on review.

What CPSA (Alberta) Says

CPSA is the most permissive of the three on advertising overall. You have more latitude on what you can describe and how you can present services. But two rules still bite hard.

One, comparative claims between practitioners remain prohibited. You cannot position yourself as better than the clinic down the street, even if it's true. AI models produce this kind of copy constantly if you don't steer them away.

Two, CPSA expects accuracy. "More permissive" does not mean "anything goes." If an AI drafts a claim about wait times, patient volume, or clinical success rates that you can't back up with data, CPSA treats it the same as any other accuracy violation.

For chiropractors in Alberta, the CCA (Canadian Chiropractic Association) voluntary ethics layer sits on top of provincial college rules. The CCA prohibits claims of "curing" specific conditions. AI-drafted chiropractic content violates this constantly, because chiropractic content on the open internet is full of cure language written by US clinics under different rules.

The Physician Review Process That Actually Works

This is the section that matters most. Rules are useful. A workable process is what keeps you compliant when you're publishing 2 to 4 pieces of patient education content a month.

Here's the process I recommend to healthcare clients who want to use AI drafting without triggering college complaints. It's built around a week-by-week cadence because consistency is what catches errors, not heroic one-time reviews.

Week 1: Topic selection and prompt construction. The clinic's marketing lead (or agency) picks the topic. You write a prompt that names the province, the practitioner type, and the explicit prohibition list. Something like: "Draft a 700-word blog post for a family medicine clinic in Ontario. Do not make outcome guarantees. Do not include testimonials. Do not compare to other clinics. Do not reference brand-name drugs. Flag any claim that would require clinical evidence."

Week 2: AI draft and first-pass marketing review. The AI produces a draft. A non-clinical reviewer (marketing lead, practice manager) reads it against a compliance checklist. I'll share the checklist below. Obvious violations get flagged and rewritten before the physician ever sees it. This saves the physician's time.

Week 3: Physician review and clinical sign-off. The named physician reads the cleaned-up draft. Not skims. Reads. They check for: clinical accuracy, scope-of-practice issues, unsupported claims, anything that feels promotional when it should be educational. Changes go back to the writer. Sign-off gets logged. This is the documentation CPSO or CPSBC will ask for if a complaint ever lands.

Week 4: Publish and archive. The final version goes live. The original AI draft, the marketing-reviewed version, and the physician-signed-off version all get archived in a shared folder with dates. If the college ever asks, you have a paper trail showing human review, not "we let the machine publish whatever it wrote."

That's roughly 2 to 4 hours of physician time per piece, across a month. For a small practice publishing one post a month, that's manageable. For a larger clinic publishing weekly, you need to either reduce volume or build a tighter system with a clinical reviewer who isn't the senior partner.

The Compliance Checklist for Every AI Draft

Use this as your non-clinical first-pass review. If any answer is "yes," the draft needs rewriting before it goes to physician review.

Does it use the words "guarantee," "cure," "eliminate," "promise," or "ensure" in a clinical context?
Does it include or paraphrase a patient testimonial about a specific outcome?
Does it compare the practice (explicitly or implicitly) to other practitioners or clinics?
Does it name specific drug brands, device brands, or equipment brands?
Does it state a clinical statistic or success rate without a cited source?
Does it describe services or outcomes outside the practitioner's scope of practice?
Does it include before-and-after imagery without the required disclaimer (Ontario)?
Does it present promotional copy as if it were neutral educational content without any distinction?

Eight questions. Five minutes per draft. This checklist alone will catch 90% of the violations I see in AI-generated healthcare content.

A Worked Example: The Cost of Getting This Wrong

Let me show you the math on why the review process is worth the time. A chiropractor in Calgary I know (not naming, not their words, paraphrasing a common scenario) had an agency run a Google Ad that promised an outcome it could not back up. The College of Chiropractors of Alberta sent a formal letter. The practice spent roughly 3 months going back and forth to respond, revise copy, and resubmit evidence of a corrective process.

The opportunity cost of that 3 months is real. At CA$39.79 CPC for "chiropractic marketing" keyword space per DataForSEO, and assuming the practice was running a modest CA$2,000/month Google Ads budget, that's CA$6,000 of ad spend paused. Plus roughly 20 to 30 hours of the owner's time dealing with the college. Plus legal consultation if they opted for it.

Compare that to the cost of building a review process. Physician review at 2 to 4 hours per piece, at their hourly equivalent (pick your own number from your practice's revenue per hour), for 12 pieces a year, is maybe 36 to 48 hours annually. Far cheaper than the three-month fallout from one bad ad.

What To Do If You're Already Publishing AI Content Without a Review Process

Don't panic. Do an audit. Pull every piece of content on your website published in the last 18 months. Run each one through the 8-question checklist above. Anything that fails gets either rewritten or pulled down.

If you find something egregious (an outcome guarantee, a comparative claim, a testimonial), pull it down first and rewrite second. The college isn't going to ding you for pulling something down. They'll ding you for leaving it up once you know.

From there, build the week-by-week process. Assign roles. Start logging physician sign-offs. If an agency is producing your content, tell them about the checklist and ask them to produce drafts that pass it before the physician ever sees them. A decent agency should welcome this. A shitty one will push back and tell you you're being too cautious. That tells you what you need to know.

If you're evaluating whether AI drafting makes sense for your specialty, our subpage on AI tools for clinic intake and scheduling covers the operational side of AI in healthcare more broadly. For specialty-specific marketing rules, see our breakdowns for chiropractic marketing, physiotherapy marketing, veterinary clinic marketing, and optometrist marketing. If you're curious about how AI is changing how patients find clinics, our clinic AI visibility audit and piece on whether ChatGPT will replace family doctors get into that.

Three Takeaways

If you remember nothing else from this article, remember these three things.

One, the ai patient education rules aren't new rules. They're existing advertising and professional conduct rules applied to AI-drafted content. Every word you publish is your responsibility, regardless of who or what drafted it.

Two, the three universal prohibitions (no outcome guarantees, no testimonials of outcomes, no comparative claims) will catch most violations. Teach these to anyone who touches your content. Build them into your AI prompt.

Three, a documented physician review process is the single best defence if a college complaint ever lands. 2 to 4 hours of physician time per piece of content, logged and archived, is cheap insurance compared to a 3-month enforcement response.

AI drafting is a useful tool for healthcare content. I use it. A lot of good clinicians use it. The clinics getting in trouble aren't the ones using AI. They're the ones publishing AI drafts without review. That's the piece most practice owners miss. Fix the process and you can publish with confidence.