The Questions Patients Ask AI Before They Ask a Clinic

Patients asking AI health questions are now normal, so many of your patients ask a chatbot before they call you. About 1 in 10 Australian adults asked ChatGPT a health question in the six months to June 2024. OpenAI says more than 300 million people a week now ask it about health. The questions are practical: what a symptom means, whether to see someone, and what it will cost. I read 59 Australian clinic websites to see whether they answer five of those questions. Only 1 site answered all five, and 10 answered none.

How many patients ask AI before they ask a clinic?

The Australian number comes from the Medical Journal of Australia. In a survey of 2,034 adults that mirrors the whole country, Ayre, Cvejic and McCaffery found that 9.9% had used ChatGPT for health information in the six months before June 2024. They put that at about 1.9 million people.

That was two years ago, and it covered one tool. The same paper looked at people who knew about ChatGPT but had not used it for health. Of those, 38.8% were thinking about doing so within six months.

The global number is bigger. In January 2026, OpenAI said 230 million people asked ChatGPT health and wellness questions each week. By July 2026 it said more than 300 million. Those are OpenAI's own figures, from its own study of chats. So you should read them as a claim, without the weight of an independent check.

In the United States, a KFF poll of 1,343 adults in early 2026 found 32% had turned to an AI chatbot for health information in the past year. That is the same share who use social media for health. Deloitte's 2024 survey of more than 2,000 US adults put use of AI tools for health reasons at 37%. The main reason people gave for not using it was that they did not trust the information (30%).

So the patient who walks into your clinic has often already talked their problem through with a chatbot. What matters for you is what they asked. It also matters whether your website helped the chatbot answer.

What do patients actually ask an AI about their health?

The MJA paper asked its 187 ChatGPT users what they used it for. Here is the list, in their order. As you read it, notice which questions your clinic could answer.

Source. Ayre, Cvejic and McCaffery, Medical Journal of Australia, 2025, Box 2. People could give more than one answer. The authors class the symptom, what-to-do, see-a-doctor and results questions as higher risk. Of all users, 61% had asked at least one of them.

The KFF poll asked people why they used AI instead of a person. Two thirds (65%) said they wanted quick or immediate information. But 41% said they wanted to look things up before deciding whether to see someone like you. About 1 in 5 said they could not afford to see a health professional (19%). A similar share had no regular doctor or could not get an appointment (18%).

OpenAI's own July 2026 post lists the common uses in one line: "understanding a lab result and preparing for an appointment to making sense of what a doctor said and building a healthier routine."

Put those together and you can see a pattern. Patients ask AI two kinds of question. The first kind is clinical, such as what this means and whether it is serious. The second kind is practical, such as whether they need to see someone, who, what it will cost and how soon. Your clinic website can answer the second kind.

Which of those questions can a clinic page answer?

Your page should leave the clinical questions alone. Say your service page tells a reader what their symptom means, or whether they should book. Then you are giving personal advice to someone you have never assessed. Section 133 of the National Law, the part that covers health ads, bans ads that create an unreasonable expectation of beneficial treatment. In plain words, you cannot lead people to expect a better result than is fair. That is a bad place to start.

The practical questions are different. They are about your clinic, they are factual, and you already know the answers. I picked five, each one drawn from the surveys above.

1. How much will it cost? Cost was a major reason for using AI for 19% of KFF respondents.
2. Do I need a referral? The MJA's "should I see a doctor" question, in its Australian form.
3. What happens at the first visit? OpenAI's "preparing for an appointment".
4. How soon can I be seen? Access was a major reason for 18% of KFF respondents.
5. Is it covered? Medicare, a rebate, bulk billing, a health fund, DVA, NDIS or WorkCover.

An AI tool can only repeat what your site says. If your fees page says "call us for pricing", the tool has nothing to quote from you. It will quote a competitor who published a number instead. So I went and looked at who publishes what.

What did I find on 59 Australian clinic websites?

I searched for a clinic type plus a city, across Sydney, Melbourne, Brisbane, Perth, Adelaide, Canberra, Hobart and the Gold Coast. The types were physio, dentist, GP or medical centre, psychologist and cosmetic clinic. I put no question words into any search. That way the sample does not lean toward clinics that already answer them. From each site I read the page the search led to. I also read up to five linked pages about fees, FAQs, new patients, referrals, Medicare or booking.

That gave 59 clinic sites, 275 pages and 151,560 words. Here is how many sites answered each question, after I read every hit by hand.

Nearly every site says something about Medicare or health funds. Only about a quarter say what happens at the first visit, or how soon a new patient can be seen. Those are 2 questions your patients ask a lot.

When you look site by site, the picture is even starker.

• Only 1 site answered all five questions. It was a GP clinic.
• Another 8 sites answered four.
• More than half, 33 sites, answered two or fewer.
• On any page I read, 10 sites answered none of them.

The average site answered 2.14 of the five. Picture a patient who asked an AI the practical questions first. On most of these sites, they would still have to call you to find out.

Which kind of clinic answers the most, and which the least?

The five clinic types behave very differently. The gaps say something about how each kind of clinic runs. Find your own type in the table below.

Psychology sites answer the most, and Medicare is the reason. The Better Access rebate needs a GP referral and a Mental Health Treatment Plan. So every psychology site has to explain referrals and rebates, and most give the session fee beside them. In all, 9 of 11 answered the referral question.

Dental sites answer cost. Of the 11, 6 publish a new-patient check-up price, usually as an offer. Not one of the 11 says whether you need a referral, because in general dentistry you almost never do. The dentist knows that. The patient asking a chatbot does not.

GP clinics publish fee tables (13 of 16). But they say almost nothing about the first visit (2 of 16) or referrals (1 of 16). The one site that answered all five was a GP clinic, so you can do it too.

Cosmetic clinics answer almost nothing. Of the 10, 6 answered no question at all, and none said how soon a patient could be seen. The pages I read were about treatments and the people who perform them. That is a choice. It hands every practical question to whoever the patient asks next.

Why does a table cell not count as an answer?

I have a confession about my method, because it turned into a finding you can use.

My first scanner split each page into sentences and dropped anything under 25 characters. It reported 13 sites that described the first visit and 35 with cost wording. Then I read the raw pages. Seven GP clinics and two physios had full fee tables the scanner had missed. The prices sat in table cells, with no sentence around them. One cell said "Standard Consultation (Level B)", and the next said "$96.20".

I fixed the scanner and ran it again. You cannot fix a chatbot. A price in a bare cell answers a person looking at the page. It is a weaker answer for any tool that reads text, because nothing says what the number is a price for. The clinics that were easy to count were the ones that wrote a sentence: "A standard consultation is 50 minutes and costs $265."

The same goes for headings. A "Referrals" heading with a form under it does not answer "do I need a referral". In all, 14 sites had a short referral heading of some kind. Of those, 5 put nothing under it that said yes or no, so I did not count them.

I have written before about what Google's AI tells patients before they click and which AI engines actually name Australian clinics. The lesson from both was that AI tools quote sentences. This study shows the other side of the same story, which is what patients ask. You know the questions. The sentence that answers each one is short. Most sites have not written it yet.

How do you answer the five questions without breaching the rules?

Answering a practical question is low risk under the AHPRA guidelines, the rules for health ads. The risk arrives when your answer is absolute, with no room for exceptions. Here are the patterns from the sample, with a line you could publish instead. Where a "use this" line is quoted, it comes from a site in the sample, which I have not named.

Every "use this" line stays factual. None promises a result. None uses a patient's words about their care, which the testimonial rule would catch. Rebate claims were the one place the sample did well. Every "rebates available" or "no gap" headline I found had an eligibility note somewhere on the same site. The wait-time claims did not, so check yours.

Maybe the pages that should hold these answers do not exist on your site yet. If so, the About page and the fees page are the two most patients read before they book. I have covered both for you.

How did I measure this, and what can it not tell you?

The method is short enough for you to repeat.

I fetched the pages on 20 and 21 September 2026. I found 74 clinic sites, but 15 would not load, which left 59. I stripped out menus, footers and scripts. If a sentence was repeated across pages, I counted it once per site. That left 10,862 sentences and 151,560 words.

A scanner flagged sentences against five fixed patterns. It found cost wording on 35 sites and referral wording on 24. It found first-visit wording on 13, wait wording on 32 and coverage wording on 46. A second pass caught short lines and table cells the first had dropped.

Then I read every hit and applied the same rule to each site. I did not count a Medicare rebate amount as a fee. I did not count a cancellation fee either. A sentence about the GP writing a referral does not say whether the patient needs one. A results-within-48-hours line does not tell you the waiting time. That read gave the published numbers: 32, 19, 15, 15 and 45.

Method. 59 Australian clinic websites, 275 pages, fetched 20 and 21 September 2026 from neutral city searches in five clinic types. Survey figures are from the Medical Journal of Australia (2025), KFF (March 2026), OpenAI (January and July 2026) and Deloitte (2024). The scripts, the site list and every verdict are saved with the post.

Now here are the limits you should know about.

• I read up to six pages a site. A fees page I did not open is not in the count. The true figures are likely higher.
• A scanner cannot read prices inside images, PDFs or booking widgets.
• The five questions are my choice, drawn from the surveys. A different five would give different numbers.
• The KFF and Deloitte figures are American. The MJA figure is Australian but from mid-2024.
• This is general information about the ad rules. It is not legal advice about your practice.

What should you check on your own site this week?

Open your own site and try to answer the five questions as if you were a stranger.

1. Cost. Find your standard fee in dollars, written in a sentence. If it only sits in a table, add the sentence.
2. Referral. Find the word "referral". If the sentence around it does not say whether your patient needs one, write that sentence.
3. First visit. Find "first appointment" or "initial consultation". It should be followed by how long, and what happens.
4. Wait. Find "same day". If it is there, check it has a qualifier and is true. If it is not there, say how soon you usually see new patients.
5. Covered. Find "Medicare". Check that you explain who is eligible right beside it.

That is five sentences in all. More than half of the sites in this sample are missing at least three of them.

You may want your whole site read this way, with every flagged line quoted and the new line written. That is what the All Clear Audit does. If you would rather start with the page patients read first, website copy is where I begin. For the rulebook behind the rewrite table, start with the AHPRA advertising guidelines explained.

If you prefer the short version, the findings are in an eight-slide web story.

Related reading:

• What patients Google before booking a clinic
• What Google's AI tells patients before they click
• Which AI engines name Australian clinics
• What patients read on a clinic's About page
• How to rank in AI search
• SEO and AI search for clinics
• View all services →