How to get your practice cited by ChatGPT and Google AI

Why does an AI name one practice and not another?

Answer engines name the practices they can identify, trust and quote.

Google’s own guide says its AI features retrieve from the normal search index and favour pages that answer a question directly. (Google Search Central, AI features and your website, updated 10 July 2026) The practices that appear are the ones whose name, address and services agree everywhere, whose pages say plainly who they treat and how to book, and who are mentioned by directories, peak bodies and referrers.

None of that is exotic. Most of it is unfinished admin.

What the engine does before it writes

When a patient types a question into ChatGPT or Google AI Mode, the engine runs several searches behind the scenes, reads the pages that come back, and writes an answer from the ones it can reconcile. It prefers pages that state a fact in one sentence, and it drops sources that disagree with each other. A practice with two spellings of its name across three directories is one of the sources it drops.

Why this favours smaller practices

Larger groups have more pages, more listings and more chances for those listings to drift. An owner-led practice with one address and six clinicians can make every listing agree in an afternoon. The engine does not care how many locations you have. It cares whether the description of you is the same everywhere it looks.

What can I check myself?

Five checks, in order, each taking under an hour.

1. Entity consistency. Search your practice name. Every listing that appears, from your site to the health directories, should show the same name, address, phone number and services. Fix the ones that differ.
2. Google Business Profile. Every category, service, hour and photo field filled, and the Q&A section answered by you. It feeds AI answers as well as Maps.
3. Structured Q&A pages. Each service page answers, in one paragraph each, the four to six questions patients ask: who it is for, what a first session involves, fees and rebates, wait time, how to book.
4. Third-party corroboration. Your peak body listing, at least two health directories and one referrer page that describe you the same way your site does.
5. Measurement. Write down ten prompts a patient would type, run them monthly in ChatGPT, Gemini and Google AI Mode, and record who is named.

The rest of this section takes each check in turn and says what done looks like.

Check 1: Entity consistency

An entity, in the engine’s terms, is one real-world thing with one name, one address and one set of attributes that every source agrees on. Your practice is an entity. So is each clinician who holds a registration.

Open a private browser window and search your practice name, then your own name with your profession. Write down every result on the first two pages: your site, Google Business Profile, Healthdirect, HotDoc, the Ahpra register, your peak body, the NDIS provider finder if you are registered, and any directory you have forgotten paying for. For each one, note the name, address, phone number, website link and the list of services.

Fix every difference at the source, starting with the listing that is wrong rather than the one that is right. Trading name and legal name can both appear, but the trading name has to be spelt one way. If you moved rooms two years ago, the old address is still out there. This is the check that takes the longest and moves the most.

Check 2: Google Business Profile

The profile feeds Maps, the local pack and the AI answers that draw on Maps data. Fill the primary category and every secondary category that applies, list each service with its own description, set hours that match your booking system, and add photographs of the rooms rather than stock images. Answer the questions in the Q&A section yourself, because anyone can, and a stranger will answer the one about parking or fees if you do not.

Post once a month. A profile last touched in 2024 reads as a practice that may have closed.

Check 3: Structured Q&A pages

Open each service page and read it as a patient who has never heard of you. Within the first screen it should answer who the service is for, what a first session involves, what it costs and which rebates apply, how long the wait is, and how to book. Each answer is one paragraph under a heading phrased as the question, in the page HTML, with no tab or accordion hiding it. Engines quote sentences. If the sentence is not on the page, there is nothing to quote.

Do not write a separate FAQ page and leave the service page thin. The answers belong on the page a patient lands on.

Check 4: Third-party corroboration

A claim that appears only on your own site is a claim. The same claim on your peak body’s directory, on two health directories and on a referrer’s page becomes a fact the engine will repeat. Ask the GP clinics that refer to you whether their website lists the allied health providers they work with, and whether your entry is current. Check that your peak body listing carries the same services your site does.

Reviews count as corroboration too, but only for what they say. Twenty reviews that mention the receptionist by name and none that mention the condition you treat tell the engine you are friendly and nothing else.

Check 5: Measurement

Write ten prompts that name a profession, a presentation or funding stream, and your suburb or region. Run each one in the three engines on the same day each month, signed out, and record every practice named. A spreadsheet with a row per prompt and a column per month is enough. You are looking for two things: whether you appear, and who appears instead.

None of that is exotic. Most of it is unfinished admin.

How long does it take to show up?

Long, specific prompts move within a quarter; broad prompts take longer because more practices compete.

Re-crawls of commercial pages happen within weeks. The first citations usually appear on prompts that name a condition, a funding stream and a suburb together. That is where a smaller practice wins, and it is why the prompt set in step five should be specific.

What moves first

The narrow prompts move first, because fewer practices are in contention and the engine has fewer sources to reconcile. A prompt that names a condition, a funding stream and a suburb often names three or four practices. A prompt that names only a profession and a state names the large groups and the directories. Aim at the first kind and let the second follow.

What a quarter looks like

Month one is the fixes in checks one and two and the first measurement. Month two is the service page rewrites and the first directory corrections landing. By month three the narrow prompts start to change and the broad ones have not. That is the expected shape. A practice that shows on a broad prompt in month one had most of the work done already.

What gets in the way?

The same four things, in most practices we look at.

Two names for the practice: the legal entity on invoices and the trading name on the door, both used online without the other. Pick the trading name for every public listing and keep the legal entity in the footer and the schema.

Answers hidden in accordions, tabs or downloadable PDFs. If a person has to click to read it, the engine may not read it at all.

Directories left to rot. Every listing you set up in your first year of practice is still out there, and the ones with the old phone number outvote the ones with the new one.

Chasing the broad prompt. The directories win a search for a psychologist near me. A practice page that says exactly that wins a prompt about a teenager with school refusal who can claim under a Medicare mental health treatment plan.

Can I pay to appear?

No. There is no paid placement in these answers.

Anything sold as ChatGPT ads for a practice is either display advertising elsewhere or nothing. Citations are earned, and they are checkable: run the prompt and read the answer.

What you can pay for is the work: the afternoon of listing fixes, the service page rewrites and the monthly measurement. Some owners do it themselves. Some hand it to us. Either way the engine sees the same thing, a practice described the same way everywhere it looks.

How do I know it is working?

The prompt spreadsheet tells you, and so do the enquiries.

The spreadsheet from check five is the first measure: the share of your ten prompts that name you, month by month. The second is the enquiry form. Add one question, how did you find us, with ChatGPT or another AI assistant as an option, and read the answers. The third is Search Console, where queries that read like full questions rather than two-word searches usually mean an answer engine sent someone to the page.

Report all three to yourself on the same day each month. If none of them has moved by month four, go back to check one, because something still disagrees.