Nobody searches for what you actually do
Over five months we tracked every search that reached a Houston vascular practice through Google Ads. The practice performs genicular artery embolization, an image guided procedure that treats knee arthritis pain without surgery.
Every search term containing “genicular artery embolization” or “GAE”, across the entire period, produced 21 clicks and zero booked consultations.
Not because the procedure is unpopular. Because almost nobody knows it exists. The people who need it are searching “knee pain treatment without surgery” and being shown knee replacements.
That is the problem AI search optimization is actually solving, and it is why the usual advice about it misses the point.
What AI search optimization means
AI search optimization is the work of getting a business named inside the answers that AI assistants generate, rather than in a list of links below them.
When somebody asks ChatGPT, Gemini or Google’s AI Mode a question, they get a written answer that draws on a handful of sources and names some of them. Ranking fourth in a list of ten is worth something. In an answer where no list appears, it is worth nothing.
The difference from ordinary search is not the technique. It is the question being asked. Somebody typing into Google has already decided what they are looking for. Somebody asking an assistant is often still working out what is wrong with them, which means the assistant is doing the naming.
The words everybody is using, in plain terms
Four terms are circulating for roughly the same work, which makes this harder to research than it should be.
AI search optimization is the broad one: making a business findable and quotable across assistants and AI results.
Answer engine optimization, or AEO, is the same idea framed around answers rather than rankings. The name comes from treating a search engine as something that answers rather than something that lists.
Generative engine optimization, or GEO, is the academic term, from a 2023 paper that tested which page changes made a model more likely to cite a source.
LLM SEO is the same thing named after the technology underneath.
They describe one job. If an agency uses them as though they are four services, that is worth asking about.
How to rank in AI search
There is no ranking, which is the first thing to understand. There is no position to hold and no submission form. There is only whether a model, answering a question, decides your page is worth quoting.
Two things make that likely, and you need both.
These models lean on what already ranks. They do not go hunting for businesses nobody else rates. Local search positions, a complete Google Business Profile, other sites mentioning you by name: that is the pool they draw from. A practice invisible in local search is invisible here too.
The page has to be readable. The answer in the first line rather than three scrolls down. Fast. Marked up so a machine can tell what your practice is and what it treats. If the answer is buried inside a page builder under four layers of markup, a competitor whose page is easier to parse gets quoted instead.
Practices with both get named regularly. Practices with one and not the other mostly do not, which is why this runs alongside local SEO rather than instead of it.
How to show up on ChatGPT
The honest answer is that you cannot make it happen, and anybody promising otherwise is selling something they do not control.
What you can do is make yourself the easiest correct source to quote on the questions your patients actually ask, then measure how often it happens.
Start with the questions themselves. Not “vein specialist Houston”, which is what somebody types when they already know what they need. The ones that come before that: why do my legs ache at the end of the day, is there anything between physical therapy and a knee replacement, do I have to have surgery for fibroids.
Those are the questions an assistant answers by naming a procedure, and then naming somebody who performs it.
How do I know if it is working?
Google added a report this year showing how often your pages appear in AI answers, and almost nobody has opened it. It is in Search Console under Performance, and it separates AI results from ordinary search. That is a real number, and it is free.
For ChatGPT, Perplexity and the others there is no report at all. The only way to know is to run the questions on a schedule and record who gets named. That is sampling rather than measurement, and any agency telling you otherwise is describing something that does not exist.
So the honest position is one hard number and one directional one, and no pretending the second is the first.
Can I pay to appear in AI answers?
Partly, and this is moving quickly.
Advertising inside assistants exists and is being rolled out. Inventory is limited, approval is not automatic, and healthcare gets reviewed case by case rather than accepted by default. We have taken a plastic surgery practice through that review.
What you cannot buy is a place in the organic answer. There is no ad slot inside the text where the assistant names three practices. That part is earned or not at all.
The reason to look at the paid side now is cost rather than volume. Ad space inside assistants is priced like something nobody has noticed yet. That does not last.
Does this replace SEO?
No, and the framing is wrong. It is the same work pointed at a different surface.
A page that answers one question plainly, loads fast, and is marked up so a machine knows what it is looking at will rank in local search, get quoted in AI answers, and convert paid traffic. One piece of work, three returns.
What changes is the emphasis. Keyword targeting matters less, because the person asking may not know the keyword. Answering the question behind the question matters more.
What this looks like for a practice
Take the vascular practice again. The fix was not to bid harder on “genicular artery embolization”. Twenty one clicks is the entire market for that phrase, and none of them converted.
The fix was to build pages around what patients arrive with. Knee pain that has not responded to physical therapy. A surgeon who has recommended a replacement. Wanting to avoid a general anesthetic. Then to structure those pages so the answer is stated plainly and early, which is what makes them quotable.
A law firm has the same problem in different words. Somebody fired after calling in sick does not search “wrongful termination attorney”. They search “can my employer fire me for being sick”, and the assistant is what closes the distance between the two.
A home services company has it too. Nobody searches “polyaspartic floor coating”. They search how long epoxy takes to cure, and find out there is an alternative that is ready to walk on the next day.
What is still unknown
A fair amount, and it is worth saying so.
Nobody outside the companies knows how much traffic assistants actually send, because most of them do not report referrals. Ad formats inside assistants are months old and will be rebuilt more than once. The share of searches happening this way is still small next to Google.
What is not in doubt is the direction, and that the work which earns a citation is the same work that has always earned a ranking: a page that answers a real question clearly, on a site that is fast and correct, for a business that other people already rate.
That part has survived every format change so far.
We run AI search marketing alongside paid ads and local SEO for medical practices and local businesses. If you want to know whether your practice is being named in AI answers today, we can pull that number from your own Search Console before quoting anything. Ask us.