AEO
When a patient asks for a recommendation, be the practice that gets named.
A patient used to type three words and read a list, and a growing share of them now type a full sentence and read a single answer with two or three sources attached to it.
Ten places to appear became three. Everything on this page follows from that.
Answer engine optimization is the work of being one of the three. It is not a new budget line and it is not a product. It is a set of decisions about how your pages are built, and most of those decisions help you in ordinary search as well.
What answer engine optimization is, and what it borrows from SEO.
It is the work of building pages an assistant can lift an answer out of, and of making sure it names you when it does. Google’s AI summaries, ChatGPT, Perplexity, Gemini. Your same pages, read by something that is assembling a reply rather than ordering a list.
The mechanic underneath is retrieval rather than ranking. A search engine puts ten results in order and lets the patient choose, while an assistant pulls a handful of passages, merges them, and credits some of what it used. Tenth place is worth something in the first case. In the second it is worth nothing at all.
Most of what is sold as AEO right now is a rename. The audits are SEO audits with the word AI in the title, and the deliverable is often a text file that no assistant has agreed to read.
Two things are genuinely different, which is a shorter list than the market implies. Pages have to finish a question instead of covering a subject, and the facts about you have to agree everywhere a machine might look, because disagreement is what makes an assistant hedge and name somebody else. The remainder is healthcare SEO done properly, and local search for the questions that end in a place rather than an answer.
What we do
Five things we build, and none of them is a file you upload.
One page, one question, answered to the end
Most practice sites cover a subject three times badly: a service page, a blog post, an accordion of half-answers. None of the three finish the thought. An assistant lifts passages out of pages, so we consolidate. The question becomes the heading, the answer sits under it in language a person could read aloud, and the exceptions come after. That construction also wins featured snippets in ordinary search, which is why the work pays for itself while the rest is still a bet.
The answer first, the argument second
A page that spends two hundred words setting up the question gets skipped by the machine and by the practice manager reading it. Definition or number in the opening sentence, then the detail underneath.
Structured data that matches the page
Schema tells a machine what your page already says, and it cannot tell it something the page does not say. We mark up services, locations, providers and questions where the text is genuinely there, and we strip the markup a previous agency left claiming reviews and prices that appear nowhere. That kind of claim earns nothing, and it is how a practice gets its rich results switched off.
One version of who you are, in every place a machine looks
An assistant assembles its answer from several sources at once. When your site, your Google profile and a directory entry from 2019 disagree about your locations or your specialties, hedging is the safe move and hedging means naming someone else. So the dull reconciliation work belongs to this service as much as it belongs to local search.
The pages you do not own
A good share of what gets cited about a practice sits on a directory, a review site, or a forum thread from four years ago. In the accounts we run, we check which sources are actually being pulled for the questions your patients ask, then correct what is wrong about you on the ones that allow it. Some of them do not allow it, and we tell you which.
A person writes and edits every page before you see a draft, the same way we run every other content engagement. Nothing clinical gets written here: it goes to your clinicians first, because we sell marketing and this is not a subject worth guessing in.
Measurement
How you would know it is working.
Not from a dashboard. There is no rank tracker for an answer that gets written fresh every time somebody asks for it, and any tool selling you one is sampling and calling it something grander. So we sample, and we call it sampling.
Each month of the engagement, we run the same list of questions your patients actually ask, through the same assistants, and log which practices get named. The list gets written once with you at the start and then left alone, because changing the question changes the answer and a moving list would prove nothing at all.
Two softer signals sit alongside that list: referral traffic from assistants, which is real and badly under-counted, and your branded search volume, which tends to move when people start hearing your name somewhere they had not heard it before.
The signal we trust most sits in your own analytics rather than in somebody else’s: how each type of page converts. If the pages built to finish a question convert better than the pages built to cover a subject, the strategy is earning its place even when no assistant sends a click. That is the one measure on this page we have already run for a client, and the account below is it.
What one behavioral health network’s own numbers showed.
We run search and the analytics under it for a 17-location behavioral health network, and that is as much as we will say about who they are. Everything quoted here comes from a single month of their reporting in spring 2026, at the values we were given.
20.8% vs 0.3%
- Share of organic visits from facility pages
- 8%
- Share of organic leads from the same pages
- 50%
- Locations in the network
- 17
Eight percent of the visits produced half of the leads. We saw it because we measured conversion page type by page type, which most monthly reports never do. That is one reason so many practices are still funding a blog nobody converts on.
Those facility pages were not written for assistants. They were written because a patient asks where, who, how soon and what it costs, and a page answering those four things is the page that gets used. That is the same construction an answer engine retrieves, and it is the argument this service rests on.
Read the limits of it with us. One month, one client, no second case behind it, and calling that a pattern would be a claim we have not earned. It shows that answer-shaped page types convert. It does not show that they get cited. We hold no vetted citation data of our own, which is why the measurement section above begins with a sampled list rather than with a number.
No traffic figure appears anywhere on this page, and that is deliberate. The network moved its site not long before this reporting, and a year on, its organic traffic has not climbed back to where it was. Printing a session count from the account without saying that alongside it would be the exact reporting habit the rest of this page argues against.
The network’s full account, migration first, is on the results page.
Who it's for
Practices whose patients ask before they search.
- Behavioral health
A family asking an assistant which facility takes their insurance, for someone they are trying to get admitted this week.
- Cosmetic dentists
High cost, one decision, three practices compared. The comparison used to happen across four tabs. Now it often happens inside one answer.
- Plastic surgeons
The patient researching from another country, in another language, with no local knowledge to check the answer against.
- Med spas
Questions about price and downtime that most practice sites still refuse to answer in writing.
- Dermatology
A cosmetic buyer who reads for weeks and wants a price before anyone picks up a phone.
Where it matters least is urgent care. That patient wants the nearest open door in the next twenty minutes rather than a recommendation, and the map decides it. For the multi-location urgent care group we run, that is where the money goes, and we would tell you the same. The services overview shows how the three fit together.
The boundary
What AEO doesn’t do.
- It does not buy you a citation
- There is no slot in an answer and no vendor selling one. What this work changes is the odds that your page is among the ones retrieved. Any agency promising you a guaranteed appearance in an assistant's answer has sold you something nobody owns, and the useful reply is to ask how they intend to verify it next month.
- It is not a channel with its own budget
- It rides on pages you already have. If the site is thin there is nothing to retrieve, and no amount of markup fixes that. Practices arrive expecting a bolt-on, and in the accounts we run what most of them turn out to need first is three pages that answer a question properly, which is ordinary SEO with a deadline on it.
- It does not report cleanly, and it may never
- Assistants pass very little referral data, and they personalise the answer they give. Two people asking the same question in the same hour can be shown different sources. You cannot rank-track this the way you rank-track a keyword, and a tidy AEO dashboard is a model rather than a measurement. We would rather hand you a sample and call it a sample, the same way we report everything else in the accounts we run.
- It does not win the map
- When somebody searches for care near them, three map results still decide it, and that is a different mechanism with a different fix. We scope it as local search and we do not let an AEO proposal quietly absorb it.
- It does not come with a shelf life
- Retrieval changes when the models change, and that happens without notice and without a changelog. A page can be cited for two months and then not be. We build for the version of this that survives an update, which means pages that are genuinely the best answer to the question asked. Everything else has an expiry date on it.
Questions
What practices ask about getting cited.
- Is this just SEO with a new name?
- Mostly it is SEO done properly, plus two things SEO never had to care about. An assistant pulls a passage out of a page rather than ranking the page, so the page has to finish one question instead of covering a subject. And it assembles from several sources at once, so your site, your profile and the directories have to agree about you. The rest is the same discipline.
- Can you guarantee we show up in ChatGPT?
- No, and neither can anyone else. There is no placement to buy and no position to check. Two people asking the same question can get different sources, because the answer is generated fresh and shaped by what the model already knows about them. What we can do is build the kind of page that gets retrieved, then measure how often you get named.
- How do we know it is working?
- By sampling, and we say so. The engagement includes a monthly run of a fixed list of the questions your patients ask, a log of which practices get named, and a comparison against the month before. Alongside that we watch the referral traffic assistants pass, which is real and under-counted. The signal we trust most is conversion by page type inside your own analytics.
- Do we need an llms.txt file?
- No. It is a proposed file that tells assistants how to read your site, and no major one has committed to honouring it. Agencies sell it because it takes ten minutes to produce and cannot be disproved. We will add it if you want it, and it will not appear on an invoice.
- If patients get the answer without clicking, does this cost us traffic?
- Some of it, yes. Informational visits are already falling across healthcare and that continues. What survives is the visit with intent behind it: someone who read an answer, saw you named in it, and arrived to book. Fewer sessions converting at a higher rate is the shape to expect, and it is why session counts are a poor way to judge this work.
- Do we have to publish medical content to get cited?
- Not written by us. We are a marketing agency with no clinical credentials, and publishing treatment advice to win a citation is a bad trade for the practice carrying the liability. We write what we can stand behind: cost, insurance, what a first visit involves, how to choose between two options. Anything clinical goes to your clinicians before it publishes.
Send us ten questions.
Your site, and the ten questions you wish patients asked about your practice. You get back who the assistants name today for each one, what those answers say about you where you appear at all, and which of your pages are close enough to be worth rebuilding.
In writing, before there is any question of a contract. If the honest answer is that your site needs ordinary SEO work first, that is what the analysis will say.