Answer engine optimisation attracts a great deal of claim and very little mechanism. The mechanism is not complicated, and stating it plainly removes most of the noise: a generative system answers a question by retrieving documents and composing an answer from passages within them. A publisher can influence whether their document is a retrieval candidate, and whether the passage inside it is easy to lift. Nothing else.
Retrieval eligibility comes first
A page that cannot be fetched is not a candidate, and a page that is not indexed by whichever index the system draws on is not a candidate either. That makes the whole of the previous section of this publication a prerequisite: crawlable, server-rendered, indexed, canonicalised and fast. Nothing about answer engines replaces those requirements, and a great deal of AEO advice quietly assumes they are already met.
Two additional access questions matter. Does your robots.txt permit the user agents the system uses, and does any bot protection at the network layer challenge them? Many clinics have a security product in front of the site that blocks unfamiliar agents by default, which is an access decision made by an infrastructure setting rather than by anyone with an opinion about it.
The unit is the passage, not the page
Generative systems compose answers from passages. That changes what good writing looks like. A page that develops one argument across two thousand words, where each paragraph depends on the last, is excellent prose and poor extraction material. A page where each section answers one question completely is less elegant and far more useful.
Copy any paragraph from your page and paste it somewhere with no surrounding context. Read it. Does it state a complete fact, with its subject named? Or does it begin "This means that" and refer to something a reader can no longer see? The first is citable. The second is not.
The properties that make a passage liftable
- Named subject. Write "polynucleotide treatments typically require a course of sessions" rather than "these usually require a course", because the pronoun does not survive extraction.
- One claim per paragraph. A paragraph making three claims can only be lifted whole or not at all.
- Question-shaped headings. A heading that asks the question is a strong signal about what the passage below answers, for both readers and machines.
- Specific rather than hedged. "It varies" is not an answer. "It varies with the depth of placement, and typically lasts between x and y for most patients" is, provided both figures are true.
- Unit and scope stated. Numbers without units, populations or time periods cannot be quoted safely, so they are less likely to be used.
- No dependence on a table or image to make sense. If the fact only exists in a graphic, the passage cannot carry it.
Consistency across your own surfaces
If a fact about your clinic appears three different ways across your site, your Business Profile and a directory listing, a retrieval system has three candidate values and no basis to prefer one. Where your opening hours, address, service list or practitioner names are concerned, the single most useful AEO action is to make every copy identical. Consistency raises confidence, and confidence is what gets a fact reused.
Structured data as disambiguation
Markup does not make a page citable and no serious reading of the documentation suggests it does. What it does is remove ambiguity about what the page is and who published it. A page marked as MedicalWebPage, with a named publisher, a date and a clear subject, is easier to classify than the same page unmarked. Google's structured data guidelines require the markup to describe visible content, so it is a description of quality rather than a substitute for it.
Provenance is part of the answer
Systems that attribute prefer sources they can characterise. A page that names its author, states their professional registration, carries a review date and identifies the organisation behind it is easier to describe in an answer than an unsigned page on an anonymous domain. For clinics this is straightforward, because the provenance is real: a registered practitioner, a registered clinic, a public register entry confirming both.
Where a clinic is registered with a body that publishes a public record, such as the CQC in England, that record is an independent confirmation of identity. Linking to it, and making sure the details agree, is a cheap and durable provenance signal.
AI answers and assistants
LOW CONTROL- Whether your content is crawlable by the relevant user agents
- How clearly a page states the fact you want quoted, and where
- Whether the same fact is consistent across your site, profile and listings
- Structured data that removes ambiguity about who and what you are
- Whether a model chooses to cite you
- Which passage it lifts, or how it paraphrases it
- Whether an answer appears at all for a given prompt
- The training data of any model already released
- Run the same prompt repeatedly and record what is cited, with dates
- Check server logs for AI user agents rather than guessing at access
- Verify the fact you want quoted appears verbatim on a crawlable page
Dates, and being able to prove them
Publish a visible published date and a visible last reviewed date, and make the structured data agree with both. Do not update the date without updating the content, which is a pattern systems can detect across a corpus and which costs more credibility than it buys.
What does not work
- Adding a block of text addressed to language models. It is content on a page like any other, and it is not a control surface.
- Publishing an llms.txt file and expecting retrieval to change. It is a proposal, not a supported standard.
- Stuffing FAQ markup onto pages with no visible questions, which breaches the structured data policies.
- Quoting statistics you cannot trace. If a system is checking, an unsourced figure is a liability, and if it is not checking, the figure did nothing for you anyway.
- Generating hundreds of thin question pages. Thin content is thin regardless of which system is reading it.
What a clinic should actually do
- Make the site server-rendered, crawlable and fast, and confirm AI user agents are not blocked at the network layer.
- Rewrite the top ten pages so each section answers one question in a self-contained passage.
- Name subjects explicitly. Remove pronouns that only resolve in sequence.
- Make every factual claim about the clinic identical across the site, the profile and the listings.
- Add accurate structured data describing the organisation, the pages and the practitioners.
- Publish and maintain visible dates and authorship.
- Record what is cited for your key questions, monthly, so you have a baseline before claiming anything about change.
None of that is exotic. It is the discipline of writing clearly about verifiable things, applied to a reader that cannot infer context. That is the whole of the technique, and treating it as anything more mysterious is how budgets get spent on claims nobody can test.