AI search optimisation: being in the answer, not just the results.

Search still works the way it did. What changed is what sits above it: an answer assembled from a handful of pages, shown before the results, which a large share of people read instead of clicking anything at all.

The mechanics matter more than the noise around them. Google's AI Overviews, ChatGPT's search, Perplexity and Bing's Copilot all do roughly the same thing. They read a small number of sources, assemble an answer, and cite some of those sources.

So there are two questions now where there used to be one. The old question was whether you rank. The new one is whether you are among the pages an answer is built from, and being cited in an answer is not the same as ranking first.

What has not changed is the underlying work. The pages that get cited are, overwhelmingly, pages that already rank well and answer a specific question clearly. There is no separate discipline here that replaces search engine optimisation. There is an extra layer on top of it.

Short, self-contained passages that answer one question completely, on pages that already rank, on a site the model can read. That is most of it, and the third condition fails more often than people expect.

The passage part is the piece worth understanding. A model assembling an answer is looking for a chunk of text it can lift and attribute. A paragraph that answers the question on its own, without needing the three paragraphs above it, is liftable. A paragraph that only makes sense in context is not.

This is why a lot of well-written marketing copy performs badly here. Copy that builds an argument across a page reads well to a person and gives a model nothing it can quote.

What makes a page citable

• One question, answered completely, in one place.

  Not spread across a section. A heading that is the question and a paragraph under it that answers it.

• Written so a passage stands alone.

  If a sentence needs the previous paragraph to make sense, it cannot be lifted.

• Specific, checkable facts.

  Numbers, conditions, exceptions and named things. Vague copy gets skipped in favour of copy that commits to something.

• Readable by a crawler at all.

  A page that needs JavaScript to show its text is invisible to several of these systems, and this is the failure nobody checks for.

AI crawlers are separate from Google's, they are blocked by default on a lot of hosting and security setups, and a site can be perfectly visible in Google while being completely invisible to every AI system reading it.

This is the most common finding and the least discussed. A robots file, a firewall rule or a bot protection setting written before any of these crawlers existed will block them without anybody deciding to.

The decision about whether to allow them is a real one, not automatic. Allowing them means your content can be summarised without anybody visiting your site. Blocking them means you cannot be cited at all. For a local service business the second is almost always worse than the first, but it is worth choosing deliberately.

There is a separate file some people will tell you to add for this. It is not read by Google, and the evidence that anything else weights it meaningfully is thin. Adding it takes about a minute, which is why it is not worth arguing about and not worth paying for either.

Check what the crawlers can see, fix anything blocking them, rewrite the pages that already rank so each question is answered in a liftable passage, and add the structured data that tells a machine what the page is about. In that order.

The order matters because the first step is free and can disqualify everything after it. There is no point rewriting a page for citation if the systems doing the citing cannot fetch it.

The rewriting is the slow half, and it is best done on the pages that already rank rather than on new ones. A page in the top ten for a question is already most of the way to being a source for that question's answer.

What this does not require

It does not require a new website, a new platform, or a separate monthly fee. It does not require publishing more. Several agencies are currently selling all three on the back of this, and none of them are necessary.

The wider picture, including what has not changed for a local business, is in Australian search in 2027.

The order worth doing it in

• Check what the AI crawlers can actually fetch.

  Free, fast, and it is the step that most often finds the real problem.

• Fix the blocking, then decide it deliberately.

  A firewall rule nobody chose is not a decision.

• Rewrite your ranking pages into answerable passages.

  One question per heading, answered under it, standing on its own.

• Add the structured data.

  It costs an afternoon and it tells a machine plainly what the page is and who wrote it.