A growing share of commercial questions now get answered rather than searched. Someone asks an assistant which firm in Dubai does a particular kind of work, and the assistant produces an answer with a handful of names in it. If you are not one of those names, no amount of ranking on page one helps you, because the page was never opened.
This is a different problem from search ranking, with a different set of causes. We audited our own site before selling this work to anyone else, and scored 42. Most of what was wrong was mechanical.
First, check you are not blocking them
The most common cause of invisibility is the least interesting one: the crawlers are blocked. Many sites added blanket AI-crawler restrictions during the period when that seemed prudent, or inherited them from a CDN's default bot rules, and then began wondering why assistants never mention them.
These are distinct crawlers with distinct names, and blocking is rarely all-or-nothing. A site can be readable by one assistant's indexing crawler and invisible to another. Check the live file rather than assuming, and check your CDN's managed bot rules separately, because those can block traffic that your robots.txt happily permits.
Worth knowing: several of the agencies currently ranking well for AI-related searches in this market block AI crawlers outright. They cannot be cited by the assistants their own clients are asking. That is an opportunity if you are competing with them.
Be a resolvable entity
An assistant recommending a business is making a claim about a real entity, and it needs to be reasonably confident the entity exists and is what it appears to be. That confidence comes from consistency across sources: the same legal name, the same address, the same founder, the same contact details, appearing in structured data on your site and matching what exists elsewhere.
Small inconsistencies do real damage here. A trading name on the website and a legal name in the registry, an old address in a directory, a phone number formatted three ways. A human resolves those instantly. A machine treats them as evidence of ambiguity.
Answer the question in the words it was asked
The structural change is that content optimised for ranking and content optimised for citation are not the same shape. Ranking rewards depth and authority signals. Citation rewards a passage that cleanly answers a specific question and can be lifted without distortion.
- Ask the actual question as a heading, in the phrasing a person would use.
- Answer it in the first two sentences underneath, before any preamble.
- Keep the answer self-contained, so it survives being quoted without the paragraph around it.
- Put facts in text, not in an image or a chart with no caption.
- If the answer is genuinely conditional, say what it depends on rather than hedging vaguely.
Answer the questions nobody else will
The highest-value content for this is the honest negative. What a service cannot do, when it is the wrong choice, what it actually costs, where it stops working. This is thinly covered on the open web because marketing pages avoid it, and it is exactly what people ask assistants, precisely because they expect a marketing page to dodge it.
Every service page on this site carries a section saying where that service stops. That is partly principle and partly strategy: it is the content most likely to be quoted, because so little of it exists.
What moved our score
From 42 to 75, in rough order of contribution: unblocking and then explicitly naming AI crawlers; adding a consistent entity graph across every page rather than only the homepage; restructuring service pages so each opens with a plain answer to the question it targets; adding explicit limitations sections; and fixing internal linking so every page is reachable and canonical URLs agree with the links pointing at them.
None of that is clever. It is mostly the removal of obstacles. The reason the score was 42 was not that the content was bad, it was that the site was quietly difficult to read.