A growing share of B2B buyers now start vendor research by asking an assistant rather than a search engine, and the assistant answers in prose with a short list of named companies. If you are not in that list, there is no second page to be on.
This is the uncomfortable part of generative engine optimisation: the familiar lever — rank higher, get more clicks — does not apply, because frequently there is no click at all. The question is not how to get traffic from an AI assistant. It is how to be the company it names.
Ranking and being cited are not the same thing
A page can rank well and never be quoted, and a page that ranks modestly can be cited constantly. The reason is that an assistant is not choosing a destination for the user; it is assembling an answer, and it needs passages it can lift.
A long essay with the answer distributed across it is hard to quote. A section with a clear heading that states a specific answer in the first two sentences is easy to quote. The practical change is structural: write so that each section can be extracted and still make sense with nothing around it.
This is also why FAQ sections earn their place. A question as a heading and a direct answer beneath it is close to the ideal shape for retrieval.
Be specific about things nobody else is specific about
Generic content is not cited, because there is nothing to distinguish it from the fifty other pages saying the same thing, and the model has no reason to prefer yours.
What gets quoted is the narrow, concrete claim: the actual sequence of a process, the thing that goes wrong at step four, the number that only someone who has done the work would know to mention. If your article could have been written by someone who has never done the job, it will not be the one that gets named.
This is where most B2B content fails long before anything about AI enters the picture.
Answer the comparison questions directly
Buyers ask assistants comparative and qualifying questions — which tool for this situation, what does this cost, what are the alternatives, is this suitable for a company of our size. Vendors tend not to answer those on their own site, because comparison feels like it invites the prospect to leave.
The consequence is that the assistant answers from somewhere else: a review site, a competitor’s comparison page, a forum thread. You are then described by a third party rather than by yourself.
Answering honestly — including where you are not the right choice — is also what makes a page quotable, because a page that claims to be best at everything reads as marketing and gets treated as such.
You are described by sources you do not own
Your website is one input. Directory listings, review platforms, professional profiles, press coverage and discussion threads are others, and collectively they often carry more weight than your own copy, because they are not self-descriptions.
The baseline work is consistency: that your company is described in the same terms, with the same service names and the same locations, everywhere it appears. An assistant assembling a description from six inconsistent sources will produce a vague one.
It is worth periodically asking the assistants themselves what they say about your company and your category, and treating a wrong or outdated answer as a content gap with an address — usually a specific page that does not exist or a listing that was never corrected.
Make the page readable to a machine
Some of this is ordinary technical hygiene that happens to matter more now.
Content that only exists after JavaScript executes may not be seen by every crawler. Organisation and FAQ structured data state plainly what a parser would otherwise have to infer. A visible, accurate last-updated date matters because assistants favour current material and have no way to judge freshness other than what you tell them. If you have deliberately blocked AI crawlers in robots.txt, you have also opted out of being cited — a legitimate choice, but it should be a decision rather than an inherited default.
Measuring it is genuinely harder
Be careful with confident claims here, including from agencies selling the service. Citation behaviour varies between assistants, changes without notice, and is not reported in the analytics you already have. An answer that cites you today may not tomorrow, for reasons nobody outside the vendor can see.
What you can do is track it directly: maintain the set of questions a buyer in your category would actually ask, run them across the main assistants on a schedule, and record whether you appear and how you are described. It is a manual, imperfect measure. It is considerably better than assuming.
And the underlying work — specific content, honest comparisons, consistent descriptions, clean markup — is the same work that has always made a site worth citing. That part has not changed.
This sits alongside the rest of the demand work in digital marketing and automation and social media management. Related: what “AI-powered” content actually automates.
Common questions
What is generative engine optimisation?
Structuring and sourcing your public content so AI assistants name and cite your company when they answer a buyer’s question. It overlaps with SEO but optimises for being quoted inside an answer rather than for a click through to your site.
Does traditional SEO still matter for AI visibility?
Yes. Assistants largely draw on indexed web content, so being findable remains a precondition. What changes is that ranking alone is not sufficient — content also has to be extractable, specific and quotable at the level of an individual passage.
How do we know if AI assistants mention our company?
Test it directly. Write the questions a buyer in your category would ask, run them across the main assistants on a regular schedule, and record whether you are named and how you are described. There is no reliable passive report, so a manual tracked list is currently the honest approach.
