Last reviewed 25 August 2026.
Six categories cannot buy ads inside ChatGPT: health, mental health, politics, gambling, legal services and financial services. For every business in those verticals, and for anyone whose budget will not stretch to paid testing yet, there is a second route into this channel. Get named in the answer itself.
Why this is a different discipline
Classical search optimisation aims to rank a page. An assistant is not ranking pages, it is composing an answer from sources it treats as authoritative and corroborated, then naming a small number of options. You are not competing for position three on a results page. You are competing to be one of perhaps four brands mentioned at all.
That shifts the work in three ways. Corroboration matters more than any single page, because being described consistently across independent sources is what makes a model confident enough to name you. Specificity beats breadth, because assistants pull from content that answers a precise question rather than content that covers a topic broadly. And structure matters, because clearly attributed, directly stated claims survive summarisation while discursive marketing prose does not.
What actually moves the needle
Answer the question in the first forty words. Under every heading, state the answer plainly and immediately, then elaborate. Content built this way gets extracted. Content that builds to a conclusion over four paragraphs does not.
Be described the same way everywhere. Your own site is one source among many, and not the most trusted one. Directory listings, review platforms, industry publications, and comparison sites all feed the picture. Inconsistency between them creates uncertainty, and uncertainty means you do not get named.
Publish things that can be cited. Original data, clear definitions, and specific comparisons get referenced. Undifferentiated overview content does not, because there are already fifty versions of it.
Date and attribute everything. Assistants weight recency heavily in fast moving categories. An undated page is a weaker source than a dated one, and a page reviewed last month beats one published two years ago.
What does not work
Keyword density does nothing. Publishing volume for its own sake actively hurts, since thin content dilutes the corroboration signal across your domain. And there is no equivalent of a backlink scheme here. You cannot manufacture the impression of authority across independent sources without actually earning it, which is unusual in this industry and largely a good thing.
Measuring it
This is the uncomfortable part. There is no rank tracker for AI answers, and results vary by user context. The workable approach is systematic manual sampling: a fixed set of buying questions in your category, run at intervals, recording whether you are named and alongside whom. Tedious, but it is real data and almost nobody is collecting it.
Separately, track referral traffic from AI assistants in your analytics as a distinct channel. It is currently small for most businesses and growing.
Realistic timelines
Slower than paid, faster than classical SEO on a new domain. Corroboration takes months to build, but a single well placed third party mention can change your visibility in a category faster than a link would change your rankings. It also compounds and does not stop the day you pause a budget.
For restricted categories this is not a consolation prize. It is the only route available, and it carries none of the compliance exposure that paid advertising in regulated verticals does.