Getting cited by ChatGPT, Gemini and Google AI Mode: a 90-day plan.

AI answer engines do two things before they write. They retrieve, which means they search the web, often through a conventional search index, and pull the pages that look relevant. Then they read those pages and decide which ones to trust and quote.

To be cited, a business has to be findable in the retrieval step and legible in the reading step. Findable is mostly the familiar search work. Legible is newer: the engine has to understand what the business is, where it is, what it does and for whom, without ambiguity, and it has no patience for gaps a human reader would fill in.

The sources engines tend to cite are the business's own pages, review platforms, directories, press, forum threads where real people answer questions, and reference sites. A business that is consistent across all of them is easy to trust. A business that is three slightly different businesses across them is not.

The first step makes the business unambiguous.

Organization and LocalBusiness structured data go on the site, with the same name, address and phone number the business uses everywhere else. The About page states what the business is in one plain sentence, because that sentence is what a model will lift. Any page that gives advice gets an author with real credentials. The Google Business Profile is completed and matched to the site. A file at /llms.txt describes the site for AI crawlers, following the llms.txt proposal.

Then the sweep: every directory and review listing is checked and corrected for the name, the category, the hours and the address. A model that finds three versions of a business trusts none of them.

The second step gives the engines something worth quoting.

FAQ pages that answer the question in the first sentence and then explain. Comparison pages that state the facts plainly, including the ones that do not flatter you. Pricing and process pages that say how it works and what it costs, in words, without a form in the way. Tables and lists wherever a table or a list is the honest format, because models lift them cleanly.

Every page is written to be understood by a reader who arrives with no context. That happens to be the same reader a search engine serves, so none of this work is wasted on people.

The third step works on the sources the engines cite that you do not control, and on knowing whether any of it is working.

Earn mentions where real answers live: trade press, local press, the review platforms for your industry, community forums, and expert roundups. Ask satisfied customers for reviews that describe the work rather than the mood, because a review that says what was done is a review a model can use.

Then measure. Ask the major engines the questions your customers ask, record who is cited, and repeat every month. Share of answer against competitors is the number to watch. It moves more slowly than rankings, and it is worth more, because a citation in an answer is the whole first page.

It is not a trick. There is no tag that makes a model cite you, and anyone selling one is selling something. It is the same discipline as search done properly, plus a layer of clarity that most sites never bothered with because a human reader could fill in the gaps. A model does not.

If your customers ask an AI which business to call and you are not in the answer, the AI search visibility service runs this plan, and the discovery call is where it starts.