Somebody in your city opened an assistant, described their problem in a sentence, and got back two or three names. Not ten links. Two or three names, with a line each about why.
If yours was not among them, nothing dramatic went wrong. There was no penalty and no algorithm change. Your business simply was not among the things the model could confidently say something about — and confidence, not ranking, is what that shortlist is made of.
How an assistant actually picks the names
An assistant answering "who should I use for X near me" is doing two things at once. It is retrieving pages that look like they answer the question, and it is checking whether what those pages say agrees with what everything else says about the same business.
The second half is the one nobody optimises for. A model that finds one page claiming you do commercial interiors, a directory listing you under residential, and a profile with a different phone number has not found evidence — it has found a disagreement. The safest output when sources disagree is to name a business it is surer about.
The four reasons it is usually not you
Your answer is not on the page in one piece. The information exists, spread across a services page, a paragraph in the footer and a PDF. A person will assemble that. A retrieval system takes the block of text under the heading that matched, and if that block does not answer the question by itself, your page contributed nothing.
Your headings are labels, not questions. "Our Services" matches nothing anybody types. "Do you take same-day emergency callouts?" matches something somebody types, and the paragraph under it is quotable as it stands.
Your facts disagree across the web. The name, the number, the description and the area you cover have to be identical everywhere they appear — your site, your Google Business Profile, every directory. Twenty listings that disagree are worse than three that match.
Nothing about you exists off your own site. A model assembling a description of a business is reconciling several sources. If the only source is the business itself, there is nothing to reconcile and little to be confident about.
What that means to do
Not much of this is exotic, and that is the useful part. Give every question a heading that matches how it is asked, put a complete answer directly underneath it, and make every public statement of your name, number and service area agree character for character.
There is no schema type that makes a page eligible to be cited, whatever anybody is selling. Google's own guidance on its AI features names none. What is left is the boring version: be the page that answers the question completely, and be describable without contradiction.
What nobody can promise you
Anybody who offers you a guaranteed mention in an AI answer is selling something they cannot deliver. These systems are not ranked lists and there is no position to buy or hold. What can be done is to remove every reason a model has to pass over you — which is a real, finite, checkable list of work, and it is most of what this studio does.