How an assistant decides who to name
An assistant does not have a ranking table. It retrieves pages, profiles, directories and reviews, then summarizes what it finds into an answer with a few names and a few citations. So the useful question is not "how do I rank" but "can a machine tell what this business is, where it works and why it should be trusted, in a form it can quote?"
That reframes the work. Instead of chasing positions, you are making your business easy to understand and easy to corroborate.
Five things that improve your odds
1. One consistent identity
Same business name, address, phone, hours and services on your website, your Google Business Profile and every directory that carries a copy. Add Organization and LocalBusiness schema to your site, with sameAs links to your real profiles. Conflicting details make you a weaker candidate to name.
2. Answers rather than brochures
Assistants quote passages. A passage only works if it stands alone, so put the answer in the first two sentences under a heading phrased the way a person actually asks. "How much does a water heater replacement cost in Tacoma?" beats "Our services".
3. Evidence somewhere other than your own site
Reviews, citations, local mentions, named clients and case studies. A claim that only exists on your homepage carries little weight. A claim corroborated across your profile, a directory and a review is much easier for a model to repeat.
4. Specifics that survive summarizing
Service areas, hours, brands you work on, what you do not do, prices where you publish them. Adjectives disappear when text is summarized. Facts survive. "Award-winning quality service" gives an assistant nothing to say about you.
5. Clean technical access
Server-rendered HTML rather than content that only appears after JavaScript runs, fast pages, a correct sitemap, and no blocking of the crawlers that feed AI answers. If a machine cannot read the page, nothing else on this list matters.
How to measure it
Pick fifteen questions a real customer would ask, run them through ChatGPT, Google AI Overviews, Perplexity and Gemini, and write down who gets named for each. That is your baseline. Re-run the same set monthly. It is unscientific in the sense that answers vary between runs, but the pattern over a few months is clear enough to act on.
Also check your analytics for referral traffic from assistant domains. It is usually small today and growing.
What nobody can sell you
There is no paid placement inside an AI answer, no submission form, and no way to guarantee a mention. Anyone promising a ranking in ChatGPT is selling something they do not control. What is controllable is everything the assistant reads, which is most of the work above, and most of it improves ordinary search rankings too.
Common questions
Is this different from SEO?
It overlaps heavily. The differences: entity consistency and schema matter more, answer-shaped passages matter more, and success is measured by whether you get named rather than by a position.
Do I need an llms.txt file?
It is cheap to add and does no harm, but it is not the deciding factor. Consistent details, quotable answers and outside corroboration matter far more.
How much traffic comes from AI search today?
For most local businesses, a small fraction, and rising. The reason to act now is that the work also improves normal search results, so you are not betting on a single channel.
Does this matter for a trade business?
Increasingly. "Near me" questions are exactly what people ask assistants, and the answer is a shortlist of two or three names. You either make it or you do not.
Where to start
Ask an assistant the question your best customer would ask, and see who it names. If it is not you, that gap is the brief. Our AI search optimization page explains how we run this as an engagement, including the baseline test.