The search that no longer happens on a results page
A growing share of the people looking for a local business never see a results page. They ask ChatGPT, Perplexity, Gemini or Google's AI Overview a question in plain language and receive an answer naming two or three businesses. If you are not one of the businesses named, you were not rejected. You were never in the conversation.
This is a different competition from ranking, and it rewards different things. A results page shows ten links and lets the person choose. An AI answer names a handful and the choice is largely already made. The shortlist is the whole game.
What makes this urgent for local service businesses specifically is that the questions people bring to these tools are exactly the high intent ones. Not "what is HVAC" but "who should I call for an AC repair in San Antonio and what should it cost." That is a buying question, and the businesses named in the answer are the ones who get the call.
What we found when we measured this for San Antonio
We ran buyer stage queries against AI search and recorded which businesses were named and, more importantly, which sources the answers were built from. The results were consistent enough to be actionable.
Across queries such as best website company for a small business in San Antonio, and San Antonio local SEO and AI visibility for contractors, the named businesses came from two source types and almost nothing else.
Directory platforms. Clutch, DesignRush, GoodFirms, the Semrush agency list, OnToplist and similar aggregators. These sites have enormous authority and structured, machine readable listings, which makes them exactly the kind of source an AI system leans on.
Articles the competitors published themselves. This was the finding that matters. A meaningful share of the cited sources were best of and top ten style articles published on the blogs of the very agencies being recommended, ranking themselves inside their own lists. The AI answer treated them as editorial sources.
Neither source type is a ranking. Neither requires domain authority in the traditional sense. Both are reachable deliberately, which is what this service does.
What actually gets a business named
Entity clarity: the machine has to know what you are
Before anything can recommend you, it has to be confident about who you are, what you do, where you work and whether the business described on your site is the same one described on Google, on Yelp and in that old directory listing from 2019. Ambiguity is not neutral. It is disqualifying, because a system that is unsure will name a business it is sure about.
The work here is structured data that describes what is visibly true on the page, consistent business information everywhere it appears, an explicit statement of services and service area, and a clear relationship between the organization, its services and its locations. Our explanation of schema markup covers the mechanics without the jargon.
Citation ready passages: write the answer, not the essay
AI systems extract passages. A page that buries its answer in the seventh paragraph loses to a page that states it in the first forty words and then supports it.
In practice this means every substantial page opens with a direct answer to the question it is named after, uses descriptive headings that match how people actually ask, and includes a real FAQ section where each answer is self contained enough to be quoted without the surrounding context. This is not a trick. It is the same structure that makes a page useful to a human who is scanning.
Specific, verifiable facts
Vague marketing language is unusable to a system trying to build an answer. Concrete facts are quotable: what a service costs in a given range, how long something takes, which permit applies, what a specific code requires. Pages built from real numbers get cited. Pages built from adjectives do not.
The corollary is that a single invented figure is fatal, because credibility is the entire asset in this strategy. We do not publish a number we cannot trace.
Being present in the sources these systems actually read
This is the highest leverage item and it happens off your own site. Since the answers are built substantially from directory platforms, being listed accurately and completely on Clutch, DesignRush, GoodFirms and the equivalent industry aggregators for your trade does more than most on-site work. It is unglamorous, it is largely form filling, and it is frequently skipped precisely because it is not interesting.
Reviews and third party mentions
Review volume and content feed prominence in both traditional local ranking and AI answers. Reviews that describe the actual work in a customer's own words carry topical information, not just a star rating. Genuine local press, association membership and supplier listings function the same way.
An llms.txt file and a crawlable site
An llms.txt file gives AI systems a plain language summary of what the business is, what it offers, what it charges and what makes it different, in one place, without them having to reconstruct it from a marketing site. It is not a magic input and it does not override anything else, but it removes ambiguity cheaply.
More important is not blocking the crawlers. We regularly find local business sites that inadvertently block AI user agents in robots.txt, usually copied from a template, and are then invisible to the exact systems they want to appear in.
What we do not claim
This field attracts more confident nonsense than any other part of search work, so here is what we will not tell you.
We cannot guarantee that ChatGPT or any other system will name your business. Nobody controls those outputs, they change without notice, and two people asking the same question can get different answers.
We cannot give you a ranking position in AI search, because there is no ranking. There is a shortlist that is regenerated each time.
We cannot make this instant. These systems re-index on their own schedule, measured in weeks rather than days.
What we can do is make the business the most machine legible, most citable, most consistently described option in its category, and then measure whether the answers change.
How we measure it
Measurement is what separates this from theatre. We run a fixed set of buyer stage queries against multiple systems on a schedule and record what happened.
Which businesses were named. Whether yours appeared, and in what position within the answer. What sources the answer drew from. How the system described your business, which is frequently more revealing than whether it named you, because a wrong description points at exactly which signal is unclear.
We run the same queries before starting and repeat them on a cycle, so the comparison is against a recorded baseline rather than a memory. When the answer does not change, we say so.
Why we know this method works
We did not learn this from a webinar. Our parent company Rendimension built and measured a large scale test of exactly this mechanism.
The starting observation was that the best of and top ten articles AI systems cite for a category are frequently published by the competitors themselves. We verified that by reading the cited sources rather than assuming. We then replicated the format across hundreds of pages, with a strict rule set: never place ourselves first, always name real verified competitors, always disclose that we published the list, and never claim superiority we could not support.
When we measured citation across a representative sample of queries, seven of eight returned a hit, meaning the AI answer named us directly. The one miss was in a category saturated by large established platforms, which is itself useful information: the method works where the category is not already owned, and it does not manufacture a win where it is.
That is the same method we apply here, adapted to the local services market. Our own San Antonio comparison article is published under that rule set, including the disclosure that we publish it and appear in it.
Where AI visibility sits relative to everything else
It is not a replacement for local search work and anyone selling it that way is overstating it. The map pack still captures roughly 44% of clicks on a local results page, and near me searches still convert within a day for about 76% of the people who run them. Traditional local search is where most of the volume is today.
The relationship is closer to overlap than competition. Consistent business information, structured data, real reviews and pages built around specific questions improve both at once. The work that gets a business into an AI answer is mostly the same work that gets it into the map pack, done with extra attention to how the page reads to a machine.
That is why this runs alongside local SEO rather than instead of it, on top of a site built by website design and kept current by website management. If your current site is the constraint, website redesign comes first, because no amount of entity work rescues a site a machine cannot read.
Getting started
The honest first step is a baseline. We run the buyer queries for your category in San Antonio, record who is being named today and which sources those answers come from, then tell you whether the category is winnable and what it would take.
Sometimes the answer is that a national platform owns the category and the effort is better spent elsewhere. We would rather say that at the start than bill for a year of work that was never going to land. Our guides on how AI Overviews cite local businesses and AI search visibility for San Antonio businesses cover what you can check yourself first.
The questions your customers are actually asking AI tools
Understanding the shape of these questions changes what you publish. They are longer, more conversational and more loaded with constraints than typed searches, and they almost always contain a decision the person is trying to make.
A typed search is two or three words. The same person asking an AI tool writes a sentence: who is reliable for emergency AC repair in San Antonio on a weekend, roughly what should it cost, and how do I know I am not being oversold. That one question contains three separate requirements, and the businesses named in the answer are the ones whose published material addresses all three.
This is why thin service pages perform badly here regardless of how well they are designed. A page that says the business does AC repair in San Antonio answers one third of that question. A page that also states realistic price ranges, explains what an honest diagnosis includes and describes weekend availability answers all of it, and becomes the source the answer is built from.
The practical instruction is to write for the question, not the keyword. The keyword is ac repair san antonio. The question is everything above.
What we fix first, in order
One: crawler access. We check robots.txt for blocked AI user agents, confirm the site returns real content without requiring JavaScript execution to read the important text, and verify pages are actually indexed. This is the cheapest possible failure to fix and one of the most common.
Two: entity consistency. Business name, phone, service area, categories and service list, made identical across the site, the structured data, the Google Business Profile and every directory we can find. Contradiction is the enemy here, and old listings from previous vendors are usually the source of it.
Three: the answer layer on every important page. A direct answer in the opening lines, descriptive headings matching real questions, and a genuine FAQ block with self contained answers. Applied to service pages first, because those are the pages with commercial intent.
Four: specific facts replacing vague claims. Ranges, timelines, requirements, processes. Every one traceable. This is the slowest step and the one that produces the most durable result.
Five: the off-site sources. Accurate, complete listings on the directory platforms the answers are actually built from, plus the comparison content published under a disclosed, honest rule set.
Six: measurement. The query set run on a schedule against a recorded baseline, reported whether or not it improved.
Common mistakes we see
Marking up FAQs that do not appear on the page. Structured data is supposed to describe what is visibly true. Invisible FAQ markup is a policy violation with real penalty risk, and it is surprisingly common on sites built by vendors chasing rich results.
Publishing a listicle where the business ranks itself first. This reads as promotional to both humans and systems, and it undermines the credibility that makes the format work at all. Placing yourself where you honestly compete, and saying plainly that you published the list, is what makes it usable as a source.
Adding an llms.txt file and treating the job as finished. The file removes ambiguity. It does not create authority, generate reviews or write the pages that get cited.
Chasing every new AI platform separately. The underlying requirements are largely the same across systems: be readable, be consistent, be specific, be present in the sources they trust. Platform specific tactics change monthly. The foundation does not.
Assuming visibility in one system means visibility in all of them. They weight sources differently. In our own measurement, one system leaned heavily on reputation and published lists while another responded much faster to the live pages on a site. That difference is why we test more than one.
What this is worth in San Antonio, in numbers
We measured local demand with Google Keyword Planner rather than estimating it, because the value of being named depends on how many people are asking.
In the San Antonio media market, web design san antonio and its close variants register roughly 1,300 searches per month with top of page bids reaching $34.65. Seo company san antonio and its variants add 320 at up to $62.24. Marketing agency san antonio adds 480. Across all geo modified variants we counted 98 keywords with at least 10 monthly searches locally.
Those click prices matter for a reason that has nothing to do with advertising. They tell you what a single decision in this category is worth to the businesses competing for it. When an AI answer names three companies and the person calls one of them, that is a decision nobody paid $62 to influence. The economics of being named are better than the economics of being clicked, and the competition for it is currently much thinner.
The same pattern holds in every trade we have measured. Hvac seo reaches $125 top of page bids nationally. Law firm website design reaches $113.70. Plumbing marketing agency reaches $60. High click prices are a reliable signal that the buyer is valuable, and every one of those categories has the same AI answer surface that almost nobody is deliberately competing for yet.
