← All insights
·11 min read·By Christopher Sanovei

Showing Up in AI Answers: What Changed and What To Do

A growing share of customers ask an AI assistant before they see a list of links. Here is what influences whether you get named, without the hype.

A real share of your customers now open ChatGPT, Gemini, Perplexity or Google's AI overview and ask a question in a sentence, instead of typing two words and scanning blue links. If the answer they get does not name your business, the search ended without you and you will never see it in your analytics.

There is a great deal of noise about this. Here is what appears to hold up and what does not.

What is genuinely different

Classic search gives a list and the user chooses. An AI assistant gives an answer and the user usually accepts it. That compresses ten results into roughly three named options, and being the fourth best answer is now much closer to being invisible than it used to be.

For local service questions, these systems lean heavily on sources they can verify: your Google Business Profile, your reviews, established directories, local news and press, and pages with clear structured data. They are noticeably more conservative about naming a business than they are about explaining a concept, which makes sense given the liability.

What seems to influence being named

  • Consistent, verifiable facts everywhere. The same name, address, phone, hours and service area across your site, your profile and the directories. Contradictions make a model less willing to assert anything about you.
  • Structured data. LocalBusiness, Service, FAQ and review markup. Machine readable claims about who you are, what you do and where.
  • Content that answers the actual question. These systems quote passages that directly answer a question. A clear question as a heading with a direct two sentence answer underneath gets pulled far more than the same information buried in a paragraph.
  • Third party corroboration. Reviews, local press, genuine directory listings, trade associations. Independent confirmation that you exist and do this work.
  • Reviews, again. Both as a quality signal and as text that gets read and summarised.

What is being oversold

You will be offered an llms.txt file as the answer to all of this. It is a proposed convention for telling language models about your site. Adding one is harmless and takes ten minutes, so we add it. But there is no good evidence it improves whether you get named, and anyone selling it as the centrepiece of an AI strategy is selling you a text file.

Be equally sceptical of guaranteed AI visibility. Nobody controls what these systems say, the systems change monthly, and the people promising it are guessing with your money. The honest version is: here is what we can measure, here is what we are doing, here is whether it moved.

How to measure it at all

This is the part that gets skipped. You cannot improve what you are not checking.

Build the list of questions a customer would actually ask: who is the best plumber in your city, who does emergency repair near a given neighbourhood, who should I call for a specific problem. Run them against the major assistants on a schedule, and record whether you were named, what was said, and who was named instead.

That gives you a baseline and a trend, which is all anyone can honestly offer right now. We track it for every client alongside ranking and map position, because it is becoming a real share of how people find a service business and pretending otherwise helps nobody.

The short version

Most of what makes you visible to AI assistants is what makes you visible in local search generally: a complete and accurate profile, real content that answers real questions, consistent facts, independent corroboration, and reviews. The foundation is the same. What changed is the cost of being fourth.

Where the assistants are getting their answers

It is worth being concrete about this, because the mental model of "the AI knows things" leads people to the wrong work. For a local service question, the assistant is usually not recalling your business from training. It is running a search, reading the results, and summarising them. What it finds is what the open web says about you right now: your profile, your reviews, the directories, any local press, and your own pages.

That has a practical consequence. The question is not "how do I get into the model" but "what does the web say when something looks me up". If the answer is a complete profile, forty recent reviews mentioning specific jobs, a site that states plainly what you do and where, and three local organisations that list you, the assistant has what it needs to name you with confidence. If the answer is a profile with no services, a phone number that differs between two directories, and a site whose home page is a slogan, it will name someone else.

Writing pages an assistant will quote

Assistants lift passages that answer a question directly. A heading phrased as the question, followed by a two or three sentence answer that stands on its own, is the format most likely to be pulled. Long paragraphs that build to the answer are less useful to them, however well written.

This happens to be good writing for people as well. A homeowner at 9pm with a leaking water heater wants the answer at the top, not after three paragraphs of context. Structure your service pages and your FAQ so each question has its answer immediately beneath it, and mark the FAQ up with structured data so the question and answer pairs are machine readable. The same page serves both readers.

Consistency is the quiet part

A model deciding whether to assert that you exist and do this work is looking for agreement between sources. Your name spelled the same way everywhere. One phone number. One address. Hours that match. Service areas that match. A single contradiction between your profile and a directory listing is enough to lower its confidence, and lowered confidence means naming a competitor with cleaner facts.

This is the citation work that was always part of local SEO, now with a second reason to do it. It is dull, it is finite, and it is one of the few things on this list with no downside.

How we measure it for clients

Every two weeks we ask the major assistants the questions a customer in each client's market would ask, and record whether the business was named, what was said about it, and who was named instead. The first run is a baseline, usually zero for a business that has not done the work, and the trend from there is the only honest measure anyone can offer today. It sits on the dashboard next to rankings and the map grid, because it is becoming part of how people find a service business and pretending otherwise helps nobody.

Christopher Sanovei

Founder, Orvelo Solutions

Chris runs Orvelo Solutions from Orlando, Florida, working with home and local service businesses on local search, Google Business Profiles and the arithmetic of what a customer costs.

Want to know where you actually rank?

We will run your business across your real service area and show you where you sit against your competitors. Free, and no pitch.

Book a free visibility audit