Be one of the names an assistant actually gives out.
Ask ChatGPT or Perplexity for a good plumber in your town and you do not get ten links. You get two or three names. Those names come from what the wider web can tell a model about you, not from your website on its own, which is why this work happens mostly off your site.
Written for local business owners who want the phone to ring. Agencies adding AI visibility to what they sell, we do the same work white-label under your brand: see the partner model.
A model names the businesses it can check.
An assistant is not reading your website the way a customer does. It has read an enormous amount of the web, and when someone asks who to call, it leans on what several independent sources appear to agree on. Your own site is a claim. A chamber roster, a trade directory, a supplier locator, a news story and a few hundred reviews that all describe the same company, at the same address, doing the same work, is a claim other people back up. Models name businesses they can back up.
Say you are a roofer in a mid-sized city. Your website says you are the best in town and you have been at it for years. Nothing else on the open web confirms it or contradicts it. The model has a claim and no way to verify it, so it reaches for a business it can, and that business gets the call.
Same roofer. Now the same facts, in the same words, sit on your profile, two trade directories, the chamber roster, a manufacturer's certified installer list and a storm-season article in the local paper. The claim is checkable, the model is confident, and you are a safe name to give a stranger.
What the models are actually reading.
Five things decide whether you come up. Four of them are work we can do. The fifth is the result of the other four, and it is the one nobody can promise.
- 01
AI citations
The sources an assistant shows underneath its answer. Perplexity lists them as a matter of course and ChatGPT links them when it has searched the web. Being one of those links for the questions in your category is the most direct version of this whole job, and it goes to pages that answer a question cleanly on a site the model has reason to trust.
In practiceYour service pages, and the third-party pages about you that already get pointed at.
- 02
LLM visibility
Whether you come up at all, across the spread of questions a customer would really type. Not one query. Thirty or forty, phrased several different ways, asked more than once, because these systems do not return the same answer every time and a single lucky result proves nothing.
In practiceA fixed prompt list, run every month, with the answers saved word for word.
- 03
Brand mentions
Being written about on pages you do not own. Directories, association rosters, supplier locators, chamber pages, event programs, local news, forum threads and roundups. A mention does not need to be a link to count. It needs to exist somewhere the model has read, and it needs to say the same thing your other mentions say.
In practiceThe slow half of this work, and the half that moves it.
- 04
Entity signals
Whether the model understands you to be one business rather than three similar ones. Same legal name, same address, same service list, same categories, same hours, in the same wording everywhere it looks. Contradictions are not simply ignored. They lower confidence, and a model that is not confident names somebody else.
In practiceOne written description, used everywhere without improvising.
- 05
AI recommendations
The end state: being one of the two or three names that come back. It is built out of the four above, it moves slowly, and it is never fully inside anyone's control. We treat it as the outcome we are working toward and the first four as the work we can actually put hours against.
In practiceReported as a share of answers over time, never as a yes or no.
Mostly fact-checking, done thoroughly, month after month.
There is no clever trick here and we are not going to invent one. Most of a month on this service goes on correcting information that is wrong in public and earning mentions from organizations that take a while to answer their email. That is the honest shape of it, and it is the order the work has to happen in.
Write the facts down once
One description, one business name, one address format, one service list, one set of hours. We agree it with you, keep it in a single document and stop improvising. It sounds too simple to matter. It is the reason everything after it holds together.
Find out what the web already says
We collect what is out there now: listings, profiles, old addresses, a previous owner's name, a number that no longer rings, two entries for one business, the wrong category. Then we ask the assistants directly and record the answers before we change anything, so there is a real starting point.
Correct the record
The unglamorous part. Claiming, editing and merging listings one form and one support ticket at a time. Some go through in a minute. Some take six weeks and a follow-up to a company that would rather not hear from you. This is the step most programs quietly skip, and skipping it caps everything else.
Earn mentions worth having
Association membership, manufacturer and supplier locators, chamber rosters, local sponsorships and press that has an actual reason to write about you. We go after places that are real, because the model's trust in the source is the entire point. Bought mentions on sites nobody reads do nothing here.
Make your own pages quotable
Clear service and area pages, plain answers to the questions you get asked on the phone, and an about page that states the facts instead of selling. This overlaps withanswer engine optimization, so we do it once and both surfaces benefit.
Ask again, every month
The same prompt list, the same assistants, the same way of recording it. Answers move on their own, and the only way to know which direction yours are moving is to keep asking in a way that is comparable month to month.
If you are the agency, not the business
Partners take this as a white-label workstream. We run the prompt list, do the listing corrections and the mention outreach, and hand back a report in your template with your name on it. You keep the client relationship and the margin. Nineteen agencies work with us this way, alongside the direct clients we have served since 2015, more than 800 businesses between them.See the partner model.
What we will not promise.
Nobody outside OpenAI, Google, Anthropic and Perplexity decides what their models say. The answers change between versions, they change between one week and the next, and two people asking the same question in the same city can get different names. Any agency that guarantees you a spot in the answer is selling control it does not have.
What we can honestly offer: we make your business easier to find, easier to describe consistently and easier to verify against sources that are not yours, and we measure whether the answers move. When they do not move, you will hear that from us rather than reading a chart that avoids the subject.
Most of this never shows up in Google Analytics.
If you judge this work by the referral numbers in your analytics, you will conclude it did nothing. That is a measurement problem rather than a result, and it is worth understanding before you start paying for it.
A fixed list of the questions your customers would really ask, run every month across ChatGPT, Perplexity, Claude and the Gemini app, with the answers saved as written. The share of those answers that name you is the headline number we report.
Which pages get shown as sources when your category comes up, including pages about you that you do not own. This is the part you can act on fastest, because a missing citation usually points at a specific page that needs writing or a specific site that has you wrong.
A running list of every place on the web that describes your business, what each one says and whether it is correct. It is dull in month one and genuinely useful by month four, when you can see it grow and see the contradictions disappear.
Some assistant traffic does arrive with a referrer you can segment, chatgpt.com and perplexity.ai among them, and it is worth splitting out. Plenty arrives with no referrer at all, because the person heard your name, closed the assistant and typed you into the address bar. Watch direct and branded search next to the referral numbers, and ask every new caller where they heard your name. That answer is worth more than the dashboard.
One of five places the answer can happen.
We split this work by where the customer's question gets answered, not by what the tactics are called. This page is the one furthest from Google: the assistants people open instead of a search engine.SEO is the foundation all five stand on, the technical, on-page and authority work that makes your pages findable in the first place.Answer engine optimization is Google quoting one of your pages directly on the results screen.AI search visibility is Google writing its own summary instead, in AI Overviews and AI Mode. Andsearch experience optimization is everything after the click, whether the page someone lands on turns into a call.
They share inputs. The consistency and mention work described on this page also helps the Google surfaces, which is why we rarely sell it on its own and never sell it as a replacement for the basics.
What owners ask us about this.
What is generative engine optimization, in plain English?
Assistants like ChatGPT, Perplexity, Claude and the Gemini app answer a question by naming two or three businesses instead of listing a page of links. Generative engine optimization is the work that makes your business one of those names. Most of it happens away from your website, on the listings, profiles, rosters and articles a model can find and cross-check.
Can you guarantee ChatGPT will recommend my business?
Nobody can, least of all the people selling the guarantee. The companies running these models do not publish how the names get picked, the answers drift week to week, and two people asking the same thing in the same city can get different businesses back. What is within reach is making you easy to find, easy to describe and easy to check, then measuring whether the answers move. Treat a promised placement as a reason to walk, or at the very least to ask for it in the contract with a refund attached.
How is this different from the AI work on your other page?
It is a different place, not a different idea. AI search visibility covers Google's own generative answers, AI Overviews and AI Mode, which appear on the search results page and lean on Google's index. This page covers the assistants outside Google: ChatGPT, Perplexity, Claude and the Gemini app. They read different sources and are measured differently, so we run them as two jobs that share one foundation.
Do I need a special file or page so the AI can read my site?
No. There is a steady supply of products promising a magic file that makes models prefer you. Google has said publicly that things like llms.txt are not required, and the assistants outside Google have not asked for one either. What actually moves this is dull: facts that match everywhere, real mentions on sites you do not own, and pages that answer the question.
How long before anything changes?
Corrections land fastest, because you are fixing something already wrong in public rather than building something new, and those can show inside a few weeks. Mentions and citations move slower, usually a quarter or two before the answers shift, and longer in a category full of established names. We would rather read what the web currently says about you and hand you a range than name a date on a first call.
We are an agency. Can you run this under our brand?
Yes. Nineteen agencies use us for fulfillment today and this goes out the same way: your brand on the report, your client relationship, our team running the prompt list, the listing corrections and the mention work. The partner model has the detail.
See what the assistants say about you today.
Tell us your business name, your city and what you do. We will ask ChatGPT, Perplexity, Claude and the Gemini app the questions your customers would ask, and send you what came back, including the answers you will not enjoy reading.
Form not loading? Email hello@beyaoshy.com and we will reply within one business day.