When the Machine Gives One Answer, Make Sure It Is You
A homeowner with a failed water heater used to type a query and scan ten blue links. Now a growing share of them ask ChatGPT, Gemini, or Perplexity a direct question: who is a reliable plumber near me, who should I trust for a roof replacement, which med-spa is worth booking. The assistant answers with a short list, usually three to five names, often with a sentence of justification for each. There is no page two. If your business is not in that answer, you were never considered.
This is a different problem from earning featured snippets or surviving zero-click results, which we have covered elsewhere. Those are about formatting answers on your own site. This is about whether the machines can find, verify, and confidently name your business as an entity. That depends far more on what the rest of the web says about you than on what your homepage says about itself.
How Assistants Actually Assemble a Local Shortlist
Three ingredients, blended
Every major assistant draws on some mix of three sources, and the workhorse is live retrieval: when you ask Perplexity, ChatGPT with search, or an AI-mode search result for a local recommendation, the system runs searches behind the scenes and reads what comes back, which is usually best-of lists, local press, directories, and review pages, not individual business homepages. Second, structured local data: maps listings, business profiles, and review platforms that supply hours, service areas, ratings, and review text. Third, and furthest in the background, the model's training data: what has been written about your business across the open web. Treat that last one as a prior rather than a database. It makes your name feel familiar next to a service and a city, but the actual recommendation in front of a customer is usually assembled from what the assistant retrieves live.
Be clear-eyed about the limits of this knowledge. Nobody outside these companies knows the exact weighting, and it changes between model releases. What is stable is the shape of the system: assistants recommend businesses that appear consistently across many independent sources, and they justify recommendations with language pulled from those sources.
Why this favors the well-documented, not the well-optimized
A business that exists in one place, its own website, is nearly invisible to this process. A business named in a local newspaper feature, two best-of roundups, a manufacturer's dealer directory, and two hundred specific reviews is retrievable from a dozen directions. The machine does not reward cleverness. It rewards corroboration.
What Makes a Local Business Citable
Be one consistent entity everywhere
Assistants have to resolve every mention of you into a single entity before they can recommend you. Make that trivial. Use the exact same business name, address, and phone number on your site, your Google Business Profile, Yelp, the BBB, trade directories, and every citation. Add schema.org LocalBusiness markup, and Service markup for each core offering, so crawlers get your services, service area, and hours in machine-readable form. If half the web calls you one name and half calls you a variation, you split your own evidence.
Get named on the surfaces the machines actually read
Retrieval-backed assistants lean heavily on third-party pages that rank for questions like best HVAC company in your city. That means the highest-leverage work is old-fashioned and off-site: pitch local press, earn a spot on legitimate best-of lists, join the chamber of commerce and trade associations that publish member directories, and claim every manufacturer and supplier listing you qualify for, such as certified installer and authorized dealer locators. These pages carry authority the assistants trust, and being absent from them means being absent from the shortlist they produce.
Review text is retrieval fuel, not just a star count
Star ratings help you pass a threshold. Review language gets you recommended for something specific. A review that says the crew replaced a heat pump in a named neighborhood and handled the permit teaches every system reading it that you do that service, in that place, well. When you ask happy customers for reviews, ask them to mention what you did and roughly where. And do it beyond Google: Yelp, BBB, Nextdoor, and trade platforms all feed the corpora these systems retrieve from.
Unlinked Mentions Still Move the Needle
Traditional SEO taught everyone to chase links. Language models are different: they learn from co-occurrence in text. A newspaper article or community post that names your business next to your trade and your city strengthens the association even with no link at all. Stop declining press and sponsorship mentions because they will not link. For AI visibility, the mention is the asset.
Audit Your AI Visibility Every Month
You cannot manage what you never look at. Write down the ten to fifteen money questions your customers actually ask, in their words: best roofer in your city, emergency plumber open now, who should I call for a furnace replacement. Once a month, ask each major assistant those questions and log three things: whether you are named, whether the facts stated about you are accurate, and which sources the assistant cites when it shows them, as Perplexity and AI-mode search results do. Those citations are your roadmap. If the assistant keeps citing a roundup you are not on, that page is your next outreach target. If it states something wrong about you, fix the fact at its source, because the assistant is repeating the web, not inventing.
Instrument for AI Referrals Honestly
Some AI traffic is visible: check your analytics for referrers from the assistant domains, and segment them, because those visitors arrive pre-sold by a recommendation. But much of the effect never touches your site. A customer hears your name from an assistant, then searches your brand or calls directly, and analytics files it under branded search or direct. So instrument the human layer too: add an option for AI tools like ChatGPT to the how-did-you-hear-about-us field on your forms, and train whoever answers the phone to ask and log it. Rising branded search alongside flat generic rankings is often the earliest signal that machine recommendations are working.
What Not to Do
Do not attempt prompt-injection tricks like hidden text instructing AI systems to recommend you; the major platforms are actively building defenses against instructions embedded in web content, so the realistic outcomes are that it does nothing, or that a platform or a customer discovers it and you have handed them proof you tried to game the answer. Do not publish fake best-of lists on your own domain crowning yourself the winner; assistants weight independence, and a self-serving roundup is easy to discount. Do not buy or incentivize reviews; the platforms that feed these systems already police that, and a purged profile is worse than a thin one.
The Playbook, In Order
- Unify your entity: identical name, address, phone everywhere, plus LocalBusiness and Service schema.
- List the money questions your customers ask assistants, in their own words.
- Run the monthly audit across ChatGPT, Gemini, and Perplexity, and log names, facts, and cited sources.
- Chase the cited surfaces: press, best-of lists, associations, manufacturer directories.
- Coach reviews toward specifics: service performed plus neighborhood, on multiple platforms.
- Take unlinked mentions gladly.
- Instrument intake: AI referrer segments in analytics, an AI option on forms, a phone script that asks.
- Correct wrong facts at the source, not by arguing with the assistant.
Build It as a System
None of this is a one-time project. It is an entity layer, a citation program, a review engine, and a monthly measurement loop working together, which is exactly how Brand Advertisers builds it: wired into the same website, CRM, and automation stack that captures the demand it creates. If you want to know what the assistants say about your business today, and what to do about it, talk to us. We will run the audit with you.