How to Get Your Business Recommended by ChatGPT (and Other AI)

First, how the AI actually decides

When someone asks ChatGPT, Claude, Gemini, or Perplexity "who is a good roofer near me" or "best injury lawyer in Buffalo," the assistant does one of two things, and the difference matters more than anything else in this guide.

With web search on (Perplexity always, and increasingly the default in ChatGPT and Google's AI answers), the assistant runs an ordinary web search for roughly the customer's question, reads the top-ranking pages, and composes an answer from what it just read, usually with citations. If your business appears in those pages, you are in the room when names get picked. If not, you do not exist for that answer.

Without search, the assistant answers from its training data: a snapshot of the web collected months before the model shipped. This layer is brand gravity. It knows businesses that were mentioned consistently across many sources for years, and it is essentially deaf to anything you did last quarter.

The playbook, in the order that pays

1. Be readable: the entry ticket

Before an engine can recommend you it has to crawl, read, and identify you. That means: AI crawlers not blocked in robots.txt (many sites block GPTBot and ClaudeBot through boilerplate without knowing), structured data that declares who you are as data rather than prose (LocalBusiness or your industry's schema type, carrying name, address, phone), a phone number and address in real text rather than images, a page title that carries your name and city, and enough substantive text to quote. None of this gets you named by itself. All of it is required before anything else works.

2. Publish the page that answers the question

Search-grounded engines read whatever ranks for the customer's question, and in most local markets we have measured, the most-read source is not a directory: it is a business's own page with a title that matches the question verbatim. A page literally titled and structured around "best [your service] in [your city]," answering plainly with real specifics, gets retrieved when customers ask exactly that, and being retrieved is being in the room.

3. Get present where the engines already look

The rest of what engines read is remarkably consistent: review platforms (Yelp, BBB, industry review sites), directories for your vertical (for lawyers: Justia, Avvo, Super Lawyers, Expertise.com, FindLaw, Martindale), and an authority layer of rankings and press (Best Lawyers style lists, Forbes roundups, Wikipedia where warranted). Every one of those is a place your business can be present through ordinary, legitimate means: claimed and completed profiles, accumulated reviews, earned mentions. When the engine's other reads corroborate your own site, naming you becomes safe; when they contradict or omit you, it is not.

4. Reviews: count, recency, and responses

Review platforms appear in the engines' reading list on nearly every commercial query. What the engines can weigh is what is visible: how many reviews, how recent, and whether the business responds. A steady flow of genuine reviews beats a burst of old ones, and review velocity is one of the few signals that feeds both mechanisms at once: retrieval today, and the corroborated reputation the next generation of models trains on.

5. Measure monthly, because the answers move

Studies of AI answers find 40 to 60 percent of cited sources change month to month. That volatility is bad news for anyone who wants to win once and coast, and good news for anyone starting behind: the answer box is re-decided constantly. Ask the engines your customers' questions on a schedule, record who gets named and which sources were read, and treat the citation list as your to-do list. That is not a metaphor: the pages an engine cites when it fails to name you are, precisely, the places you need to be.

The two clocks, honestly stated

The fast clock (weeks to a few months): everything above moves the search-grounded layer as fast as your pages can rank and your profiles can fill. This is where effort shows up first and where progress is measurable almost immediately.

The slow clock (12 to 24 months): training memory only learns you when your presence is consistent across many sources before the next crawl and training cycle. There is no shortcut. The consolation is that the fast-clock work, sustained, is exactly what the slow clock eventually bakes in. The businesses that dominate model memory today simply started accumulating corroborated presence years before anyone was measuring it.

What does not work

• Keyword stuffing and invisible text. Engines read pages the way people do, and the synthesis step actively filters spam patterns.
• Guarantees. Answers are probabilistic and citation sets churn. Anyone guaranteeing "#1 in ChatGPT" is guaranteeing something they cannot control.
• One-time fixes. The technical layer is a fix; visibility is a program. Both are real work, and they are different work.
• Waiting for certainty. The training-memory clock means that starting a year from now costs a model generation, not a year.

Where to start today

Check where you stand. Ask ChatGPT, in a private window, exactly what your customer would ask, and see who gets named. Then check the technical half in about a minute: our free scan runs live checks against your actual site, whether AI crawlers can get in, whether your identity is machine-readable, whether there is anything worth citing, and tells you exactly what to fix. Then read why businesses end up invisible to AI for the five failure patterns we see most.