A growing share of “who should I call” questions never touch a search results page anymore. People ask an assistant. “Find me a good plumber near me.” “Which coffee shop nearby is open right now.” “Who does emergency HVAC in this area.” ChatGPT, Google’s AI Overviews, Gemini, and Perplexity answer in a sentence or two, naming a handful of businesses.
If your business is not in that handful, you may as well not exist for that customer. So the obvious question: how do these assistants decide who to name?
AI assistants do not “know” your business. They retrieve it.
The first thing to understand is that a language model does not carry a live directory of local businesses in its head. When you ask for a recommendation, the assistant gathers current information from the web and from structured data sources, then writes an answer grounded in what it found. This is retrieval, and then generation.
That means the assistant is only ever as right about your business as the information it can retrieve. If the web says you close at 5 when you close at 7, the assistant will confidently tell a customer you are closed. It is not lying. It is repeating the best information available to it.
Four things that decide whether you make the answer
Across the major assistants, the same signals keep mattering:
1. Consistency of your core facts. Your name, address, phone, hours, and category need to agree across the places an assistant looks: Google, Apple Maps, Bing, and the directories that feed them. When the sources agree, the assistant treats the information as reliable and uses it. When they conflict, the assistant hedges, picks one at random, or skips you for a business it is more sure about. Contradictory data is the quiet killer here.
2. Structured data it can read without guessing. Assistants strongly prefer information that is labeled, not just visible. A clearly marked business name, address, hours, and category in structured data (schema) removes ambiguity. A page that only shows those details as styled text forces the model to infer them, and inference is where mistakes happen.
3. Corroboration from sources it trusts. An assistant is more confident naming a business that shows up consistently across several independent, credible sources than one that appears in a single place. Being present and accurate in more of the places that matter raises the odds you are the safe answer.
4. Reviews and reputation signals. Ratings and review sentiment feed into which businesses get called “good” or “top rated.” This is the part you influence through service, not settings, but it only counts if the profile carrying those reviews is correctly tied to your business in the first place.
The uncomfortable pattern
Notice what three of those four signals have in common. Consistency, structured data, and corroboration are all downstream of one thing: is your business information accurate and the same everywhere it appears?
You can have great reviews and still lose the recommendation, because the assistant found two different phone numbers for you and could not tell which business was which. You can be the best plumber in town and get skipped, because your hours on one directory say you closed permanently three years ago.
Being the recommended answer is not a trick you play on the model. It is the byproduct of the model being able to find one clear, consistent, correct version of your business.
What this means for a small business
You do not need to reverse-engineer each assistant. Their incentives all point the same way: they want to give the person a correct, confident answer, and they reward businesses that make that easy. The practical work is unglamorous:
- Get your core facts right, and identical, on Google, Apple Maps, and Bing.
- Make sure structured data on your own site states those same facts plainly.
- Fix the stale copies and duplicates sitting in directories you forgot about.
- Keep it that way, because a single edit or a moved location can put you right back into “the sources disagree” territory.
That last point is the hard one. Doing this once is a weekend. Keeping it true forever, across every platform, as details drift, is the actual job.
This is what LocalBasics is for. You set your name, address, phone, hours, and categories once in a single protected record, and that record stays accurate and consistent everywhere an assistant looks, monitored for life, on a one-time payment. It does not promise the assistant will pick you. It makes sure that when the assistant looks, it finds one clear, correct version of your business instead of three conflicting ones. That is the part you control, and it is the part that decides whether you are even in the running.
The bottom line
AI assistants recommend the businesses they can understand with confidence. Confidence comes from consistency. If your information is accurate and the same everywhere, you are eligible to be the answer. If it is scattered and contradictory, the assistant will quietly recommend someone else, and you will never see the customer you lost.