7 minute read
How AI assistants decide which businesses to recommend
AI assistants name a few businesses instead of ranking ten links. What shapes that short list, and what a multi-location business can control.
Ask an AI assistant for "a good estate planning attorney in Boulder" and you won't get a page of links. You'll get three or four names, a sentence about each, and maybe a few citations. If you're not one of those names, the conversation is over before a buyer ever sees your website.
That short list is now a real channel. Google reported AI Overviews reaching more than 2 billion monthly users in 2025, and by mid-2026 said its AI Mode had passed 1 billion monthly users. OpenAI has described ChatGPT at roughly 900 million weekly users. BrightLocal's 2026 consumer survey found the share of people using generative AI for local business recommendations jumped from 6% to 45% in a year.
Two sources of knowledge
Assistants answer from two places. The first is what the model learned in training: a broad, somewhat dated picture of which businesses exist and what people say about them. The second is live retrieval: when the assistant searches the web in the moment and reads a handful of pages before answering.
For anything local or recent, retrieval does most of the work. That's good news, because retrieval reads the same public sources you can influence today.
What the assistant reads
- Maps and business profiles. Name, category, hours, address and attributes from Google Business Profile and the major map providers.
- Directories and data aggregators. Industry directories and the data networks that feed them. Consistent facts across them build confidence; contradictions erode it.
- Reviews. Ratings, volume, recency and what reviewers actually say. "Pre-planning", "bilingual staff" or "same-day appointments" in reviews become reasons an assistant gives for naming you.
- Your own site. Clear service and location pages, plus structured data (schema) that states facts in a machine-readable way.
- Third-party pages that rank. "Best of" lists, local news, association pages and comparison articles. Assistants often cite these directly.
Why multi-location businesses get left out
The most common failure we see isn't bad content. It's inconsistent facts. One location moved and three directories still show the old address. An acquired practice kept its old name on Yelp. Holiday hours were updated on Google but nowhere else. To a retrieval system, those look like different businesses or unreliable data, and the safest move for the assistant is to recommend someone else.
What you can control
- Make the facts agree everywhere. One master record per location, pushed to every listing and kept in sync.
- Mark up your site. Organization, LocalBusiness and Service schema on the right pages, matching your listings exactly.
- Answer real questions on your pages. Prices or price ranges, service areas, what's included, who it's for. Specific answers get quoted; vague ones don't.
- Earn and answer reviews. Ask at the right moment, respond to every one, and let customers describe what you're good at in their own words.
- Get mentioned where assistants look. Local press, associations and credible comparison pages in your category.
How to measure it
Pick 25 to 50 questions your buyers actually ask. Ask each assistant the same way every month. Log who gets named, in what order, and which sources get cited. That log tells you where you stand, who you're losing to, and which sources to fix next. It's the core of how we track AI visibility for clients.